混合效应模型的多惩罚回归过程及其算法收敛性研究

Research of Multi-penalty Regression Process of Mixed Effects Models and Its Convergence

针对混合效应模型中固定效应与随机效应同时选择问题,提出了施加多个惩罚项的回归过程,同时给出了参数估计的交替迭代算法,并证明了算法的收敛性。针对两种特殊的多惩罚回归过程,分别利用计算机模拟数据进行了比较分析,结果显示新方法在各种不同条件下均有良好的表现,尤其是能处理高维稀疏的混合效应模型。最后通过一个实际数据演示了新方法的应用。

For the problem of selecting fixed and random effects in the mixed effects model,the paper proposes a multiple penalization regression process and gives the iterative algorithm of parameter estimation,the convergence of algorithm is also proved.For two kinds of special multi-penalized regression process,the paper makes a comparison analysis using the simulation data,the results show that the new method has good performance in different conditions,especially for its ability to deal with the high dimensional sparse mixed effect models.Finally,the application of the new method is demonstrated by a real data.