New version of vegan released to CRAN (1.17-9)

Gavin Simpson

Yesterday Jari packaged up the latest release in the current stable branch of the vegan package. Version 1.17-9 of vegan is now on CRAN as a source tarball with binaries for MS Windows and MacOS X to follow soon.

New releases in the current, stable branch of the package tend to be for bug fixes but we sometimes include new functionality considered low impact and important enough to warrant early release. This release sees a fix to a bug in the anova() method for rda(), cca() and capscale(), which would produce incorrect results when partial models were supplied and testing was performed by axes (e.g. anova.cca(..., by = "axis")).

From a user’s perspective, there were three important additions to vegan with this release:

ordisurf() gained several new arguments to allow control of the methods used by gam() when fitting response surfaces to ordination configurations. The smoothness selection method can now be specified via a new argument 'method' and an additional penalty can be applied to smooth terms so that they can be penalized out (i.e. removed from) the model if there really is no relationship between the response and the ordination configuration. If this happens, ordisurf() won’t draw a response surface at all.
Automatic plotting of the fitted surface as a side-effect of ordisurf() can now be turned off, and a new plot() method allows later plotting of response surfaces without having to refit the entire model again.
metaMDSrotate() can now rotate an nMDS ordination with an arbitrary dimensional solution whereas before it would only rotate 2-D solutions.
diversity() (and related functions: rarefy(), rrarefy(), and specnumber()) are now more intuitive and work with vector inputs. As a result many sites/samples can be analysed at once now.

Full details of the release can be seen in the NEWS file, the relevant portion of which appears at the bottom of this post. We’ve got some exciting new features planned for the next version of vegan, which I’ll blog about when they start appearing in the development version on R-forge.

            VEGAN RELEASE VERSIONS 
            ======================

                   CHANGES IN VEGAN 1.17-9

- anova of cca/rda/capscale results gave wrong results in partial
  models. The bug was introduced in vegan 1.17-7. 

- diversity and related functions rarefy, rrarefy and specnumber
  now accept vector input. Earlier a single site had to be
  analysed either as a single-row matrix or using the non-default
  setting MARGIN = 2.

- drarefy: new function that returns a matrix of probabilities
  that a species occurs in a rarefied sample of a given size.

- metaMDS: it is possible to supply a starting configuration with
  argument 'previous.best'. A previous metaMDS or isoMDS result can
  also be given as a starting configuration. If the starting
  configuration has a higher number of dimensions than requested,
  the extra ones are dropped, and if the starting configuration
  has fewer dimensions, random scores for extra dimensions will be
  added. This may help in running metaMDS over a range of
  dimensionalities. 

- metaMDSrotate: can now rotate metaMDS solutions with any number
  of dimensions so that the first axis is parallel to a fitted
  environmental vector. Previously, only two dimensional solutions
  worked.

- ordilabel: gained argument 'xpd' that allows labels outside the
  plot area. This allows labels above axes, for instance.

- ordisurf: gained several new arguments to control the mgcv::gam
  fitting. Also gained an argument to suppress plotting, and a new
  plot method. The fitted model can be specified with a formula. 

- prestonfit and friends: default is now 'tiesplit = TRUE' (which
  was a new feature introduced in vegan 1.17-8).