0

I am new to statistical analysis and R. I have data of relative abundance (in percentage) of 5 taxa in 8 ponds. The number of samples in the 8 ponds ranges from 7 to 12 (making the design unbalanced). Can I use PERMANOVA (vegan::adonis2) and PERMDISP to test if the composition of the 5 taxa is different in the 8 ponds?

Thank you!

I have tried as follows:

matrix<-as.matrix(Data[,c(5:9)]) # columns 5:9 refers to relative abundance of the 5 taxa in percentage

dist<-vegdist(matrix, method="robust.aitchison", na.rm = F)

adonis2(dist~ Pond, data=Data)

anova(betadisper(dist,Data$Pond))

enter image description here

Does the result mean that there are no difference between the ponds as a whole? And how can I analyse the differences of individual taxon among the ponds?

Thank you!

Vons
  • 3,277
  • 2
  • 16
  • 19
Aston
  • 3
  • 2
  • FYI; you don't need to take a screen shot of your output from the R console. It's just text, so copy it to the clipboard and then paste it into your question and wrap it in backticks to indicate it is code – Gavin Simpson Jun 02 '23 at 17:23

1 Answers1

0

Yeah, that's fine. But you don't want the anova() method for betadisper() objects; you should use permutest() on the betadisper() object to get the permutation test of heterogeneous dispersions.

The results of the PERMANOVA mean you fail to reject the null hypothesis that the variation explained by Pond is consistent with the Null hypothesis of no effect. The results do not indicate there is no difference between ponds; there are differences (otherwise the SumOfSqs would be 0), but they are just small.

Analysing the per taxon differences would just be a univariate model so not really something you would do with adonis2()...

Gavin Simpson
  • 170,508
  • 25
  • 396
  • 453
  • Thank you!! Would you please also recommend any ways of doing the univariate tests (with such relative abundance data and unbalanced data)? – Aston Jun 05 '23 at 04:09
  • And do you know if 'pairwise.adonis2' can be used with such unbalanced data? Thanks a lot! – Aston Jun 05 '23 at 05:17
  • As far as I understand what `pairwise.adonis2()` does, it should be fine for unbalanced data (though your power to detect a true effect will obviously change). Univariate tests would be via generalized linear models on the abundances for example. – Gavin Simpson Jun 06 '23 at 07:44