Adam Kustka
2014-Jun-01 16:39 UTC
[R] adonis output not significant even with dummy input?
I used adonis to test tested for differences in the relative species (OTU)
abundances from incubation experiments with a common starting water source
and duplicates for each of three treatments (a one way test with three
levels in the independent variable). For full disclosure (but it may not
be relevant) I came up with a metric to reflect change in rel species
abundance from t0 to t9 days, weighted to not overly inflate significance
of rare species. The Bray Curtis similarity table looked compelling (where
A and B represent replicates). I was surprised to see p ~0.06. So, I made
a dummy file with a blatant differences (dummy file below) and only
obtained a p value of 0.05. I’m clearly missing something here. BTW I also
noticed if you purposefully mismatch your env file you can still get a
“result”. I'm so grateful for your help!!!
B-C output from real data (three treatments each replicated twice)
St24FeA
St24FeB
St24NoFeA
St24NoFeB
St24EntA
St24FeB
0.081774
St24NoFeA
0.322509
0.282081
St24NoFeB
0.343357
0.301258
0.064281
St24EntA
0.268338
0.221075
0.098628
0.161846
St24EntB
0.203852
0.194518
0.166727
0.221682
0.096111
Dummy file used as
OTU1
OTU2
OTU3
OTU5
St24FeA
100
20
100
20
St24FeB
100
25
100
25
St24NoFeA
5
80
5
80
St24NoFeB
5
80
5
80
St24EntA
7
100
7
100
St24EntB
7
100
7
100
BC of dummy file (BCdum) is even more compelling than the real data (no
transforms done for this).
Output from adonis using BCdum as input.
Call:
adonis(formula = Bcdum ~ treatment data=st24.env, permutation=2000)
Terms
added
sequentially
(first
to
last)
Df
SumsOfSqs
MeanSqs
F.Model
R2
Pr(>F)
treatment
2
0.73764
0.36882
5313.2
0.99972
0.05147
Residuals
3
0.00021
0.00007
0.00028
Total
5
0.73785
1
The entire script is here. Note the “abundance” data is not transformed in
R. Rather, I calculate a metric of change in percent abundance
t= 9 d(counts(i)/counts(total) /
t = 0 d(counts(i)/counts(total)
which is then normalized so small abs changes in rare species don’t have an
undue influence.
St24Y<-read.table("st24relchangeforR.txt")
BC24<-vegdist(St24Y,method=”bray”,binary=FALSE, diag=FALSE,
upper=FALSE,na.rm = FALSE)
st24.env<-read.table("st24envfileforR.txt")
adonis(BC24~ treatment,st24.env,perm=2000)
[[alternative HTML version deleted]]