About a month ago, I gave a talk at the Research Synthesis 2019 conference, organized by the Leibniz Institute for Psychology Information (ZPID), in Dubrovnik, Croatia. In this talk, I describe the past, present, and future of the metafor package. Although I find it extremely embarrassing to see me on video like this, you can see the talk here. If you just want the slides, you can download them here. Abstracts and several other recorded talks from the conference can be found here.
Quite frequently, the question arises how one should interpret the results from models that include an intercept term (as is done by default in the model fitting functions) and those where the intercept term was removed. In part, this issue was already addressed in two entries under the tips and notes section (Testing Factors and Linear Combinations of Parameters and Models with Multiple Factors and Their Interaction), but I now wrote up a further discussion of this that uses a simpler example: Meta-Regression Models With or Without an Intercept.
I've added another example of a contour-enhanced funnel plot to the section showing various plots and figures that can be created with the metafor package. See here. This one provides a very close re-creation of Figure 2(b) from Peters et al. (2008). Note that you will have to install the development version of the metafor package for this code to work, since I just added the dataset to the package.
After almost 2 years, I have finally released a new version of the metafor package (version 2.1-0). Lots of updates have been made to the package in the meantime (if I didn't miscount, 195 commits since the last version). The package now includes 33,146 lines of code (strictly speaking, 23,769 once we remove empty lines and 20,682 if we also remove comment lines). The full changelog can be found here.
Some of the more interesting (and user-visible) additions are the
vif() function (for variance inflation factors); the
reporter() function (for dynamically generating analysis reports; note that this is work in progress); output can be styled if the crayon package is loaded;
rstudent() can compute cluster-level Cook's distances and standardized/studentized residual values for
rma.mv() now allows for continuous-time autoregressive and various spatial correlation structures;
escalc() can now compute the coefficient of variation ratio and the variability ratio for pre-post or matched designs; and various improvements to the
I have updated the page on how to conduct model selection and multimodel inference (see here), which now also discusses how to use the MuMIn package for these analyses. In addition, I've added a new page, discussing the use of the mice package in combination with metafor for multiple imputations in the context of a meta-analysis (see here).