Small updates to geomorph (plotOutliers)

[Mike Collyer]

Dear geomorph community,

I wish to alert you to some small but perhaps important updates to the plotOuliers function.  Here is a list of updates:

1. Use of a power-transformation to find the upper limit for detecting potential outliers, based on Tukey’s method (box-plot upper fence).  Because a distribution of shape distances in inherently skewed, the transformation assures a better estimation of an upper limit on symmetrical data, which can be back-transformed to a distance.

2. Elimination of the median and lower limit in plots, as they offer nothing useful for outlier detection but make the plot unnecessarily busy.  Also, thank you to Frédéric Santos for catching that the limits were mislabeled as quartiles.

3. An option to use a portion of principal components (PCs) to calculate distances.  This might be useful for, e.g., examining digitizing error made on repeated digitizations of the same research subjects along PC2, if PC1 tends to capture variation among subjects.  The argument default, PC = NULL, means using all dimensions of shape, which is consistent with the previous version of plotOutliers.

4. A change of “Procrustes distance” to just “Distance” as an axis label in the plot, owing to the change made in point #3.  These distances are only Procrustes distances if all coordinates are used, and if they are not projected configurations (tangent space data).

The updated function is currently available on Github (https://github.com/geomorphR/geomorph), pending submission of a package update to CRAN.  Either devtools::install_github() or remotes::install_github() can be used to install the package.  The repository address is “geomorphR/geomorph” and the git reference is “Stable”, when using these functions.

Happy computing!

Mike (on behalf of the geomorph team)