中国农业科学2018，Vol.51Issue(4)：728-737,10.DOI:10.3864/j.issn.0578-1752.2018.04.012
三元肥效模型异方差诊断及其可行广义最小二乘回归建模研究
Heteroskedasticity Diagnosis and Feasible Generalized Least Squares Regression Modeling of Ternary Fertilizer Efficiency Model
摘要
Abstract
[Objective] This paper presented the new strategy to develop fertilizer efficiency models,so as to improve the scientificityoffertilizereffectmodelsandthesuccessrateofthefieldexperiment.[Method] According to the results from the 171 "3414" NPK field experiments with early rice,using the modem regression analysis theory,the reason of the ternary model heteroscedasticity and the statistical test methods were studied in the paper.Feasible generalized least squares (FGLS) was used to eliminate the heteroscedasticity effect.[Result] According to the professional characteristics of fertilizer efficiency models,the field experiment design and its management practices,soil fertility and its spatial heterogeneity were the key reasons for the models' heteroscedasticity.Based on the White test,Park test and KB test,the heteroscedasticity of the 25.15％ ternary fertilizer efficiency model showed the statistical significantly level.In view of the serious heteroscedastic of the field experimental data,FGLS regression model could convert part of field experimental results into significant level,whereas the Ordinary Least Square (OLS) could not;or the FGLS could convert the no-typical model into typical model;the probability of typical models increased from 27.68％ to 34.99％.For the typical ternary model,the recommended fertilization rate of FGLS regression model roughly was equate to the rate of OLS model.Compared to the OLS modeling,the FGLS regression modeling could decrease the Mean Square Error (MSE) greatly,at the same time,which also could decease the variance from the parameters.Thus the FGLS regression model could improve the fitting precision and the predication ability.[Conclusion] FGLS is an effect way to eliminate or ease the heteroscedasticity from the ternary quadratic polynomial model,and can obviously improve the success rate of the field experimental data modeling.关键词
早稻/氮、磷、钾肥/肥效模型/异方差/广义最小二乘法Key words
early rice/N, P, K fertilizer/fertilizer efficiency model/heteroscedasticity/feasible generalized least squares引用本文复制引用
孔庆波,章明清,李娟,许文江,章赞德,姚建族..三元肥效模型异方差诊断及其可行广义最小二乘回归建模研究[J].中国农业科学,2018,51(4):728-737,10.基金项目
国家自然科学基金项目(31572203)、福建省农业科学院PI项目(2016PI-31)、农业部测土配方施肥(2011-2015) （31572203）
