高维纵向数据的亚组识别方法及应用

Subgroup identification of high-dimensional longitudinal data and its application

在高维纵向数据建模的背景下,构建了一种数据驱动的亚组识别方法,将极大极小凹惩罚方法和同质划分方法结合起来,并基于二值分割法对回归系数之间的变点进行识别。通过统计模拟实验,将所构建的亚组识别方法和其他6种方法进行对比,检验了所构建的亚组识别方法的性能。通过一个实例数据的分析,即国内各地区生产总值和产业结构的建模,进一步阐述了该方法的优势。

This article introduces a data-driven subgroup identification method in the context of high-dimensional longitudinal data. It combines the minimax concave penalty and the homogeneous pursuit for detecting the change points between the regression coefficients by the binary segmentation method. In addition, the proposed subgroup identification is compared with other competitive methods. Some simulation studies are carried out to measure the performances of the proposed subgroup identification method and competitors. Finally, a real data analysis on the relationship between regional GDP and industrial data has been conducted, in which the advantages of the proposed method are elaborated in details.