I know that many biostatisticians prefer using MEM rather than PCNM in spatial ecology and biogeography, although the later belongs to the same MEM family. Why?
I am quite familiar using PCNM but I have realized that the PCNM package is no more supported by the latest release of R program (version 3.0.0). Is there any other package?
If so, is this new spatial analysis method based on Euclidean distance measure?
In addition, is this new package still can support the forward selection to select the key spatial variables?
Hi Salman. I agree with Francois. MEM is a refinement of PCNM but still suffers from potential Type 1 error and interpretation problems. I've attached the Gilbert and Bennett paper. Generalized dissimilarity modelling (check out papers by Simon Ferrier) might be an alternative. Or you could use both techniques and see if your results agree. If the patterns are strong, then the results should be fairly consistent. Good luck with your analyses!
Thanks Francois and Joseph for the feedback and discussion.
Interestingly, I have found that both PCNM and MEM still not completely free of risk as both methods may overestimate R2. Unfortunately, no other method available until now.
Do you think is the forward selection will reduce the Type 1 error?
Based on the spatial ecology R-froge, PCNM, spacemakeR, packfor packges still being developed. The old versions of these packages are not supported by R version 3.0.0.
Hence, I may wait for while until these packages being released.
Thanks again for your valuable feedback.
You're welcome! Good luck with your analyses. And yes, I think forward selection will reduce Type-1 error, but it won't eliminate it.
All the best,
Joe
hi! does anyone have a good tutorial (step by step) for PCNM and variation partitioning in R?it would be great help. thanks!
You can find the reference and R code in Dr. Pierre Legendre' s webpage
http://adn.biol.umontreal.ca/~numericalecology/indexEn.html
Hi,
Regarding "Interestingly, I have found that both PCNM and MEM still not completely free of risk as both methods may overestimate R2. Unfortunately, no other method available until now.Do you think is the forward selection will reduce the Type 1 error?
"
This is probably very late in the discussion so might be too late to help you out. However, you could first use variography to look for lag distance at which the sill is reached (distance at which there is no more spatial autocorrelation) to generate a priori hypotheses about which principal co-ordinates to include in your subsequent analysis.
-Steve
All the packages from the old r-forge page are now available in the github . You can download the entire repository for example and compile the packages from source code. I just did it and still works in R 3.3.0.
https://github.com/rforge/sedar
Hello everyone,
I was looking for an answer to a question I have and I reached to your conversation, so I thought you might be able to help me with it.
I need to analyse the role of spatial vs environmental effect (through variation partitioning) on Notonecta species distribution among fishless ponds. I have been using adespatial package to do that. My problem is after I calculate the MEMs how can I evaluate the scale of each one of them (fine versus large) to form the submodels to be used in variation partitioning? I have read and tried most of the adespatial function but I could not find the function which can help me calculate the scale of each MEM.
I really appreciate if anyone can help.
Mitra
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology