Hello, I have been told by a colleague that interpreting the ANOSIM R value as a measure of importance is not good practice as stats does not necessarily tell you anything about biology. Would I be more correct in interpreting the R value as a measure of predictive power of the model? Is anyone able to clarify? Thank you in advance.
Dear Patrick,
High R shows that the groups are very dissimilar and low R the opposite. This is based on ranking the distance between items, and looking at the ranks of distances between items within groups vs distance between groups.
The reason why your colleague (and many others) would say that direct interpretation of the R is unhelpful is that the asymptotic values change with the number of observations and the number of groups (not to mention their relationships and questions about how the distances were derived). So we cannot take an R value, say .5, as being particularly large or small, or even impressive compared to the R from some other study.
If you really wanted to interpret such a value you would probably have to do some simulation to work out the possible distribution and likely range for your setup. However, doing this requires assumptions that more or less defeat the reasons for ANOSIM in the first place.
Comparing the R with that of other models on the same data can be down however.
As to the broader point that "stats does not necessarily tell you anything about biology"... statistics is not substantive in that sense. A statistic can, however, be incredibly informative when biology is understood well enough to put theories to a test!
Hope this helps.
Dear Patrick,
High R shows that the groups are very dissimilar and low R the opposite. This is based on ranking the distance between items, and looking at the ranks of distances between items within groups vs distance between groups.
The reason why your colleague (and many others) would say that direct interpretation of the R is unhelpful is that the asymptotic values change with the number of observations and the number of groups (not to mention their relationships and questions about how the distances were derived). So we cannot take an R value, say .5, as being particularly large or small, or even impressive compared to the R from some other study.
If you really wanted to interpret such a value you would probably have to do some simulation to work out the possible distribution and likely range for your setup. However, doing this requires assumptions that more or less defeat the reasons for ANOSIM in the first place.
Comparing the R with that of other models on the same data can be down however.
As to the broader point that "stats does not necessarily tell you anything about biology"... statistics is not substantive in that sense. A statistic can, however, be incredibly informative when biology is understood well enough to put theories to a test!
Hope this helps.
There is few discussions on ANOSIM in RG that you may find interesting to read:
https://www.researchgate.net/post/Can_anyone_help_me_in_understanding_and_clearly_interpreting_ANOSIM_Analysis_of_Similarityand_SIMPER_Similarity_percentage_analysisresults/1
https://www.researchgate.net/post/Why_do_I_obtain_different_results_for_MDS_and_PERMANOVA_in_PRIMER/1
https://www.researchgate.net/post/Which_R_value_is_considered_to_show_a_strong_difference_between_the_groups_in_ANOSIM
https://www.researchgate.net/post/Any_advice_on_Primer_7_in_running_an_ANOSIM_Two_way_Nested_Crossed_R-Value_Interpretation
Thanks for this Bruce and Aimeric, this has been very helpful, especially the other discussions you have pointed out Aimeric.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques