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—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特点,对现代回归设计的农业应用进行了讨论。 Based on the experimental error theory, through the division of error types and degrees of freedom, the essence of regression design is demonstrated by increasing the determinant of information matrix A to reduce the model error and improve the fit of the regression equation, but not with the experimental random error ; As to whether the fit test of the regression equation depends on whether there is a residual degree of freedom. Saturated D-Optimal Design A has the largest determinant and requires the least amount of experimental processing for a given regression mode and set-up iterations, making it the optimal design. This paper also discusses the agricultural application of modern regression design based on the characteristics of fertilizer field trials.