2005 Volume 17 Issue 2 Pages 127-146
The conventional methods for analysis of longitudinal data are not flexible enough for exploring the nonlinear underlying data structures. But interactions among time and other covariates exist in many longitudinal data, the way of evaluation of the interactions will show the superiority or inferiority of the analysis methods. Furthermore, time trend of observations has nonlinear structure in many cases, describing this structure is connected with describing the mean structure of observations correctly. In this paper, as a convenient method to explore the nonlinear structure and, especially, interactions structure in longitudinal data, we focus on Multivariate Adaptive Regression Splines (MARS). And we present MARS for analysis of longitudinal data (L-MARS), and we demonstrate the methodology. Moreover, we evaluate the trial performance through case examination and the simulation.
