Second-stage nMDS (Morlaix macrofauna)
The illustration below has calculated all combinations of species (sp), genus (gn) and family (fm) level data, under no transform (no), square-root (sqr), fourth-root (4th), log(x+1) (log) transforms and reduction to presence/absence (pa), with similarity sheets sp-no to fm-pa. [Actually, all these have already been calculated for you, as PRIMER format *.sid files in the directory C:\Examples v7\Morlaix macrofauna\Morlaix similarities, and you can open as many of them as you need into the workspace in one batch by File>Open, highlighting them all and taking Open]. With one of these as the active matrix (it does not matter which – one of the very few routines for which that is true), run Analyse>2STAGE>(Data•Multiple matrices)>(Other resemblance matrices: ✓fm-4th & ✓fm-log & ✓fm-no & …) & (Correlation method: Spearman rank). This returns a second-stage resemblance sheet of matrix correlations $\rho$, all of which are positive, with some very close to 1 (e.g. species and genus level under no transform; 4th root and log transforms for any of the taxon levels, etc), indicating robustness of the conclusions to those particular choices. Now run nMDS on this matrix to obtain the 2nd stage MDS plot – note that the plot below has had its boundary shape changed with Graph>Special>Main>(Plot type•2D)>(Aspect ratio: 2.0). The main conclusions are that: transform choice and taxonomic level tend to have orthogonal effects (transformations run across the page, taxon levels run up the page); transform choice generally makes a larger difference to the outcome than taxon level (the exception being between 4th root and log, which are more or less equivalent – log being more severe than 4th root on very large abundances but less severe than 4th root for small counts); the difference between taxonomic levels increases with the severity of the transformation. The latter is to be expected, since untransformed analysis tends to be dominated by a handful of species with the largest abundances – when these are in different genera or families their contribution is unchanged by aggregation. Save Morlaix ws for later this section, and close it.
