Tree-structured modelling of varying coefficients

被引：7

作者：

Berger, Moritz ^{[1
]}

Tutz, Gerhard ^{[2
]}

Schmid, Matthias ^{[1
]}

机构：

[1] Univ Klinikum Bonn, Inst Med Biometrie Informat & Epidemiol, Sigmund Freud Str 25, D-53105 Bonn, Germany

[2] Ludwig Maximilians Univ Munchen, Inst Stat, Akad Str 1, D-80799 Munich, Germany

来源：

STATISTICS AND COMPUTING | 2019年 / 29卷 / 02期

关键词：

Varying-coefficient models; Interactions; Recursive partitioning; Tree-based models; VARIABLE SELECTION; REGRESSION; BIAS; REGULARIZATION;

D O I：

10.1007/s11222-018-9804-8

中图分类号：

TP301 [理论、方法];

学科分类号：

081202 ;

摘要：

The varying-coefficient model is a strong tool for the modelling of interactions in generalized regression. It is easy to apply if both the variables that are modified as well as the effect modifiers are known. However, in general one has a set of explanatory variables, and it is unknown which covariates are modified by which variables. A recursive partitioning strategy is proposed that is able to deal with this complex selection problem. The tree-structured modelling yields for each covariate, which is modified by other variables, a tree that visualizes the modified effects. The performance of the method is investigated in simulations, and two applications illustrate its usefulness.

引用

页码：217 / 229

页数：13

共 31 条

[21]

Ripley Brian D., 1996, Pattern Recognition and Neural Networks

[22] Variable selection bias in regression trees with constant fits [J].