| 摘要: |
| 【目的】提出针对无重复试验数据进行分析的方法，并演示用SAS提供的Proc mixed程序分析无重复试验数据的过程。【方法】基于国际标准统计分析软件（SAS）的Proc mixed过程和实际试验设计，应用线性混合模型对作物育种中的无重复试验数据进行分析，在分析植物育种无重复试验数据特点和传统方差分析法应用缺陷的基础上，将线性混合模型分析法中利用协方差结构反映试验误差特征的原理，用于植物育种无重复试验数据的统计与分析中，采用模型拟合信息量准则选择最优的试验误差协方差结构模型，最后进行实证分析。【结果】线性混合模型分析法给出了植物育种无重复试验品系产量效应估计及其差异显著性测验的结果；采用效应估计值得到的品系效应排序及入选优良系，与直接采用品系产量观测值法所得的结果存在较大差异；误差协方差结构模型的选择，对无重复试验分析结果的影响较大。【结论】利用线性混合模型原理和SAS 软件的Proc mixed 程序，可实现对植物育种无重复试验数据的分析，能解决植物育种无重复试验品系间可比性差及不能进行效应差异显著性统计测验的问题。 |
| 关键词: 作物育种 无重复试验数据 模型选择 协方差结构 信息量准则 线性混合模型 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金项目(30571072);教育部留学回归人员基金项目（2004） |
|
| Empirical study on analyzing unreplicated trials data of crop breeding based on linear mixed model |
|
|
| Abstract: |
| 【Objective】The study provied a new method for analysing unreplicated data,and showed how to use the Proc mixed program provieded by SAS to analyse the unreplicated data.【Method】We used linear mixed models to analyze unreplicated experimental data from crop breeding,the corresponding program and process to analyze the data of unreplicated trials based on an international standard statistical software(SAS).The data characteristics of unreplicated trials and the drawbacks of classical analysis of variance were indicated.The principle of covariance structures of linear mixed models was used to analyze unreplicated trial data.The information criteria of model fit was used to select the covariance structure model.Finally,the emirical analyzsis was presented.【Result】The empirical study showed that the linear mixed model analysis provided estimates and tests of line effects for the data of unreplicated plant breeding trials;The outcome in ranking and selection of the lines tested based on the estimates was different from that based on the observed value of the lines;The choice of covariance models had important impact on the results of analysis of unreplicated trials.【Conclusion】The unreplicated experimental data could be analyzed by using the Proc mixed procedure in SAS based on the principle of the linear mixed model,so that the method can resolve the inferior comparable problems of unreplicated breeding lines and the problems of unreplicated trail data cannot be performed in significant test. |
| Key words: crop breeding unreplicated trial data model selection covariance structure information criterion linear mixed model |
