[R-sig-eco] Measuring community similarity when total abundance differs

[Dixon, Philip M [STAT]]

Tim,

I don't know of any automatic way to deal with unequal sampling effort.  Even if there were, I would argue strongly that you need to think carefully about the data and how they relate to how you want to define 'similar' and 'less similar' community composition.  

Fundamentally, this is an issue about how to define the distance between two vectors of species abundances.  Bray-Curtis is sensitive to differences in total abundance.  Perhaps the difference between a sample with 100 individuals and a sample with 12000 individuals is informative.  Probably it isn't.  If so, you probably (but not certainly) want to compute percent similarity (Bray-Curtis after dividing each abundance by total abundance in the sample).  That's a very common recommendation, but even that may be misleading if the reason for the 12000 is an emergence flush of one species.  Adjusting by total sqrt(abundance) may be appropriate, but it is certainly not common.  Similar levels of abundance heterogeneity occur in benthic macrofauna data sets.  Clarke is fond of Bray-Curtis (without total abundance correction, if I remember correctly) of sqrt(sqrt(abundance)), i.e. abundance^(1/4).  

Anne Chou and Robin Chazdon have developed and used abundance corrected similarity measures, similar in spirit to the Chou measures of richness.  I'm not fond of them because I believe the adjustments for sampling effort presume that two samples come from the same underlying community.  

Any distance measure reduces a pair of vectors to a single number.  It always throws away information.  You need to make sure that your choice of measure keeps the features you feel are important.  I strongly suggest you look at the abundance vectors for pairs of samples with small distances and pairs of samples with large distances to make sure that your quantitative measure matches your intuitive sense of 'similar' and 'not similar'.

Philip Dixon