RESUMO
Data for milking parameters were obtained from three different milking clusters using a modified Latin square design to take into account the real milking time regimen of the farm and the time requirements to sort cows. In this modification, cows entered the milking parlor in a random sequence rather than by group. In addition, the clusters were assigned to milking sites. Parameters were compared for 554 cows from one dairy herd that was divided into two smaller herds of 410 and 144 cows. A modified Latin square design was used to compare clusters. The results revealed that 92.1 and 95.5% of cows in Experiments 1 and 2, respectively, were treated by all three clusters; 7.9 and 4.5% of the cows, respectively, were treated by only two clusters; and no cow was treated by only one cluster. The milking parameters for both experiments were in good agreement. Results demonstrated that the modified Latin square design was equivalent to the original Latin square design under the constraint that the number of cows included in one experiment should be > 100 and that the number of milkings should not be < 200 per cluster.
