Dear Vindoggy, I wondered how do you *know* that species three is influential. Your analysis shows that it is linearly well explained by the ordination, but it does *not* show that the species influences the ordination.
We can study the influence of a species by seeing how much the result will change if we remove the species: removal of an influential species will change the result much. We can assess the change in ordination using the residual sum of squares of Procrustes ordination where we compare results of full data to results when we remove one species. The following continues from your analysis where you had the ordination result 'mds': impo <- numeric(4) for(i in 1:4) impo[i] <- procrustes(mds, metaMDS(DF[,-i]))$ss This gives us importance values: impo # [1] 0.13510647 0.05256924 0.00666738 0.03128587 Third value is lowest. This means that removing of species 3 hardly influences the results, and the species is not influential. By far the most important species is number 1. Cheers, Jari Oksanen On 19/02/2014, at 06:10 AM, Vindoggy ! wrote: > > > I have a data set for benthic cover made up of 4 species, over 9 years. I > have been using the vegan package to create nMDS plots for this data, but am > a little confused by some of my results. > > > Here is the data for 4 species (Sp 1-4) over the course of 9 years > > > Sp1<-c(27.76 ,25.44, 34.72, 41.28, 44.00, 31.28, 32.00, 38.72, 40.40) > Sp2<-c(36.72, 40.08, 36.96, 26.40, 34.40, 33.28, 31.28, 22.72, 14.00) > Sp3<-c(0.16, 0.08, 0.32, 1.28, 1.76, 1.84, 2.24, 4.40, 1.76) > Sp4<-c( 1.12, 2.24, 4.00 , 2.56, 4.40, 8.96, 12.24, 13.60, 23.36) > > DF<-data.frame(Sp1,Sp2,Sp3,Sp4) > > row.names(DF)<-c(2006:2014) > > > > At first I was advised to transform my data to relative abundance like so: > > > > DF<- decostand(DF, method='total') > > > > set.seed(999) > mds<- metaMDS(DF, dist='bray',trymax=99) > > envfitHab<-envfit(mds, DF,perm=999) > > plot(mds,display='sites',type='t') > > plot(envfitHab) > > envfitHab > > > > > > > > > > But I found that Species 3 seemed to be having a greater impact on the plot > (as I judged by looking at the blue vectors, p values, and scores data) than > I would have gathered. Considering Sp3 has a relatively low abundance and > does not change much over time, I would not expect it to influence the > dissimilarity matrix much at all. And Sp1, which has a higher abundance and > changes significantly over time, seems to have a much smaller impact and > larger p value. > > I came to the conclusion that I should not be using relative abundance data, > as I do care about total abundance. So I gave up on decostand and used raw > abundance data instead. I was surprised to find that Sp3 played a similar if > even larger role in impacting the plot. Which runs counterintuitive to my > (admittedly limited) understanding of nMDS. > > So my question is why is a species with low abundance that does not vary much > from year to year impacting my nMDS plot so much? > > > > > [[alternative HTML version deleted]] > > _______________________________________________ > R-sig-ecology mailing list > [email protected] > https://stat.ethz.ch/mailman/listinfo/r-sig-ecology _______________________________________________ R-sig-ecology mailing list [email protected] https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
