CAJ | 학술논문

论述了线性混合模型方差协方差结构与作物品种区域试验分析模型的对应关系,以我国2005-2006年东北华北玉米8组区域试验资料为例,按照线性混合模型分析原理及模型拟合信息量准则与似然比测验,对区域试验品种方差协方差的结构特性及不同方差协方差结构模型在品种效应估计与评价的差异状况进行了探讨.结果表明,在分析的所有试验中,环境间品种效应方差协方差均不符合方差分析模型假设的同质性结构,而是呈现为各种异质性结构;产量效应测验差异显著的品种对数目在方差分析模型与最佳方差协方差结构线性混合模型间的一致率平均为86%,品种产量效应排序在两种模型间也存在明显不同,品种产量效应估计的平均误差在最佳方差协方差结构线性混合模型小于在方差分析模型.
논술료선성혼합모형방차협방차결구여작물품충구역시험분석모형적대응관계,이아국2005-2006년동북화북옥미8조구역시험자료위례,안조선성혼합모형분석원리급모형의합신식량준칙여사연비측험,대구역시험품충방차협방차적결구특성급불동방차협방차결구모형재품충효응고계여평개적차이상황진행료탐토.결과표명,재분석적소유시험중,배경간품충효응방차협방차균불부합방차분석모형가설적동질성결구,이시정현위각충이질성결구;산량효응측험차이현저적품충대수목재방차분석모형여최가방차협방차결구선성혼합모형간적일치솔평균위86%,품충산량효응배서재량충모형간야존재명현불동,품충산량효응고계적평균오차재최가방차협방차결구선성혼합모형소우재방차분석모형.
The method mainly used for analyzing crop variety regional trials is based on analysis of variance (ANOVA), which requires a homogenous variance-covariance of data. Now, other models different from the ANOVA model are available. However, the problems that how the models should be assessed and that which model is more suitable for given trial data are not solved and hence restrict the applicability of the models in practices. This paper tried to solve these problems on the basis of liner mixed models. Relations between various variance-covariance structures of linear mixed model and models available for analyzing crop regional trials were discussed. Then on the basis of analyses of the corn regional trials in northeast and north China, using the information criterion and likelihood-ratio-test, the characteristics of variance-covariance structures of regional trial data and the performance difference between the ANOVA model and the linear mixed model with optimal variance-covariance structure were assessed. The results showed that the variance-covariance of variety effect over environments was not homogeneous as defined in the ANOVA model, but heterogeneous in all the considered trials. The ratio of the same variety contrast with significant difference between the ANOVA model and the optimal linear mixed model averagely reached 86%. Also, there was obvious difference in the yield ranking of varieties between the two models. The error of variety effect estimation in the optimal linear mixed model was smaller than that in the ANOVA model.