
植物营养与肥料学报2010,16(6):1535-1538 Plant Nutrition and Fertilizer Science
关于饱和D一最优设计的再商榨
王兴仁',申建波1*,刘小虎2
(1中国农业大学资源与环境学院,植物-土壤相互作用教育部重点实验室,农业部植物营养学重点实验室,北京100193:
2沈阳农业大学土地与环境学院,辽宁沈阳110161)
摘要:以试验误差理论为基础,通过误差类型和自由度的划分,论证回归设计的实质是通过增大信息矩阵A的行列式来减少模型误差,提高回归方程的拟合性,而与试验随机误差无关;至于能否对回归方程进行拟合性检验,则取决于是否有剩余自由度。饱和D一最优设计A的行列式最大,在给定回归模式和设置重复的条件下所需的试验处理数也最少,因而是最优设计。本文还结合肥料田间试验待点,对现代回归设计的农业应用进行了讨论。
关键词:饱和D一最优设计;模型误差;信息矩阵行列式;肥料田间试验
中图分类号:Q24
文献标识码:A
文章编号:1008505X(2010)06153504
Reconsideration of saturated D-optimal design
WANC Xing-ren', SHEN Jian-bo'', LIU Xiao-hu*
(I College of Resoturces and Enwironmental Sciences, China Agricultural Uniersity/Key Laboratory of Plant-Soil Interactions,
MOE/Key Laboratory of Plant Nutrition,MOA,Bejing 100193,China;
2 College of Land and Environmental Sciences, Shenyang Agricultural Unitersity, Shenyang 110161, China)
Abstract: Based on experimental error theory, the objective of regression design is to improve fitness of equation through enlarging the determination of error type information matrix A and thus reduce model eror on basis of clas-sification of error types and degree of freedom, which is independent of random error. Whether the significant difference of regression equation in fitness can be tested is dependent on residual degree of freedom. Saturated D-optimal design has the maximum determinant of information matrix A. Under the given regression model and num-ber of replication, there was a minimum number of treatments in the experiment of Saturated D optimal design which is characterized of optimization. Agricultural application of modern regression design was discussed in combi-nation with properties of fertilizer field experiment.
Key words: Saturated D-optimal design; regression design, model error; determinant of information matrix,
fertilizer field experiment
回归饱和D一最优设计,由于其在无重复且无剩余自由度的条件下无法统计检验而引起颇多争议,认为D一最优是不存在的,因而D一最优设计是最不优设计。因此本文拟对D一最优设计的优良性及与此有关的肥料试验设计问题作进一步讨论。
收稿日期:2010-05-26
接受日期:2010-08-10
误差类型及其与试验设计的关系 1
1.1误差类型
试验误差即真值与观测值之间的偏差,可以划分或分解为三类:随机误差、系统误差和模型误差(或称失拟误差)[-2)。其中,随机误差是受试验材料、试验条件等偶然因素影响而产生的变异,其大小
基金项目:农业部行业专项计划(200803029-01-02,200803030);科技部重大基础研究"973"项目课题(973-2007CB109302);国家科技支撑
计划课题(2006BAD25B02)资助。
作者简介:王兴仁(1940一),男,辽宁东港人,教授,长期从事植物营养和施肥技术及其研究方法的教学和科研工作。
E-mail; wangxingrenfe 163. com 万方数据
*通讯作者E-mail:jbehen@cau.edu.cn