I think that the answers to your question (as many other in biological science) is depends. I think that rules of thumb like the one provided by Dennis Goss de Souza are useful, although dangerous. To interpret any statistic you have to first determine if is significant given your hypothesis. if it is you can try to make sense of if. In the ANOSIM particularly, the degree of similarity depends on how much overlap you can allow in your particular subject of study. It also depends on the vagility of your system. All these issues need to be explained and defined before you run the experiments and during the interpretation of the results. Also take into account that R in ANOSIM is (aprox.):

[(similarity between groups) - (similarity within groups)]/[n(n-1)/4]

is bounded between -1 and 1 and can be interpreted similar to a correlation value.

I will also urge you to read Anderson 2013, about drawbacks and rejection rates for ANOSIM in particular situations.

Cheers!

http://www.esajournals.org/doi/pdf/10.1890/12-2010.1

Hi Lidimara,

There's a good previous discussion on this topic - please see attached link. In short, the higher the R value, the more of a difference between samples (the maximum value is 1). See this discussion also for some advice on negative R values if you happen to get any.

https://www.researchgate.net/post/Can_anyone_help_me_in_understanding_and_clearly_interpreting_ANOSIM_Analysis_of_Similarityand_SIMPER_Similarity_percentage_analysisresults

0.75 < R < 1 - highly different

0.5 < R < 0.75 -  different

0.25 < R < 0.5 -  different with some overlap

0.1 < R < 0.25 - similar with some differences (or high overlap)

R < 0.1 - similar

I think that the answers to your question (as many other in biological science) is depends. I think that rules of thumb like the one provided by Dennis Goss de Souza are useful, although dangerous. To interpret any statistic you have to first determine if is significant given your hypothesis. if it is you can try to make sense of if. In the ANOSIM particularly, the degree of similarity depends on how much overlap you can allow in your particular subject of study. It also depends on the vagility of your system. All these issues need to be explained and defined before you run the experiments and during the interpretation of the results. Also take into account that R in ANOSIM is (aprox.):

[(similarity between groups) - (similarity within groups)]/[n(n-1)/4]

is bounded between -1 and 1 and can be interpreted similar to a correlation value.

I will also urge you to read Anderson 2013, about drawbacks and rejection rates for ANOSIM in particular situations.

Cheers!

http://www.esajournals.org/doi/pdf/10.1890/12-2010.1

I second Jose Sergio's reply - IT DEPENDS. I think it is worthwhile listing any R values that you WANT to discuss according to your hypothesis, and making statements along the lines of "just missing the CONVENTIONAL level of .. and the p value of 0.05, but still indicating some pattern". That's of course not the strictly mathematical/statistical way, but as USERS of statistics, I think we have to apply some pragmatism as well - all the more so when the data are far from what a statistician / mathematician would consider data-worth-analyzing-and-not-just-house numbers ..

I thank you for the answers. They helped a lot.

As Jose Sergio and Susanne said, it depends and you can’t give a clear cut. As Jesse mentioned, there is already a good discussion related to ANOSIM and R values on RG.

I would add that the strength of the R values will also depends on the number of replicates you have (which subsequently determined the number of permutations possible) and the “nature” of your data (i.e. the number of variables you have, and if you expect your data to show lot of variability).

If you find a R value = 0.4 comparing the effect of one treatment to a control, R=0.4 will not have the same meaning if you have 10 samples or 100 samples and if you have 10 variables or 1000 for each sample. So you need to take into consideration all these parameters specific to your study, your research question, and see if all these make sense.

Actually R is a scaled measure of difference between groups, so it is not really influenced by numbers of replicates, nor by the number of variables you have (as it is calculated from resemblances, not variables directly).  That being said, the TEST will be influenced by the number of samples, and inter-sample variability.  The test, however, is of the hypothesis that the difference between groups is zero.  The discussion here is about interpreting levels of R, which is not the same thing.  There is no hard and fast guidance here, so Dennis' framework is as good as any.