多元回归中因子分析和选择问题初步探讨

讨论了多元线性回归分析中因子选择存在的问题．建议在选择因子时，将Fa作为选择因子的门限值．而不作为假设检验的显著性界限．在逐步回归的基础上，提出以剩余标准差最小为准则的改进算法．文中计算了四例，全部取得最优因子．

Primary Study on the Analysis and Selection of Variables in Multiple Regression

This paper suggests that Fa be acted as the threshold for variable selection, but not as the limit for significance tests. Based on the stepwise regression procedures an im- proved algorithm is presented here with a criteria of the least residual standard deviation. The results of the four examples show that all the best variables have been obtained by this new approach.