Skip to content
Prev 6473 / 7419 Next
Original Raw

pattern in a few interdependent variables

Martin Weiser 
Dear list members,

could You please advise me upon a not-so-multivariate problem?

We have run an experiment: we cultivated 10 plant species, each with 28 
replicates, we applied two treatments (say, fertilization F and adding 
the symbionts S) in a fully factorial design (F+S+, F-S+, F+S-, F-S-).

We were interested in one plant trait (T1) in response to those 
treatments, but we also measured trait2 (T2) - size of the plant - as 
the covariate.

We have analyzed the data in an design-based way:
T1 ~ T2 * F * S * Species (or something similar)

But, it turns out that in S+ treatments, the plants and the species 
varied a lot both in T1, but also in T2 (the size) and in the level of 
plant colonization by symbionts. I believe that the relation is not 
unidirectional, and I would like to see how (and if) species differ in 
the  "symbiont colonisation-size-trait1" interplay.

One of my ideas was to take the symbiont colonisation, individual size 
and trait1 as a multivariate matrix and run redundancy analysis (RDA; or 
CCA) with Species identity as the canonical (constraining) variable.

Another option could be to construct the regression model II for each of 
the relationships (symbionts~size, size~trait1, symbionts~trait1) and 
plot the slopes for each of the species.

Do you have any other idea how to inspect the patterns?

With kind regards,

Martin Weiser