Full text: Download
This study aimed to analyse the sources of variation of milking speed assessed through automatic computerised devices included in milking machines, to study the relationships between this trait and milking speed assessed through stopwatch and to develop statistical procedures useful for converting automatic device milking time into stopwatch milking time in order to obtain a fast, simple and cheap collection of milking time records for genetic evaluation purposes. A total of 571 records of stopwatch milking time (SMT), device milking time (DMT) and milk yield at milking were collected in 23 herds of the Trentino Alto Adige region in Italy equipped with two types of automatic milking devices. After log-transformation of SMT (lnSMT) and DMT (lnDMT) and a preliminary analysis of sources of variation of lnDMT, dataset was partitioned into two mutually exclusive subsets: a calibration one, used for statistical analysis, and a validation one, used as test set to validate the prediction models. This procedure was replicated 6 times in order to repeat the cross validation accordingly. Three conversion models have been compared, based on different combinations of the effects of lnDMT, milking device and herd within milking device on lnSMT. Solutions of the models have been applied for each replicate to the validation dataset for estimating lnSMT and the soundness of conversion equations have been evaluated considering the correlation between estimated and actual lnSMT and bias and precision of estimates. Milking time assessed through different procedures resulted in differences between methods for both mean and distribution, and these suggested the need of developing statistical procedures aimed to the conversion of DMT into SMT before their use in sire evaulation. The soundness of the models tended to slightly increase with the increase in the number of effects considered. The correlation between estimated and actual SMT was in the range of 0.80 to 0.86, the estimated bias was close to 0 for all models and the precision, i.e. the average standard deviation of the difference between estimated and actual SMT, in the range of 8-9% of the mean of actual SMT. In conclusion, conversion equations proposed for joining the two sources of information performed satisfactorily, giving rise to SMT accurate estimates, which were not distorted and fairly precise. The use of such equations can support the integration of automatically acquired milking time records into breeding schemes, which is advisable for increasing the number of sires progeny tested and the accuracy of breeding values estimated.