metaforest: Exploring Heterogeneity in Meta-Analysis using Random Forests

Conduct random forests-based meta-analysis, obtain partial dependence plots for metaforest and classic meta-analyses, and cross-validate and tune metaforest- and classic meta-analyses in conjunction with the caret package. A requirement of classic meta-analysis is that the studies being aggregated are conceptually similar, and ideally, close replications. However, in many fields, there is substantial heterogeneity between studies on the same topic. Classic meta-analysis lacks the power to assess more than a handful of univariate moderators. MetaForest, by contrast, has substantial power to explore heterogeneity in meta-analysis. It can identify important moderators from a larger set of potential candidates, even with as little as 20 studies (Van Lissa, in preparation). This is an appealing quality, because many meta-analyses have small sample sizes. Moreover, MetaForest yields a measure of variable importance which can be used to identify important moderators, and offers partial prediction plots to explore the shape of the marginal relationship between moderators and effect size.

Version: 0.1.2
Depends: R (≥ 3.4.0), ggplot2, metafor, ranger, mmpf
Imports: gtable, grid
Published: 2018-05-31
Author: Caspar J. van Lissa
Maintainer: Caspar J. van Lissa <c.j.vanlissa at gmail.com>
License: GPL-3
NeedsCompilation: no
Materials: README NEWS
In views: MetaAnalysis
CRAN checks: metaforest results

Downloads:

Reference manual: metaforest.pdf
Package source: metaforest_0.1.2.tar.gz
Windows binaries: r-devel: metaforest_0.1.2.zip, r-release: metaforest_0.1.2.zip, r-oldrel: metaforest_0.1.2.zip
OS X binaries: r-release: metaforest_0.1.2.tgz, r-oldrel: metaforest_0.1.2.tgz
Old sources: metaforest archive

Linking:

Please use the canonical form https://CRAN.R-project.org/package=metaforest to link to this page.