Even if you do not habitually read Nick Gotelli's work, chances are good that if you are an ecologist you have used his software, have one of his books on your shelf (or at least studied it in Ecology class), or have had him as an editor. I don't know all that much about him personally, but when I met him I learned he is also a bluegrass musician and has a fondness for Hieronymus Bosch. This indicates a creative personality which may explain why he has been such a force in methods development and refinement in ecology, particularly involving null models.
Briefly this is how a null model works: 1. You determine a pattern that you expect to arise in your data if a certain process is occurring; classic example is if there is competition among species then you expect spatial segregation among different species, 2. You develop a metric which tells you how strong this patterning is in a given sample, 3. You calculate the metric for a given sample, 4. You compare the calculated metric to those for a large sample of simulated data (usually based on resampling of your real data) which are patterned only by random chance; the algorithm that generates these random samples is your null model, 5. If your real value is more patterned than a large proportion (let's say 95%...but not necessarily) of the simulated samples, you would infer that the process is actively patterning your samples. Not so conceptually challenging, right?
The idea came from Diamond's checkerboard concept (Diamond 1975) that certain species combinations do not occur due to competition (based on bird distribution across New Guinean islands). Connor and Simberloff (1979) created the null model to test whether the number of checkerboard pairs differed from chance alone...it didn't in their test. So in this case, the checkerboard (a pair of species never in the same sample) was the pattern being detected, the metric was the number of checkerboards, and it was compared to randomly structured data. This set off a storm of controversy and one of ecology's greatest debates. But regardless of where you stand on that particular debate, there is considerable utility in the null model approach and they can be applied to a wide variety of questions.
It's really cool that out of all the possible biota, Drs. Gotelli and Maestre have developed a new method using biocrusts as a model system. They applied it to field data, and Andrea Castillo's constructed experimental crusts. Every chance I get, I like to tell people what a great model biocrusts are in community ecology (and maybe landscape ecology too) for empirical tests of theory, e.g. biodiversity effects on function, the stress gradient hypothesis, intransitive competition as a diversity conserving mechanism to name a few. Fernando has really been a pioneer in this arena. I'm glad to see Dr. Gotelli also sees the utility in this study system.
Briefly, this method compares the difference in a given functional response (e.g. photosynthetic rate, enzyme activities, etc.) in samples that contain a particular species and samples which do not. It can be applied to non-experimental field data, taking advantage of natural variation (i.e. a natural experiment). For each species, it calculates a difference (D) between the mean value of some function in samples with the species present and the mean value of the function in samples with the species absent. Then to create the random expectation, the values of function are scrambled randomly among samples, and D is calculated. This process is repeated a large number of times usually until you have thousands of these randomly structured D values. Thus you have an entire distribution of null D-values with which to compare to the real D-value calculated from your data. A standardized effect size can be calculated by subtracting the mean random D from the real D-value, and dividing by the standard deviation of the randomized D-values.
Here's a video prepared by the authors about the technique:
This paper is especially interesting to me, since I have also worked on some of the data that were used to develop the method in this paper (Bowker et al. 2011). I also have tried to tease out the impact of individual species on ecosystem functions. I had the sense that there was a low level of redundancy among the macroscopic crust components, the mosses and lichens. My reasoning was that if different species exhibit unique functional profiles (a suite of effects on different ecosystem functions) then they were not redundant. My approach to this problem was to seek correlations between certain species and higher functional values in ordination space. The biggest flaw with my approach is that it is difficult to get much information on the rarest or less abundant species. I'm not sure if Gotelli's method gets around this issue. It seems it would be difficult to get a good estimate of D if you have relatively few presences of a particular species. I think it's important to note that in both papers (Bowker et al. 2011, Gotelli et al. 2011), the importance of a species must be understood in a community context. For example if Squamarina lentigera has a strong positive relationship with phosphatase activity, it does not mean that Squamarina is making large amounts of phosphatases. It may be, but it may also be altering the abundance of other organisms which do influence phosphatase. In other words the effects can be indirect. Also, the authors are careful to note that null models can be deceptive. For example, an apparent effect of the species on a function may be observed if: a. the function being considered is actually influencing the abundance of the species, not vice-versa, or b. both arise due to a third mechanism such as spatial variation in soil properties. But if you take steps in your sampling design to minimize these possibilities, you are on firmer ground. For these inferential weaknesses, null models will not supplant experimentation, but they complement it well. An experiment adding or removing all community members to determine their effect on ecosystem function can get overwhelming and impractical fast. A null model analysis can help identify a smaller set of species which can be examined in an experimental context. Furthermore, you may already have a dataset amenable to this null model analysis (I do), whereas designing and running an experiment de novo may take years. In any case, I'm looking forward to applying this new technique and the FORTRAN program supplied with the paper.
Bowker MA, Mau RL, Maestre FT, Escolar C, Castillo AP. 2011. Functional profiles reveal unique ecological roles of various biological soil crust organisms. Functional Ecology 25: 787-795.
Connor EF, Simberloff D. 1979. The assembly of species communities: chance or competition? Ecology 60: 1132-1140.
Diamond JM. 1975. Assembly of species communities. Pages 342-444 in Cody ML, Diamond JM, eds. Ecology and evolution of communities. Harvard University Press, Cambridge, Massachusetts, USA.
Gotelli, N., Ulrich, W., & Maestre, F. (2011). Randomization tests for quantifying species importance to ecosystem function Methods in Ecology and Evolution DOI: 10.1111/j.2041-210X.2011.00121.x