ABSTRACT
There are many R packages for meta-analysis and we have so far illustrated
three commonly-used ones: rmeta by Lumley (2009), meta by Schwarzer (2010)
and metafor by Viechtbauer (2010), in the previous chapters. As seen from
the illustration in the previous chapter, all three packages can serve as“general
purpose” packages for arbitrary effect-size and different outcome measures to
fit fixed-effects and random-effects meta-analysis models. From our experience,
all three packages are easy to use for meta-analysis with metafor having more
methods implemented. For example in random-effect meta-analysis, rmeta
only implemented the DerSimonian-Laird estimator to estimate the between-
study variance of τ2 where metafor and meta included several other methods,
such as the restricted maximum-likelihood estimator, the maximum-likelihood
estimator, the Hunter-Schmidt estimator, the Sidik-Jonkman estimator, the
Hedges estimator, and the Empirical Bayes estimator. In addition, metafor
has a great feature for meta-regression as illustrated in Chapter 7 and can
be used to include multiple continuous or categorical regression covariates as
well as mixed-effects models, whereas meta can only include a single categor-
ical regression covariate in fixed-effects, and no meta-regression is included in
rmeta. A comprehensive comparison and discussion of these three packages
can be found in Table 2 of Viechtbauer (2010).