Adonis and Permdisp results

[Inga]

Dear QIIME Community

I have 30 faecal samples from slugs from 10 different locations (3 per site) and want to find out if grouping them by site is significant.
I used the methods adonis and permdisp, both of which were highly significant (adonis: R2 = 0.44, p = 0.001, permdisp: F = 4.61, p = 0.004). If I understand correctly, that means that the significant result I observed with adonis is due to unequal variances between the samples? In my PCoA plots there are no clear clusters visible, which made me suspicious in the first place. I then looked at the results of the pairwise permdisp, filtered out the samples which had significantly different variances and ran both tests on the filtered distance matrix again. The new results are: R2 = 0.47, p = 0.001 and F = 5.14, p = 0.002. However, looking at the pairwise comparisons,  none of them give a p-value below 0.05.
I am confused! Did I misunderstand something fundamental here?
Thanks!
Inga

[Jai Ram Rideout]

Hi Inga,

That's my understanding too: adonis can be sensitive to differences in dispersion, so using something like permdisp can help determine whether the differences you're seeing are due to dispersion (spread) or center.

I'm confused by your results as well. I wonder if the pairwise permdisp p-values are being corrected for multiple comparisons? This could make the pairwise p-values greater than 0.05. Another thing to consider is the permutational scheme being used. The QIIME script uses a simple permutation scheme that assumes independence between your samples; this may or may not be the correct permutation scheme to use for your adonis/permdisp tests (it is not currently possible to specify a different permutation scheme using QIIME however, so you'd need to run this in R).

I recommend checking out GUSTA ME, which has info about applying these sorts of methods to microbial ecology datasets. For example, there are pages for adonis and different permutation schemes. GUSTA ME has a user forum where you might post your question. I am not a statistician and have only a basic understanding of these methods, so you may be better helped by one of the GUSTA ME statisticians there.

Good luck and sorry to not be of more help!

Best,

Jai

--

---
You received this message because you are subscribed to the Google Groups "Qiime Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qiime-forum...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Inga]

Thanks Jai, I'll see what I can learn :).

Best wishes,
Inga