Gretchen Brownstein and company have an accepted paper in the Journal titled “Chance in plant communities: a new approach to its measurement using the nugget from spatial autocorrelation “. Read the abstract here.
The authors have provided a short synopsis of the paper and a photo of one of their study sites on the South Island, New Zealand.
Photo credit: J. Bastow Wilson
Community composition, i.e. species presence or absence, is controlled by two types of factor: deterministic ones and random ones. There is a large body of work examining both types of factors, with evidence supporting both. Generally this work has looked for evidence of either random factors or deterministic factors, but it seems increasingly clear that both are at work simultaneously.
The discussion has now shifted towards examining whether random or deterministic factors are more important in structuring communities. Do different community or habitat types differ in their degree of randomness? Does species richness or time since disturbance influence the degree of randomness? These are fundamental ecological questions, however until now there has been no clear way to quantify the degree to which randomness or ‘chance variation’ determines community composition.
In our paper we describe a method for quantifying randomness using the distance/dissimilarity relation. We tested our method in 16 sites around the South Island of New Zealand, including forests, shrublands, grasslands and wetlands, to see if and how chance varied across plant community types and at different spatial grains (scales).
We found that randomness was not related to community type but rather that randomness was correlated with the species richness of the whole community: communities with more species have a greater degree of randomness. Our method provides ecologists with a tool to quantify randomness in communities and our analysis lends support to the idea that chance, redundancy and the size of the species pool are all connected.
Journal of Ecology Blog 
Hmm…the idea here is similar in spirit to variance partitioning methods that try to separate distance effects from environmental effects on species composition. Unfortunately, those methods don’t work as advertised–you can’t reliably infer the strength or importance of the underlying processes that generated the patterns from the patterns themselves: http://onlinelibrary.wiley.com/doi/10.1111/j.1600-0587.2009.06105.x/abstract The goal of understanding how deterministic and random processes together determine species composition is a great goal. But it’s not clear to me that one can achieve, or even help achieve, that goal with variance-partitioning type methods applied to purely observational data. There’s just no straightforward mapping from the underlying processes to their observed consequences.
Good point Jeremy. What do you propose as an alternative? I imagine manipulate experiments at large scales are prohibitive.
Not clear why the experiments have to be at “large scales”, if by “large scale expt.” you mean, say, “removing a species over a vast area.” For instance, if you think that “randomness” increases with species richness b/c population sizes decline as richness increases, well, why don’t you go out and manipulate absolute population sizes while holding densities and species richness constant? Do a small-scale expt. to directly test for the underlying processes that you think are at work, to complement your observations of the large-scale patterns. See, for instance, Jon Shurin’s 2000 work in Ecology on local-regional richness relationships. Jon complemented very large-scale documentation of that relationship with small-scale experiments directly testing a key mechanism (invasion resistance by species-rich local communities) thought to affect the shape of that relationship. This prevented him from misinterpreting the large-scale relationship.
In the context of this study, one thing you could manipulate is absolute population sizes. If you think that more species-rich sites exhibit more “randomness” because population sizes are smaller and so stochastic drift is relatively more important, well, manipulate absolute population size while holding densities and species richness constant (I’ve seen grant proposals from other researchers to do precisely this with plants).
I’ve written about this kind of thing on the Oikos blog, and probably need to do so again. Just because you’re interested in “large scale” patterns doesn’t mean “small scale” expts. are somehow irrelevant. In particular, if you want to tell a story about how underlying “small scale” processes of birth, death, and dispersal gave rise to your large scale pattern (which is something macroecologists very often want to do), then “small scale” experiments directly testing whether those processes operate the way you think they do are not just relevant but surely essential.
Shorter version: if you think that the large scale is just the aggregate outcome of what happens at smaller scales (and how could it be otherwise?), then you surely ought to directly study what happens at the small scales, as well as the large-scale consequences.
http://oikosjournal.wordpress.com/2011/11/22/species-pools-and-the-fallacy-of-composition/
http://oikosjournal.wordpress.com/2011/05/03/why-macroecology-needs-microecology-revisiting-an-oikos-classic/
http://oikosjournal.wordpress.com/2012/02/07/drilling-down-vs-scaling-up/