线性混合效应模型的惩罚变量选择

目的研究基于惩罚的线性混合效应模型变量选择原理和方法。方法对线性混合效应模型中的固定效应施加惩罚,采用Lasso和SCAD进行变量选择,通过两步迭代算法估计惩罚似然,利用BIC原则选择惩罚参数。通过广泛的模拟研究评价Lasso和SCAD在线性混合效应模型变量选择中的性质表现,并应用于真实数据的数量性状位点选择。结果模拟研究和实际应用显示,在线性混合效应模型中,两步迭代算法简单可行,基于惩罚的变量选择方法能够有效识别有意义的协变量。结论基于惩罚的策略为线性混合效应模型提供了行之有效的变量选择方法。

Variable Selection for Linear Mixed Effects Model Via Penalization Approaches

Objective To investigate variable selection approaches for linear mixed effects model via penalization-based strategies. Method Lasso and SCAD were used to select important variables for linear mixed effects model, a new two-step iteration algorithm was developed to maximize the penalized likelihood, and the penalization parameter was chosen via the BIC procedure. Extensive simulations were implemented to evaluate the performance of the proposed approaches for variable selection. An application to quantitative trait loci was given to demonstrate these penalization approaches. Results Simulations and application have shown that the proposed two-step iteration algorithm is effective and feasible for maximization the penalized likelihood and the penalization-based methods are promising and powerful approaches for variable selection of linear mixed effects model. Conclusions Penalization-based strategies are powerful approaches for variable selection of linear mixed effects model.