Hello, i'm using metaMDS to ordinate trait mean data of plants in different ecosystems. I used different calibrations but the stress differs only between 0,001 an 0,1e-19. I have never seen such a good stress, so I wonder about possible mistakes, but i cannot find one. Data looks like that: plots trait1 trait2 traits3 1 2,4 0,2 20 2 0,5 0,3 21 . . . . 300 . . . My standard calibration is: metaMDS(test ,distance = "euclidean" # because metric data , k = 3 # because I have 3 traits, but I couldn't plot a scree plot , trymax = 3-5 # not more, because too large dataset , autotransform =TRUE , noshare =1 # Ordination may be very difficult if a large proportion of sites have no shared species. In this case, the results may be improved with stepacross dissimilarities, or flexible shortest paths among all sites. No manipilation: noshare=1.0 ,trace = TRUE # will monitor the final stresses ,plot = TRUE ,zerodist="add" ) thanks for helping me out Daniel ----------------------------------------------------------------- Daniel Hornstein University of Bayreuth, Germany Department of Disturbance Ecology phone: 0921//55-2173
metaMDS: unrealistic good stress
2 messages · Daniel, Tyler Smith
Daniel <daniel_ho at web.de> writes:
i'm using metaMDS to ordinate trait mean data of plants in different ecosystems. I used different calibrations but the stress differs only between 0,001 an 0,1e-19. I have never seen such a good stress, so I wonder about possible mistakes, but i cannot find one.
You've got three traits, and you're using MDS to project them into three dimensions (k = 3). There's no stress, because you've got as many dimensions as variables. You only get stress when you try and squeeze the variation from three variables into less than three dimensions. Try again with k = 2 and you should get some stress. Put another way, why are you trying to project your data into three dimensions? If you want to plot it on paper, you need to project it into two dimensions. If you actually are using the three ordination axes in some subsequent analysis, you could just use your variables (possibly transformed) instead. HTH, Tyler
Data looks like that: plots trait1 trait2 traits3 1 2,4 0,2 20 2 0,5 0,3 21 . . . . 300 . . . My standard calibration is: metaMDS(test ,distance = "euclidean" # because metric data , k = 3 # because I have 3 traits, but I couldn't plot a scree plot , trymax = 3-5 # not more, because too large dataset , autotransform =TRUE , noshare =1 # Ordination may be very difficult if a large proportion of sites have no shared species. In this case, the results may be improved with stepacross dissimilarities, or flexible shortest paths among all sites. No manipilation: noshare=1.0 ,trace = TRUE # will monitor the final stresses ,plot = TRUE ,zerodist="add" ) thanks for helping me out Daniel ----------------------------------------------------------------- Daniel Hornstein University of Bayreuth, Germany Department of Disturbance Ecology phone: 0921//55-2173
There is something fascinating about science. One gets such wholesale
returns of conjecture out of such a trifling investment of fact.
--Mark Twain