Work

Recursive partitioning of growth curve models with generalised linear mixed-effects regression trees

Year: 2018
Type: dissertation
Abstract: In this study we investigated how to best fit a recursive partitioning growth curve model (RP-GCM) with the R package glmertree on longitudinal data.We used data on reading, math and science ability of children from kindergarten through eighth grade on five measurement occasions in the Early Childhood Longitudinal Study-Kindergarten class of 1998-99 (ECLS-K; National Center for Education Statistics, 2016).We used reading, math and science ability as the response variable and different child characteristics as partitioning variables (such as motor skills and gender).We investigated how time and clustering of observations can be accounted for in a RP-GCM.Specifically, we assessed the effect of two parameters in the recursive partitioning algorithm: I) Whether cluster-(C) or observation-level (O) parameter stability tests are employed for selecting partitioning variables and II) whether initializing model estimation with the random effects (R) or the tree structure (T) yields a more accurate and/or less complex model.The effects of the two parameters were assessed in both a random intercept (RI) and a random intercept and slope (RIS) model.In the RI models, CT and CR yielded higher predictive accuracy and lower complexity.In the RIS models OR performed best with highest predictive accuracy and lowest complexity.The best RI and RIS model yielded similar substantive results: The detected subgroups differed mostly in socioeconomic status, fine motor skills and race.Thus, these variables were the most important predictors of reading, math and science trajectories, but it should be noted that the subgroups differed more strongly in their baseline performance than in their growth rate over time.Although the results of the best RI and RIS models were comparable, the OR RIS model may be preferred over the CT/CR RI model, because of its lower MSE and tree size.Table of contents Introduction ....
Language: English
Open Access status: green
OpenAlex ID: W6986796343
Cites: 18
Cited by:
Related to: 10