Extended Data Fig. 9: MDS reliability tests.

The MDS pattern in Fig. 2 (using all nodes in our phylogeny and the geomean statistic) is largely robust to the statistic phylogeny or procedure used, including: (a) mean, (b) median or (c) harmonic mean; temporal selection of input nodes: (d) restricted to <=65 Ma or (e) > 3 and <=65 Ma; (f) the mean of 100 jackknifed (90% without replacement) datasets (see methods for details); (g) addition of dummy samples to reflect the known distribution of species across biomes, and (h) the use of a non-ultrametric (RAxML v8.1.20) phylogeny with branch lengths based on site substitution rates (rather than ages). The position of the Arctic bathyal biome is labile reflecting its varying relationship to Arctic shelf and Eastern Atlantic bathyal biomes. The temperate shelf biomes are notably more dispersed when older nodes are excluded (d, e) or down-weighted (c).