[R-sig-eco] Is adonis the right test for me?

[Eduard Szöcs]

Hai Kari,

as far as i know SimPer is not available in vegan.

I have written some code for myself a few months ago to do this.
I think its doing the job, but coding is poor (3 for-loops...).


- Eduard


############################################################
######         SIMPER      , Clarke (1993)                           
            ########
###### V 0.3   29.04.2011  Szöcs, Eduard                                 
########
############################################################


###############
#### Code #####
###############

simper <- function(comm, groups, select){
   group.a <- as.matrix(comm[groups == select[1], ])
   group.b <- as.matrix(comm[groups == select[2], ])
   n.a <- nrow(group.a)
   n.b <- nrow(group.b)
   P <- ncol(comm)
   me <- matrix(ncol = P)
   md <- matrix(ncol = P)
   contr <- matrix(ncol = P, nrow = n.a * n.b)
   for(j in 1:n.b) {
       for(k in 1:n.a) {
           for(s in 1:P) {
               md[s] <- abs(group.a[k, s] - group.b[j, s])
               me[s] <- group.a[k, s] + group.b[j, s]
               a <- rowSums(me)
               c <- md / a
               contr[(j-1)*n.a+k, ] <- md / a
   }}}
   av.contr <- apply(contr, 2, mean)*100
   ov.av.dis <- sum(av.contr)
   sdi <- apply(contr, 2, sd)
   sdi.av <- sdi / av.contr
   av.a <- colMeans(group.a)
   av.b <- colMeans(group.b)
   dat <- data.frame(av.contr, sdi, sdi.av, av.a, av.b)
   dat <- dat[order(dat$av.contr, decreasing = TRUE),]
   cum <-  cumsum(dat$av.contr / ov.av.dis)*100
   out <- data.frame(dat, cum)
   out
}


###############
#### Usage ####
###############

#  simper(comm, groups, select)

# Arguments :
############

#   * comm : commumity data. species in columns, observations in rows
#   * groups : vector containing treatment per observation
#   * select : vector with length 2, indicating the two groups to compare


# Output :
##########

#   * av.contr : average contribution to overall similarity
#   * sdi : standard deviation of contribution
#   * sdi.va : ratio mean / sd = sdi / av.contr.
#   * av.a : average abundance in first group
#   * av.b : average abundance in second group
#   * cum : cumulative contribution


#################
#### Example ####
#################

require(vegan)
data(dune)
data(dune.env)
adonis(dune ~ Management, data=dune.env, permutations=99)
levels(dune.env$Management)
simper(dune, dune.env$Management, c("BF", "SF"))




Am 29.04.2011 06:34, schrieb Kari Lintulaakso:
> Dear list,
>
> I'm not so familiar with ecological statistics (though I should be),
> so I'm looking for a confirmation and support to my decision to move
> on with the adonis path.
>
> I've been trying to find a method to analyse differences between
> mammalian communities and I think adonis would do the work for me.
>
> My data consists of different localities (each of assigned into a
> vegetation class) and number of species in different ecomorphological
> classes (groups) I have generated.
> There are nine vegetation classes, 52 localities, six mammalian groups
> having some 200-300 species in them.
> My idea is to use these mammalian groups as "species" in the data
> matrix which would look like this:
>
> L	V	G1	G2	G3	...	G6
> 1	1	0	1	2	...	8
> 2	2	1	2	1	...	3
> 3	2	1	3	2	...	4
> 4	3	5	2	2	...	9
> 5	3	5	3	3	...	8
> 6	3	4	2	3	...	7
> ...		
> 52	9	8	8	1	...	0
>
> L=locality id
> V = vegetation class
> G1...G6 mammalian groups, and in the matrix the counts of species in the group.
>
> First I'm trying to test if there are any differences between the
> vegetation groups. For this I'm planning to use a pairwise adonis
> comparisons between the different vegetation groups, i.e V1 vs V2, V1
> vs V3, ... V8 vs V9. From each of these comparisons I'm taking the Df,
> F, P>F etc values into a matrix so that I'm able to see the values
> later.
>
> When I get the P-values I suppose I can say that there are differences
> between the vegetation classes when the P-value is<  0.05, right?
> And if so, I need to find a way to explain which mammalian groups are
> responsible for these differences? Does anyone have a good suggestion
> for this? I've found one approach which is called SimPer (Similarity
> Percentage Analysis), but I'm not sure is it in vegan.
>
> And finally, is there any requirement, how many different data rows I
> need for one type of vegetation? In some cases I only have one which
> would mean that the within group variation is zero?
>
> Any suggestions are welcome!
> -Kari
>
> Kari Lintulaakso, M.Sc.(Biosciences)
>
> Doctoral student
> Paleontology and Paleoecology
> Department of Geosciences and Geography
> University of Helsinki
>
> * Email: kari.lintulaakso at helsinki.fi
> * Post: Department of Geology, Gustaf Hällströmin katu 2a (P.O Box
> 64), 00014 University of Helsinki
> * Web page: http://blogs.helsinki.fi/lintulaa/
>
> _______________________________________________
> R-sig-ecology mailing list
> R-sig-ecology at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>