Paper on linear growth curve modeling with lmertree() published:
Fokkema M, Zeileis A (2024). "Subgroup Detection in Linear Growth
Curve Models with Generalized Linear Mixed Model (GLMM) Trees."
Behavior Research Methods, 56(7), 6759-6780.
doi:10.3758/s13428-024-02389-1
Reduce precision in tests in order to avoid NOTEs in CRAN checks.
Added functions cv.lmertree() and cv.glmertree(), implementing honest estimation of
model coefficients per Athey & Imbens (2016)
Fixed bug in coef() method betamertree().
Additional arguments to lattice::dotplot can now be passed to
plot.lmertree()/plot.glmertree().
Dedicated plotting function for partitioning growth curves has
been added, see which = "growth" in ?plot.lmertree and
?plot.glmertree.
Minor bug fixes.
Experimental function betamertree() added, for partitioning
mixed-effects beta regression trees.
Bug fix for missing data: Listwise deletion over all predictor, response, and splitting variables is now applied. Otherwise, global and local parts of (g)lmertree model could be estimated based on a different sample size or an error could occur.
Fixed bug: If dot . was used in formula for specifying partitioning
variables, tree fitting would pick up tree structure from previous
iteration. Fixed now.
Fixed bug: When type = "simple" in call to function plot.lmertree()
and plot.glmertree() is specified, further arguments are now passed
correctly.
Bug fix in predict() method: Erroneous allocation of new observations
to tree nodes could occur, which has been fixed now.
New plotting method for (g)lmertrees. Arguments which, type, and fitted
now support wider range of plots. E.g., fitted argument provides
different ways to compute fitted values in terminal panel plots.
E.g., type argument now supports the plotting of fixed-effects
coefficients with standard-error bars.
Two artificial example datasets added to illustate fitting of constant
fits (MHserviceDemo) and growth-curve models (GrowthCurveDemo) in
terminal nodes.
New sections added to vignette, illustrating how mixed-effects regression trees with constant fits and growth curve models in terminal nodes can be fitted.
Functions glmertree() and lmertree() now take offset arguments.
Argument ranefstart of functions glmertree() and lmertree() can now be
set to TRUE. As a result, the random effects will be estimated
before fitting the tree in the first iteration. This may yield better
results when random effects are expected to be substantial.
Functions glmertree() and lmertree() now take cluster argument, an optional
vector with cluster IDs to be employed for clustered covariances in the parameter
stability tests.
Bugs fixed: Argument dfsplit is now passed correctly to tree fitting functions,
additional arguments are now passed correctly to lmer() and glmer().
Arguments lmer.control in lmertree() and glmer.control in glmertree()
are now actually passed to lmer() and glmer() internally.
First CRAN release of the glmertree package for fitting generalized linear
mixed-effects model trees in R. For an introduction to the underlying methods see
Fokkema, Smits, Zeileis, Hothorn, Kelderman (2015). Detecting Treatment-Subgroup Interactions in Clustered Data with Generalized Linear Mixed-Effects Model Trees. Working Paper 2015-10. Working Papers in Economics and Statistics, Research Platform Empirical and Experimental Economics, Universitaet Innsbruck. URL https://EconPapers.RePEc.org/RePEc:inn:wpaper:2015-10
The package is under development on R-Forge at https://R-Forge.R-project.org/projects/partykit/