不同方差协方差结构线性混合模型在玉米品种区域试验中的适用性及模型误用对分析结果的影响

The Applicability of Different Liner Mixed Model with Variance-covariance Structures in Regional Corn Trials and Influence of Model Misuse on the Analysis Results

为研究不同方差协方差结构线性混合模型对区试分析的影响。采用2002-2006年中国40个玉米区试组数据,用7种不同结构的线性混合模型分别对数据进行拟合分析并利用信息量准则选择每组数据的最佳模型,对在不同方差协方差结构模型下区试品种效应估计与评价的差异进行探讨。结果表明,不同方差协方差结构模型对试验数据拟合的效果不一样,没有一个模型对所有试验资料均是最佳的;AMMI-1和FA(2)结构模型作为最佳模型的百分率最大,适用性最广泛;不同结构模型对区试品种效应的估计及评价不同,FA(1)模型的稳健性最好,方差分析模型对品种进行差异显著性分析时的误差不同程度地大于其他结构模型。不同方差协方差结构线性混合模型适用于不同试验数据,模型误用对品种效应的估计和评价有较大影响;在实际区试中,应利用信息量准则选用最佳结构模型进行分析,以提高作物区试分析及品种评价的准确性。

Probe the applicability of different liner mixed model with variance-covariance structures in variety regional test and misuse model influence on the analysis results. Use the data of 40 groups material in corn variety regional trial from the year 2002 to 2006 in China, the applicability of different liner mixed model with variance-covariance structures in variety regional test and the performance difference during the linear mixed model with different variance-covariance structure were assessed. Using different liner mixed model with variance-covariance structures, the data fitting effect has difference. There has no model always be the best for all the trial materials. The AMMI-1 and FA（2） models which have different variance-covariance simple structures are used widely. Different structural models give different results for estimation and evaluation of the effects of different varieties in re- gional test, Different liner mixed model with variance-covariance structures are applicable for different trial data sets. Model misuse can make a greater influence on the estimation and evaluation of varieties. Therefore, in olqrder to improve the accuracy of regional test analysis and variety evaluation, it is necessary to choose the best model which is abtained according to the information criterion and likelihood-ratio-test in the actual corn variety regional trial.