Up to now, I had been using a sort of home-brewed version control system for the development of the metafor package. In the meantime, I was watching all the cool kids jump on the git / GitHub bandwagon and was really starting to feel left out. But no longer! I have finally moved development to git / GitHub, so now I am cool again, right?!? I thought so ...
Anyway – you can find the master (i.e., development) branch at https://github.com/wviechtb/metafor. I'll keep things simple and just have that one branch (for now), with official releases still made via CRAN once a sufficient number of changes have been accrued and/or a particular milestone has been reached.
A new version of the package is up on CRAN now. The most interesting changes from my perspective are the addition of the
robust() function for obtaining (cluster) robust tests and confidence intervals for
rma.mv models (along the lines of the Eicker-Huber-White method) and that
confint() now works for models fitted with the
rma.mv() function (providing profile likelihood confidence intervals for variance components and correlation parameters). Full changelog can be found here.
I just noticed that the JSS article about the metafor package has now been cited more than 1000 times according to Google Scholar. My thanks to all the people who have cited the package – I very much appreciate it!
Had to update the package due to some tests failing on Sparc Solaris. As far as I can tell, the issue seems to be resolved now (see the CRAN Package Check Results). Also, after some extensive testing, I decided to change the default optimizer for
nlminb(). This is in principle a non-backwards compatible change, which I am not crazy about doing, but really a necessary one. Extensive testing indicated that
nlminb() (and also
"BFGS") is typically quicker and more robust than the implementation of the Nelder-Mead algorithm in
optim(). Full changelog here.
A new version of the metafor package is out (version 1.9-6). The most important changes from my perspective are:
rma.mv()function now allows for two terms of the form
~ inner | outer. I added this so that users can fit network meta-analysis models with random inconsistency effects (more details about this in a publication that is under preparation), but this also comes in handy when dealing with other complex data structures.
profile()functions now can do parallel (multicore) processing (which is especially relevant for
rma.mvobjects, where profiling is crucial and model fitting can be slow).
There are of course a number of additional changes and improvements to the package – the full changelog can be found here.
This has been something I meant to add for quite some time: A to-do list and a description of planned features that I would like to implement in the metafor package at some point in the future. The list is a bit disorganized at the moment and not complete, so I'll be updating this on and off.
I previously had added a page under the tips and notes section illustrating how to assemble a dataset for a meta-analysis of standardized mean differences from various pieces of information. I have now added another page illustrating how to assemble a dataset for a meta-analysis of (log) odds ratios (and the same principle also applies to risk ratios).
I have started to put the development version of the metafor package up on this website. You can download and/or install it directly from the website as described under the download and installation section. Note that the version number of the development version is the same as that of the next upcoming official release. However, the actual code may still change before the next official release is made available via CRAN.
Under the Tips and Notes section, you will now find another entry, this one illustrating how to use the metafor package in combination with the glmulti package for model selection and multimodel inference based on an information-theoretic approach.
I added a section to the website with links to other websites that are in some way relevant/related to the metafor package or other software packages for meta-analysis.
Just a quick addition to the analysis examples section. I made an entry for the analyses I conducted in my 2007 article for the Zeitschrift für Psychologie (Journal of Psychology), which illustrates some basic applications of the random/mixed-effects model (Viechtbauer, 2007).