Skip to content

multinomial regression for community distances turned into a factor

1 message · Tim Richter-Heitmann

#
Dear list,

i need to evaluate the influence of temporal and spatial distance 
between two communities on their abundance weighted Raup Crick distance.

The abundance weighted Raup Crick distance (see Stegen et al, 2012, 13, 
ISMEJ)? is formulated as a difference between observed and the mean of 
simulated bray curtis distances,
is scaled to the interval -1 and +1, and knows by definition only three 
outcomes:

A value of <-0.95 represents homogenizing dispersal between two communities
A value of > 0.95 indicates a dispersal limitation between two communities
A value of < |0.95| is undefined ("undominated")

Thus, i have actually no interest in the actual value, but only in the 
outcome said value is associated with.
Since i have many observation pairs (70,000) i would like to avoid 
linear regression as in distance-decay regression.
What i want to know is:
i) if low spatial and/or high temporal distances faciliate homogeneity 
by dispersal and
ii) if high spatial distances and/or low temporal distances faciliate 
dispersal limitation.

Is it valid to turn my outcome variable into a factor with the 
aforementioned three levels, and then formulate two logistic regression 
models (for case >0.95 and for case <-0.95) or a multinomial regression 
model for all three levels,
e.g. multinom(Factor~Space*Time)?
If so, what would a model look like in R?
Thank you for your advice.