摘要
本文对于回归中自变量的选择提出了一个新方法,这个方法是逐步回归的推广。假定自变量中有一组k个变量,如果因变量与组中每个自变量的相关系数都很大,而组中任意两个自变量之间的相关系数都很小,则称这组k个自变量为一个基组。传统的逐步回归每次吸收或剔除一个自变量,而本文提出的方法则对每个可能的k元基组考虑吸收或剔除。本文通过一个实例说明,与传统的逐步回归方法相比,新方法更为灵活并能得到更好的结果。 关键词 变量选择(variable selection),逐步回归(stepwise regression),变量的基(baseof variables)。
This paper presents a new methed of variable selection in regression. It is a extension of the stepwise regression. A group of k variables is called a base if the coprelation coefficients of response and each variable in the group are big, and the correlation coefficients between the pairs of variables in the group are small. The methed is starting from each possible k-variable base other than only one variable in the traditional stepwise regression. A real example indicated that the new method is more flexible and can get better results than traditional stepwise regression.
