The metafor Package

A new version of the metafor package has been released on CRAN. This is version 4.8-0 and it includes a whole bunch of updates.

Quite importantly, the various forest() functions have received some TLC. In particular, the header argument is now TRUE by default, the y-axis is now created with yaxs="i", and the y-axis limits have been slightly tweaked in accordance. This has the potential to introduce some minor backward incompatibilities (in terms of the look of the resulting plot). I really try to avoid such changes, but some early design decisions really needed to be revised.

There are now also multiple styles for drawing the prediction interval (via the predstyle argument). Instead of a dotted line superimposed on the polygon (which is easily confused with a confidence interval for the average effect), predstyle="bar" draws a bar below the polygon, predstyle="shade" shades the bar in accordance with the density of the prediction distribution, and predstyle="dist" draws out the entire prediction distribution (and shades the regions beyond the prediction interval bounds and the region below/above zero). The forest plot below illustrates these different possibilities.

Another graphing function that received some attention is labbe(), which now has a bunch of new arguments for showing the confidence and prediction interval regions around the line that indicates the estimated effect. An example of this is shown below.

In terms of statistical methods, the selmodel() gains a subset argument, to specify a subset of studies to which the selection function should apply. The idea is that some studies may be assumed to be free of publication bias – for example, pre-registered studies – and we only want the selection model to apply to the non-preregistered studies (this idea is based on a recent talk by Robbie van Aert on the meta-analysis of non-preregistered and preregistered studies). Also, for the beta selection model, one can now also specify two steps values to fit a truncated beta selection model (this idea is based on a blog post by James Pustejovsky).

A whole bunch of new outcome measures have also been added to escalc(), including the probit transformed proportion (PRZ), the common language effect size (CLES), the area under the curve (AUC), and the coefficient of determination (i.e., $R^2$) in the case where predictor values are treated as fixed (R2F).

As a result of some recent questions on the R-sig-meta-analysis mailing list, I've also added a nice little helper function called deltamethod() to the package, which can be used to apply the (multivariate) delta method to a set of estimates.

Some smaller changes include the addition of an adjust argument to the anova() and predict() functions for adjusting p-values and interval bounds for multiple testing and the addition of the se() extractor function for extracting standard errors from model objects.

The full changelog can be found here.