多种分布下选择后变量显著性分析及其在CEPS数据中的应用

在使用变量选择方法选出模型后,如何评价模型中变量系数的显著性是统计学重点关注的前沿问题之一.文章从适应性Lasso变量选择方法的选择结果出发,在考虑实践中误差分布多样性的前提下,基于选择事件构造了模型保留变量系数的条件检验统计量,并给出了该统计量的一致收敛性质的证明过程.模拟研究显示,在多种误差分布下所提方法均可进一步优化变量选择结果,有较强的实用价值.应用此方法对CEPS学生数据进行了实证分析,最终选取了学生认知能力等10个变量作为影响中学生成绩的主要因素,为相关研究提供了有益的参考.

Variable Significance Test After Selection Under Various Distributions and Its Application in CEPS Data

After using the variable selection method to select the model,how to evaluate the significance of the variable coefficient in the model has always been one of the key issues in the statistical frontier.Based on the selection results of the adaptive Lasso variable selection method,under the assumption of the diversity of error distribution,the conditional statistics of the model retained variable coefficients are constructed based on the selection events,and the uniform convergence of the statistic is proved.Applying the method proposed in this paper,the CEPS student data is used to empirically analyze the main factors affecting the performance of middle school students.Finally,10 variables such as students’ cognitive ability are selected,which provides a useful reference for related research.