Agricultural Institute of Slovenia, Hacquetova 17, 1000 Ljubljana, Slovenia.
J Dairy Res. 2010 May;77(2):137-43. doi: 10.1017/S0022029909990586. Epub 2009 Dec 24.
Three models for the estimation of milk, fat and protein daily yield (DY) based on a.m. (AM) or p.m. (PM) milkings were compared. A total of 518 766 test-day records from 5078 dairy cattle farms obtained between March 2004 and April 2008 were analysed. The DY model was a linear model with DY as a dependent variable. In the PYR model and the DYR model, partial yield ratios (AM:DY and PM:DY) and daily yield ratios (DY:AM and DY:PM), respectively, were used as a dependent variable in the first step. In the second step, DY was estimated as a partial yield divided (PYR model) or multiplied (DYR model) by the estimated yield ratio from the first step. Models included the effect of partial yield (only in the DY model), milking interval, stage (month) of lactation and parity. Analysis of variance indicated that partial yield was the most important source of variation for the DY model whereas milking interval had the biggest effect in the PYR model and the DYR model. Differences in accuracy (correlation between the true and the estimated DY) between the models were negligible. On the other hand, models differed in the amount of bias (average error). The DYR model on average overestimated DY by 0.13 kg, 0.01 kg and 0.01 kg for milk, fat and protein, respectively. For the other two models the overall bias was almost zero. However, the DY model overestimated low and underestimated high DY owing to the well known regression property. The DYR model progressively overestimated high DY. These problems were not observed with the PYR model which seemed to be the best model. In this paper a relatively old topic was analysed and discussed from a new point of view, where the estimation of DY is based on modelling biologically more stable partial yield ratios rather than yield values from a.m. or p.m. milking.
比较了基于上午（AM）或下午（PM）挤奶的三种估计牛奶、脂肪和蛋白质日产量（DY）的模型。分析了 2004 年 3 月至 2008 年 4 月期间从 5078 个奶牛场获得的总共 518766 个测试日记录。DY 模型是一个线性模型，DY 为因变量。在 PYR 模型和 DYR 模型中，第一阶段分别使用部分产量比（AM:DY 和 PM:DY）和日产量比（DY:AM 和 DY:PM）作为因变量。在第二阶段，通过从第一阶段估计的产量比估计 DY 作为部分产量的商（PYR 模型）或积（DYR 模型）。模型包括部分产量的影响（仅在 DY 模型中）、挤奶间隔、泌乳阶段（月）和胎次。方差分析表明，部分产量是 DY 模型中最重要的变异来源，而挤奶间隔对 PYR 模型和 DYR 模型的影响最大。模型之间的准确性（真实和估计的 DY 之间的相关性）差异可以忽略不计。另一方面，模型之间的偏差（平均误差）有所不同。对于牛奶、脂肪和蛋白质，DYR 模型平均分别高估 DY 0.13 公斤、0.01 公斤和 0.01 公斤。对于其他两个模型，整体偏差几乎为零。然而，由于众所周知的回归特性，DY 模型高估了低 DY 并低估了高 DY。DYR 模型逐渐高估了高 DY。这些问题在 PYR 模型中没有观察到，PYR 模型似乎是最好的模型。在本文中，我们从一个新的角度分析和讨论了一个相对较旧的话题，即基于对生物上更稳定的部分产量比的建模来估计 DY，而不是基于 AM 或 PM 挤奶的产量值。
