Selection of Risk factors for Regression Analysis with Missing Data

摘要：

We propose a penalized estimated likelihood method for variable selection in the regression model with some risk factors being missing or mismeasured. Under some general regularity conditions, we show the consistency and asymptotic normality of the penalized likelihood estimator. We further propose that, for certain penalty functions with proper choices of regularization parameters, a sophisticated selection approach of tuning parameter is given by local quadratic approximation. Some simulation study results show that the proposed method performs well under reasonable finite sample results. We illustrate the proposed method by analyzing a data set from the Collaborative Perinatal Project (CPP).

关键词： Risk factors Estimated likelihood Nonconcave panelization Oracle property Asymptotic normality

作者: Liming Wang Jinhong You Yong Zhou

作者单位: Department of Statistics,Shanghai University of Finance & Economics Department of Biostatistics,University of North Carolina,USA Chinese Academy of Science

会议类型: 国际会议

会议名称: International Symposium on Financial Engineering and Risk Management(2008年金融工程与风险管理)

会议地点: 上海

会议语种:英文

页码: 29-33

在线出版日期: 2008-06-01（万方平台首次上网日期，不代表论文的发表时间）