An embedded and charset-unspecified text was scrubbed... Name: not available URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20111004/5a922777/attachment.pl>
Adonis and nmds help and questions for a novice.
4 messages · Ashley Houlden, Gavin Simpson, Sarah Goslee
On Tue, 2011-10-04 at 08:45 +0000, Ashley Houlden wrote:
Hi,
forgive me if someone has already posted about this but I have had a
look and cannot find the answer, also I am very new to R and been
getting the grips with this.
I have been trying to use Adonis to find out if there are significant
difference between groups on data that I have analyses with NMDS, and
have been struggling with getting this to work and understanding what
is going on. I am looking at diversity in different soils with either
woodland or grassland habitats.
I have run the scripts
library(vegan)
library(ecodist)
library(MASS)
mydata <- read.table("ash_data.csv", header=TRUE, sep=",",
row.names="Site")
envdata_fit <- read.table("ash_env.csv", header=TRUE, sep=",",
row.names="Site")
#distance matrix of samples using bray curtis
d= bcdist(mydata, rmzero=FALSE)
And then using the distance matrix from this to use for adoins? Is
this correct.
With this I have then run Adonis
results = adonis(d ~ wood, envdata_fit, permutations = 1000)
and get significant values to see if sig diff in diversity between
wood and grass habitat.
However I have been reading about combining the variables, but there
seems to be different ways for example
results = adonis(d ~ wood+soil, envdata_fit, permutations = 1000)
so get sig values for Wood and soil
or
results = adonis(d ~ wood*soil, envdata_fit, permutations = 1000)
And I get sig values for wood, soil, and wood soil interaction.
This seems to make sense, however for both if I put the variable the
other way around (soil+wood or soil*wood) I get very different sig
values, even accounting for the fact they vary slightly due to the
permutations. So whats is going on and why to the the values change so
much?
You can isolate the effects due to different permutations being used by setting a seed via set.seed(). As ?adonis says, sequential sums of squares are used. If there is imbalance in your design it isn't surprising that the results are not invariant to the ordering of terms in the formula.
I was also wondering in Adonis, can you nest treatments, so see effect of soil removing the effect of woodland as you can with anova?
Not 100% sure what you mean by nested, but adonis() uses the full functionality of R's formula interface. See The R manual for details or ?formula. ?adonis also has details of how you might test a nested design in the Details section - this might not be what you want but it does allow you to test for an effect of one variable by conditioning the permutations on another.
Another general questions as well, if I have more than two groups in a treatment, say for soil, clay, sand, loam and do the stats, and I get a significant value, what does it actually mean, is it that soil generally has an effect, with each group separate, or there are general differences between soils which may be one group is very different to the other two?
The permutation test, test at the level of the factor, not pairwise comparisons of the levels within the factor. So you get information on Soil, not on Clay, Sand, Loam levels. This is the same as you would get if you did anova(mod) where mod was a linear model with a factor predictor. betadisper() the sister function to adonis() which tests for differences of multivariate dispersions, not differences of multivariate means, does allow the sorts of pairwise tests you are thinking of, but we haven't implemented this in adonis yet I'm afraid. HTH G
Many many thanks to anyone who can help me as I have asked people who use R near me and no-one is sure and uses Adonis.. Ash
______________________________________________ R-help at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~% Dr. Gavin Simpson [t] +44 (0)20 7679 0522 ECRC, UCL Geography, [f] +44 (0)20 7679 0565 Pearson Building, [e] gavin.simpsonATNOSPAMucl.ac.uk Gower Street, London [w] http://www.ucl.ac.uk/~ucfagls/ UK. WC1E 6BT. [w] http://www.freshwaters.org.uk %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
On Tue, 2011-10-04 at 08:45 +0000, Ashley Houlden wrote:
Hi,
<snip />
#distance matrix of samples using bray curtis d= bcdist(mydata, rmzero=FALSE)
In addition, you don't necessarily need ecodist for the bray curtis distance. vegdist() in vegan will compute this for you. Not that there is anything wrong with ecodist I hasten to add - just that you can do this all in vegan if you wanted. G
And then using the distance matrix from this to use for adoins? Is this correct. With this I have then run Adonis results = adonis(d ~ wood, envdata_fit, permutations = 1000) and get significant values to see if sig diff in diversity between wood and grass habitat. However I have been reading about combining the variables, but there seems to be different ways for example results = adonis(d ~ wood+soil, envdata_fit, permutations = 1000) so get sig values for Wood and soil or results = adonis(d ~ wood*soil, envdata_fit, permutations = 1000) And I get sig values for wood, soil, and wood soil interaction. This seems to make sense, however for both if I put the variable the other way around (soil+wood or soil*wood) I get very different sig values, even accounting for the fact they vary slightly due to the permutations. So whats is going on and why to the the values change so much? I was also wondering in Adonis, can you nest treatments, so see effect of soil removing the effect of woodland as you can with anova? Another general questions as well, if I have more than two groups in a treatment, say for soil, clay, sand, loam and do the stats, and I get a significant value, what does it actually mean, is it that soil generally has an effect, with each group separate, or there are general differences between soils which may be one group is very different to the other two? Many many thanks to anyone who can help me as I have asked people who use R near me and no-one is sure and uses Adonis.. Ash
________________________________ No virus found in this message. Checked by AVG - www.avg.com<http://www.avg.com> [[alternative HTML version deleted]] ______________________________________________ R-help at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~% Dr. Gavin Simpson [t] +44 (0)20 7679 0522 ECRC, UCL Geography, [f] +44 (0)20 7679 0565 Pearson Building, [e] gavin.simpsonATNOSPAMucl.ac.uk Gower Street, London [w] http://www.ucl.ac.uk/~ucfagls/ UK. WC1E 6BT. [w] http://www.freshwaters.org.uk %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
On Tue, Oct 4, 2011 at 4:41 PM, Gavin Simpson <gavin.simpson at ucl.ac.uk> wrote:
On Tue, 2011-10-04 at 08:45 +0000, Ashley Houlden wrote:
Hi,
<snip />
#distance matrix of samples using bray curtis d= bcdist(mydata, rmzero=FALSE)
In addition, you don't necessarily need ecodist for the bray curtis distance. vegdist() in vegan will compute this for you. Not that there is anything wrong with ecodist I hasten to add - just that you can do this all in vegan if you wanted.
Because ecodist is awesome. :) But there's no need to mix and match; vegan and ecodist do many of the same things (for historical reasons). Sarah
Sarah Goslee http://www.functionaldiversity.org