I conducted a partial CCA analysis in Canoco 5 on a dataset with environmental variables belonging to three distinct spatial scales. When doing variation partitioning it is normal to get some overlap in variation that is accounted for by two or three different components (or spatial scales). This can then be used to illustrate the overlap graphically by overlapping circles (although it is very difficult if not impossible to get it up to the right scale). What does it mean however if the programme returns a negative value for overlap and how do I interpret that ?
It is not possible to have negative numbers for species abundance in CCA, so recheck the data and analysis
Lets say you apply variance partitioning (VP) on two groups of parameters and the individual and combination of the groups (models) explains: A = 10%, B = 10% and AB = 25%.
Explained by A = AB - B = 25 - 10 = 15%
Explained by B = AB - A = 25 - 10 = 15%
Explained by AB = A + B - AB = 10 + 10 - 25 = -5%
Unexplained = 1-AB = 100 - 25 = 75%
A negative value to me seems than that the combination of the two models explains less than when applied separately. It is also argued that they can be set to zero, although not really satisfying. Yet, quoted from GUSTA ME, "These are artifacts of the VP procedure which arise when certain relationships are present in the data. No solution is available to remove them meaningfully. " (https://mb3is.megx.net/gustame/constrained-analyses/variation-partitioning).
- Badges
- Science topic