Weller J I
Institute of Animal Sciences, Agricultural Research Organization, Bet Dagan, Israel.
J Dairy Sci. 1988 Jul;71(7):1873-9. doi: 10.3168/jds.S0022-0302(88)79757-X.
Mixed model equations are constructed using the convention of regression on dummy variables that are given values of either unity (presence of the effect) or zero (absence of the effect). In the proposed method, incomplete records were included by computing regression coefficients of sire effects as the regression of the effect on the partial record on the same effect on the complete record. Partial and complete records were treated equally for other effects. Regression and the error components of variance were estimated as simple functions of the length of the partial records. The only additional computation required in sire evaluation was the differential weighting of records in the construction of the mixed model equations. This method was tested on field data and was slightly more accurate than evaluations including partial records without differential weighting and significantly more accurate than evaluations obtained with partial records deleted.
混合模型方程是按照虚拟变量回归的惯例构建的，虚拟变量的值为1（效应存在）或0（效应不存在）。在该方法中，不完整记录通过计算父系效应的回归系数纳入 ，即将部分记录上效应的回归作为完整记录上相同效应的回归。对于其他效应，部分记录和完整记录同等对待。回归和方差的误差成分作为部分记录长度的简单函数进行估计。在父系评估中唯一需要的额外计算是在构建混合模型方程时对记录进行差异加权。该方法在田间数据上进行了测试，比不进行差异加权包含部分记录的评估略准确，且比删除部分记录得到的评估显著更准确。
