GrowthCurveDemo: Artificial dataset for partitioning of linear growth curve...

In glmertree: Generalized Linear Mixed Model Trees

Artificial dataset for partitioning of linear growth curve models

Description

Artificial dataset to illustrate fitting of LMM trees with growth curve models in the terminal nodes.

Usage

data("GrowthCurveDemo")

Format

A data frame containing 1250 repeated observations on 250 persons. x1 - x8 are time-invariant partitioning variables. Thus, they are measurements on the person (i.e., cluster) level, not on the individual observation level.

person

numeric. Indicator linking repeated measurements to persons.

time

factor. Indicator for timepoint.

y

numeric. Response variable.

x1

numeric. Potential partitioning variable.

x2

numeric. Potential partitioning variable.

x3

numeric. Potential partitioning variable.

x4

numeric. Potential partitioning variable.

x5

numeric. Potential partitioning variable.

x6

numeric. Potential partitioning variable.

x7

numeric. Potential partitioning variable.

x8

numeric. Potential partitioning variable.

Details

Data were generated so that x1, x2 and x3 are true partitioning variables, x4 through x8 are noise variables. The (potential) partitioning variables are time invariant. Time-varying covariates can also be included in the model. For partitioning growth curves these should probably not be potential partitioning variables, as this could result in observations from the same person ending up in different terminal nodes. Thus, time-varying covariates are probably best included as predictors in the node-specific regression model. E.g.: y ~ time + timevarying_cov | person | x1 + x2 + x3 + x4.

References

Fokkema M & Zeileis A (2024). Subgroup detection in linear growth curve models with generalized linear mixed model (GLMM) trees. Behavior Research Methods. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.3758/s13428-024-02389-1")}

See Also

lmertree, glmertree

Examples

data("GrowthCurveDemo", package = "glmertree")
head(GrowthCurveDemo)

## Fit LMM tree with a random intercept w.r.t. person:
form <- y ~ time | person | x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8
lt.default <- lmertree(form, data = GrowthCurveDemo)
plot(lt.default, which = "tree") ## yields too large tree
VarCorr(lt.default)

## Account for measurement level of the partitioning variables:
lt.cluster <- lmertree(form, cluster = person, data = GrowthCurveDemo)
plot(lt.cluster, which = "tree") ## yields correct tree
plot(lt.cluster, which = "growth") ## plot individual growth curves not datapoints 
coef(lt.cluster) ## node-specific fixed effects
VarCorr(lt.cluster) ## with smaller trees random effects explain more variance

## Fit LMM tree with random intercept and random slope of time w.r.t. person:
form.s <- y ~ time | (1 + time | person) | x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8
lt.s.cluster <- lmertree(form.s, cluster = person, data = GrowthCurveDemo)
plot(lt.s.cluster, which = "tree") ## same tree as before
coef(lt.cluster) 
VarCorr(lt.s.cluster)

glmertree documentation built on June 28, 2024, 5:08 p.m.