Hi. I am working on avifaunal assemblage of an estuary and I have 3 seasons data across 5 different habitat types. I used nmds ordination to look for any gradient in the assemblage across the habitats but in both 2-D and 3-D ordinations my stress values were high, almost equal to 0.3.
In such a case, reducing the number of species (using the most abundant species) would lead to a better ordination of the habitats or do I need to transform the abundances of all species before computing the dissimilarity matrix? I tried the second option using fourth root transformation, still the stress was higher than the non-transformed data. Or do I need to use a different ordination technique or a different dissimilarity matrix (I used Bray-Curtis)? If so please give some suggestions. (I use PAST and R for my analysis).
In my own data (plankton are highly seasonal), I presorted by season (a summer NMDS, winter NMDS, etc). Do you have enough seasons to do this?
Hi Edward. I have data for the three main seasons - summer, monsoon/rainy and winter but unfortunately not monthly.
Its better that you can compare the seasons, or you can compare among the habitats. Higher stress value may be due to dominance of some species.
Are you using season and habitats as grouping factors at the same time. As all answers suggest, u simply need to do saparate NMDS for different seasons.
Also u can try Horn-Morisita index, which is independent of sample size.
Thank you everyone for your replies. I am performing separate nmds analysis for habitats and seasons. In case of latter im yet to do the analysis. In case of habitats I'm using pooled data of all the three seasons. However my samples are unbalanced due to logistic issues. Few of them have just 3 sampling points whereas one habitat has 11 sampling points. Does an unbalanced design makes any difference for performing nMDS?
In case of number of species the total number of species is around 130 which I reduced to 40 most abundant ones for nMDS analysis. Interestingly the high stress values which I got was when I performed the test in PAST software but when I performed the same test in R (vegan) I got stress values less than 0.2 in 3-D ordination. If using different softwares give different results then which is the more reliable one?
@Avadhoot: thanks for suggestion but as I mentioned above my samples are unbalanced. I do not know whether Morisita's index can be performed in such a case.
Hi, You need first to pretransform your data if there is strong dominance if suit to your hypotheses. Ordination, like nMDS are not test for difference. Use ANOSIM or PERMANOVA to test if there is difference among your factors with an appropriate design. Whatever the results, you can interpret it. You can illustrate the results with an nMDS only if it make sens stress below 0.25, otherwise do not present it. In your results you can describe differences in the main text without figures. Remember that only the tests for differences among factors count, and do not use the ordination to describe where are they.
Hello,
I would follow the recommendation of Mathieu Cusson. It is not necessary eliminate species, or separate for seasons. Rather I would propose to start with a strong transformation to moderate the effect of the dominant species. I also think you should make before, a PERMANOVA test, to identify whether there differences and at what level these differences occur.
The values of your variables also show large differences in proportion? (Eg. Var. A = 25, 28, 31 ..., var. B = 0.5, 1, 0.7 ....) in this case is also necessary to apply a transformation to the ambiental matrix.
Good luck
@Paromita : I hope u mean different sampling efforts by unbalanced samples. In that case Horn-Morisita will work.
http://cc.oulu.fi/~jarioksa/softhelp/vegan/html/vegdist.html
as Mathieu Cusson suggested, you should stress on "appropriate design to test your factors". and nMDS is just a scatterplot, whether or not there is actual difference can only be confirmed by ANOSIM or PERMANOVA. I am more inclined towards ANOSIM as R statistic of ANOSIM also gives degree of saparation among groups.
But without saparating seasons i feel u might get negative R values, indicating higher between group similarities than within group. following article illustrates interpreting negative R values if u get.
http://www.int-res.com/articles/meps/180/m180p257.pdf
I hope this helps.
Kudos
Avadhoot
I suggest you filter the data set to exclude very rare taxa. Say the taxa must occur in 10% of samples and provide at least one percent of total abundance, also try dispersion index pre-processing
@Mathieu Cusson, I fully agree with your suggestion that ordinations without testing for different factors are meaningless. I am performing nMDS as the first step and want to follow it with either PERMANOVA or a related non parametric test. I have actually been able to get an interesting result regarding a particular habitat through nMDS but I'm getting high stress values in PAST software. I then did the same test using R and got better stress values but the results are almost same. But thanks for your reply.
@Avadhoot Velanker: thanks for the suggestions. I will also try Horn-Morisita's index as you suggested. In case of testing for differences between the habitats I too am thinking to first test overall differences between the groups followed by seasonal difference within each group,especially in one particular group which has shown high scatter in the ordination. Thanks.
the stress values issued by different softwares might not be directly comparable - not only because different algorithms are used, but also, because variants of the stress value formula exist. So the guidelines only apply in certain cases (see e.g. https://homepage.uni-tuebingen.de/florian.wickelmaier/pubs/Wickelmaier2003SQRU.pdf) and it is necessary to refer to the manuals to clarify how the stress is actually calculated in the specific case and with the given settings.
Hello. I have question about significance of nMDS and ANOSIM. Did ANOSIM follow Kruskal rules (MDS and nMDS) about model significance? Thanks
In the context that three years of samples with a large number of taxa, with replacement or lots of rare species, filter to remove the noise using various transformations & shade plot to find the opitmal best filter. With a situation with high stress try several tools: nMDS, bootstrap average MDS, Permanova- PCO ordinations. In the Permanova ( the factors that are making meaningful influence can be identified and quatified.