基于变量选择事件的新弹性网方法

Elastic Net Method Based on Variable Selection Events

从弹性网(Elastic net)方法所选择的模型出发,构造基于模型选择条件下的系数的精确分布,并通过分布进行推断从而得到检验系数显著性的P值及模型系数的置信区间等.通过方法可对传统弹性网方法所选模型做进一步调整,模拟研究说明了本文所提方法在变量选择中的适用性,如对噪声变量有较强的识别能力等.在实证分析中,使用基于变量选择事件的弹性网方法对我国劳动者工资收入的影响原因进行了筛选,分析表明在传统弹性网方法选取的解释变量中,宗教活动频率、工龄、身体健康程度以及个体身高不是影响劳动收入的最主要原因,可依据实际情况剔除这些变量,减少研究成本且提高分析效率,在实际应用中有一定的参考价值.

Based on the model selected by the Elastic net method,this paper constructs the exact distribution of the coefficients under the condition of model selection,by using the distribution,we can obtain the significant p-value and the confidence intervals of the model coefficients.This method can be used to adjust the model selected by the traditional elastic net method.The simulation study shows the applicability of the proposed method in variable selection,such as strong identification of noise variables.In the empirical analysis,we use the proposed method to screen the explanatory variable of the income of laborers in our country.The analysis shows that among the explanatory variables selected by the traditional elastic net method,the frequency of religious activities,length of service,health condition,individual height is not the main reason that affects laborer’s income.These variables can be removed according to the actual situation to reduce the cost of research and improve the efficiency of analysis,which has certain reference value in practical application.