Coefficient-wise tree-based varying coefficient regression with vcrpart

Abstract

The tree-based TVCM algorithm and its implementation in the R package vcrpart are introduced for generalized linear models. The purpose of TVCM is to learn whether and how the coefficients of a regression model vary by moderating variables. A separate partition is built for each potentially varying coefficient, allowing the user to specify coefficient-specific sets of potential moderators, and allowing the algorithm to select moderators individually by coefficient. In addition to describing the algorithm, the TVCM is evaluated using a benchmark comparison and a simulation study and the R commands are demonstrated by means of empirical applications.