rstanarm 2.9.0-3 released on CRAN (major bug fixed)

[Ben Goodrich]

A new version of rstanarm is now on CRAN and its binaries have mostly been mirrored out. If you are not familiar with rstanarm, see

https://groups.google.com/d/msg/stan-users/HUwIufOT97E/bLp-2_bwAAAJ

This is the second official release of rstanarm and fixes a major bug that was found with lme4-style models that estimate group-specific coefficients, such as y ~ x + (1 + x | group). The correct model was being estimated but the group-specific coefficients were often mislabeled, so they might be mixed up with group-specific intercepts in the summary output. Also, ranef() would extract mis-labeled estimates and the posterior predictions may well be wrong. In short, if you used a previous version of rstanarm from CRAN or GitHub to estimate a model with group-specific coefficients, you should download the new version and reestimate that model.

Various other smaller bugs have been fixed as is mentioned in

https://cran.r-project.org/web/packages/rstanarm/NEWS

Also, there is a new vignette on pooling in hierarchical models that is available from browseVignettes(package = "rstanarm") or from

https://cran.r-project.org/web/packages/rstanarm/index.html

So, we would recommend that everyone update to the new version of rstanarm by executing

update.packages("rstanarm")

or, if you have not installed rstanarm before,

install.packages("rstanarm")

Due to some changes in other packages, this version of rstanarm requires R version 3.2.0 (which was released in April 2015) or later. In particular, it imports from the most recent version of nlme, which is a recommended package that comes with R. Hence, the old version of nlme might be installed in a different directory than rstanarm and you might not have write permissions to the directory that nlme is in. If so, ask your system administrator to update nlme or install nlme to a personal directory that appears first on .libPaths().

Ben