repeated measures NMDS?

hi eduard,

i faced similar problems recently and came to the below solution.
i only try to address the pseudoreplication with an appropiate  
permutation scheme.
when it comes to testing the interactions, things may get more complicated.

the code is in no way approven of, but at least it maybe good enough  
for a starter.

best,
kay

From what Eduard and you describe, the code you show is not the correct

library(vegan)

### species matrix with 5 sp.
### one env.variable
### a factor denoting blocks of repeated measurments

sp<-matrix(runif(24*3*5,0,100),24*3,5)
env<-rnorm(24*3,10,2)
rep.mes<-gl(24,3)

### NMDS:
sol<-metaMDS(sp,trymax=5)

fit<-envfit(sol~env)
plot(sol)
plot(fit)

### testing code for appropiate randomization,
### permuting blocks of 3 as a whole:
permuted.index2(nrow(sp),permControl(strata = rep.mes,permute.strata=T))

B=4999

### setting up frame for population of r2 values:
pop<-rep(NA,B+1)
pop[1]<-fit$vectors$r

### loop:
for(i in 2:(B+1)){
fit.rand<-envfit(sol~env[permuted.index2(nrow(sp),
                          permControl(strata = rep.mes,permute.strata=T))])
           pop[i]<-fit.rand$vectors$r
}

### p-value:
print(pval<-sum(pop>pop[1])/B+1)

### compare to anti-conservative p-value from envfit(),
### not restricting permutations:

envfit(sol~env,perm=B)


Zitat von Eduard Sz?cs <szoe8822 at uni-landau.de>:

Thanks, that helped.

permuted.index2() generates these types of permutations. But  
envfit() does not use this yet.
What if I modify vectorfit() (used by envfit() ) in such a way that  
it uses permuted.index2() instead of permuted.index()?


Eduard Sz?cs





Am 08.11.2010 22:01, schrieb Gavin Simpson:

On Mon, 2010-11-08 at 15:39 +0100, Eduard Sz?cs wrote:

Hi listers,

I have species and environmental data for 24 sites that were sampled
thrice. If I want to analyze the data with NMDS I could run metaMDS on
the whole dataset (24 sites x 3 times = 72) and then fit environmental
data, but this would be some kind of pseudoreplication given that the
samplings are not independent and the gradients may be overestimated,
wouldn`t it?

For environmental data a factor could be included for the sampling
dates - but this would not be possible for species data.

Is there an elegant way either to aggregate data before ordination or
to conduct sth. like a repeated measures NMDS?

Thank you in advance,
Eduard Sz?cs

Depends on how you want to fit the env data - the pseudo-replication
isn't relevant o the nMDS. If you are doing it via function `envfit()`,
then look at argument `'strata'` which should, in your case, be set to a
factor with 24 levels. This won't be perfect because your data are a
timeseries and, strictly, one should permute them whilst maintaining
their ordering in time, but as yet we don't have these types of
permutations hooked into vegan.

If you are doing the fitting some other way you'll need to include
"site" as a fixed effect factor to account for the within site
correlation.

You don't need to worry about the species data and accounting for
sampling interval. You aren't testing the nMDS "axes" or anything like
that, and all the species info has been reduced to dissimilarities and
thence to a set of nMDS coordinates. You need to account for the pseudo
rep at the environmental modelling level, not the species level.

HTH

G