Variable screening and model selection in censored quantile regression via sparse penalties and stepwise refinement

Variable Screening;Censored Data;Quantile Regression;Least Absolute Selection and Shrinkage Operator (LASSO);Smoothly Clipped Absolute Deviation (SCAD);Portnoy;Peng and Huang;Stepwise Regression;Bidirectional;Backward;Left Censoring;Random Censoring

Gan, Lu

关键词: Variable Screening;    Censored Data;    Quantile Regression;    Least Absolute Selection and Shrinkage Operator (LASSO);    Smoothly Clipped Absolute Deviation (SCAD);    Portnoy;    Peng and Huang;    Stepwise Regression;    Bidirectional;    Backward;    Left Censoring;    Random Censoring;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/49692/Lu_Gan.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Many variable selection methods are available for linear regression but very little has been developed for quantile regression, especially for the censored problems.This study will look at the possibilities of utilizing some existing penalty variable selection methods on censored quantile regression problems.In the situation when censored values are not known for each observation, it is common to model the censoring as random.Under the assumption that y_i and C_i are conditionally independent given x_i, we use the random censored quantile regression Portnoy estimators (2010). This method simplifies the censored problem into a weight problem. When combined with the penalized regression method: LASSO and SCAD, one can perform variable screening for the censored data at quantiles of interest.Furthermore, we establish the asymptotic property, and illustrate the methodology in the context of ultrasound safety study.