posthoc for anosim

[Vinn]

Hi,

I am comparing fungal endophytic communities from 7 plant species. I ran anosim and permanova to test if the communities are different, and the tests suggested that the communities are different (p = 0.019 for anosim and 0.021 for permanova). However, it turned out no significant difference for all pairs when I ran make_distance_boxplots as a post-doc. I am thinking about other tests but my statistics skill is so limited. Can you explain and perhaps suggest which test I should run to solve the mystery?

Thanks in advance!
 

[TonyWalters]

Hello Vinn,

Apologies for the delay in answering. I talking to another person here, and there are a couple of things here.
make_distance_boxplots.py compares the distributions of within and
between distances for a variety of different groupings of samples, and
IIRC, it uses a non parametric test to assess differences between these
distributions.

ANOSIM, uses a different approach, it computes the R statistic, that
takes the differences of the "between group mean distance" and the
"within group mean distance", and then divides this by n(n - 1), where n
is the number of samples being compared. Then permutes the labels and
asks if there is a "random" permutation that makes up for a better R
statistic.

I would also note that there are not multiple comparison corrections for permanova and ANOSIM, so you'd want to consider that when looking at your p-values (and it might be worth increasing the number of permutations being used on those calculations).

-Tony

And thanks Yoshiiki.

[Vinn]

Hi Tony and Yoshiiki,

Thanks for your reply. I always use nonparametric with n = 999, including with the above results. Both statistics and R are not my strength. I am dealing with fungal ITS so I guess beta_significance.py is not an option. I very much appreciate for any advice you might have.

-Vinn