Characterizing direction and magnitude of community change in response to a continuous predictor?

Dear Brian and Pedro,

I think it is not a good idea to regress an unconstrained ordination axis,
especially a PCoA axis, on a linear predictor to test for the effect of an
external gradient on species composition. If the compositional turnover is
high enough, the gradient will need several PCoA axes to unfold
between-site distances in a metric space, so the first axis will not
represent the gradient in full length. This property of distance-based
ordinations is widely known as the horseshoe effect.
Why not use a constrained ordination (CCA, RDA, db-RDA) instead, with a
formula SPECIES ~ PREDICTOR separately for each region or SPECIES ~
PREDICTOR * REGION for the entire sample? Then you could partition the
total variation in species data into portions explained by the PREDICTOR
(or by REGION).

Another possibility is to relate between-site compositional distances with
environmental distances using either correlation or regression. See methods
for 'distance decay' modelling as a form of beta diversity.

Best regards,

Attila

Pedro Pequeno <pacolipe at gmail.com> ezt ?rta (id?pont: 2022. j?n. 7., K,
20:14):

Characterizing direction and magnitude of community change in response to a continuous predictor?

Thread (4 messages)