[R] Inconsistent results between CAP (capscale) and RDA
philippe.janssen at irstea.fr
Tue Apr 28 17:04:17 CEST 2015
I have inconsistent results using vegan's capscale() and rda() on a hellinger distance matrix, based on presence/absence data matrix of 192 plants species (columns) x 70 sites (rows).
HellingerFloreDist <-dist(decostand(flore[12:204], method="hellinger"))
I first wanted to perform a Canonical Analysis of Principal coordinates (CAP) using capscale() function. However, based on Jari Oksanen comment, i.e. "Constrained Analysis of Principal Coordinates (CAP) is an ordination method similar to Redundancy Analysis (rda)... If called with Euclidean distance, the results are identical to rda, but capscale will be much more inefficient.", I decided to perform RDA using rda().
anova(CAP) #Test of the significance of the analysis
anova(CAP, by="axis", perm.max=999) #test axes for significance
anova(CAP, by="terms", permu=999) #test environment variables for significance
anova(RDA) #Test of the significance of the analysis
anova(RDA, by="axis", perm.max=999) #test axes for significance
anova(RDA, by="terms", permu=999) #test environment variables for significance
However, I was quite surprised by the results, i.e. the difference between the total inertia of CAP and RDA.
Results from CAP (capscale) :
Partitioning of squared Euclidean distance:
Total 35.957 1.00000
Constrained 2.052 0.05706
Unconstrained 33.905 0.94294
Results from RDA :
Partitioning of variance:
Total 1.5384 1.00000
Constrained 0.1096 0.07124
Unconstrained 1.4289 0.92876
Searching all day long for an explanation, I still don't understand those results.
Any explanation would be greatly appreciated.
Thanks in advance.
Doctorant - UR Ecosystèmes Montagnards
Irstea - Centre de Grenoble
+33 (0)4 76 76 28 79
[[alternative HTML version deleted]]
More information about the R-help