simper {vegan}

Similarity Percentages

Description

Discriminating species between two groups using Bray-Curtis dissimilarities

Usage

simper(comm, group, permutations = 999, parallel = 1, ...)
## S3 method for class 'simper'
summary(object, ordered = TRUE,
    digits = max(3,getOption("digits") - 3), ...)

Arguments

comm

Community data.

group

Factor describing the group structure. If this is missing or has only one level, contributions are estimated for non-grouped data and dissimilarities only show the overall heterogeneity in species abundances.

permutations

a list of control values for the permutations as returned by the function how, or the number of permutations required, or a permutation matrix where each row gives the permuted indices.

object

an object returned by simper.

ordered

Logical; Should the species be ordered by their average contribution?

digits

Number of digits in output.

parallel

Number of parallel processes or a predefined socket cluster. With parallel = 1 uses ordinary, non-parallel processing. (Not yet implemented).

...

Parameters passed to other functions. In simper the extra parameters are passed to shuffleSet if permutations are used.

Details

Similarity percentage, simper (Clarke 1993) is based on the decomposition of Bray-Curtis dissimilarity index (see vegdist, designdist). The contribution of individual species i to the overall Bray-Curtis dissimilarity d_{jk} is given by

d_{ijk} = \frac{|x_{ij}-x_{ik}|}{\sum_{i=1}^S (x_{ij}+x_{ik})}

where x is the abundance of species i in sampling units j and k. The overall index is the sum of the individual contributions over all S species d_{jk}=\sum_{i=1}^S d_{ijk}.

The simper functions performs pairwise comparisons of groups of sampling units and finds the contribution of each species to the average between-group Bray-Curtis dissimilarity. Although the method is called “Similarity Percentages”, it really studied dissimilarities instead of similarities (Clarke 1993).

The function displays most important species for each pair of groups. These species contribute at least to 70 % of the differences between groups. The function returns much more extensive results (including all species) which can be accessed directly from the result object (see section Value). Function summary transforms the result to a list of data frames. With argument ordered = TRUE the data frames also include the cumulative contributions and are ordered by species contribution.

The results of simper can be very difficult to interpret and they are often misunderstood even in publications. The method gives the contribution of each species to overall dissimilarities, but these are caused by variation in species abundances, and only partly by differences among groups. Even if you make groups that are copies of each other, the method will single out species with high contribution, but these are not contributions to non-existing between-group differences but to random noise variation in species abundances. The most abundant species usually have highest variances, and they have high contributions even when they do not differ among groups. Permutation tests study the differences among groups, and they can be used to find out the species for which the differences among groups is an important component of their contribution to dissimilarities. Analysis without group argument will find species contributions to the average overall dissimilarity among sampling units. These non-grouped contributions can be compared to grouped contributions to see how much added value the grouping has for each species.

Value

A list of class "simper" with following items:

species

The species names.

average

Species contribution to average between-group dissimilarity.

overall

The average between-group dissimilarity. This is the sum of the item average.

sd

Standard deviation of contribution.

ratio

Average to sd ratio.

ava, avb

Average abundances per group.

ord

An index vector to order vectors by their contribution or order cusum back to the original data order.

cusum

Ordered cumulative contribution. These are based on item average, but they sum up to total 1.

p

Permutation p-value. Probability of getting a larger or equal average contribution in random permutation of the group factor. These area only available if permutations were used (default: not calculated).

Author(s)

Eduard Szöcs and Jari Oksanen.

References

Clarke, K.R. 1993. Non-parametric multivariate analyses of changes in community structure. Australian Journal of Ecology, 18, 117–143.

See Also

Function meandist shows the average between-group dissimilarities (as well as the within-group dissimilarities).

Examples

data(dune)
data(dune.env)
(sim <- with(dune.env, simper(dune, Management, permutations = 99)))
## IGNORE_RDIFF_BEGIN
summary(sim)
## IGNORE_RDIFF_END