Print PageThe analysis of species co-occurrence patterns has a long history in ecology, dating to the early 1970’s. Analysis of co-occurrence patterns played a central role in earlier debates about the importance of competition in structuring ecological communities. These early studies include some of the classic work in null model ecology. Many of these null models and their deterministic-model counterparts were context specific without any formal generalized framework. Analysis of co-occurrence patterns eventually evolved into the more formal approach of testing for nestedness in species presence-absence matrices. This remains a huge area of ecological-statistical endeavor. Ecologists have proposed many metrics for quantifying non-randomness in these matrices and have developed various computer algorithms for randomizing presence-absence matrices. These metrics and algorithms all have different strengths and weaknesses regarding their statistical properties, computational complexities, and biological realism. However, most of the metrics are designed to test for pattern at the level of the entire matrix and all of the algorithms randomize species among sites in a way in which it is not possible to know if the entire null space has been realized. This black box of simulation is at the heart of Type I and II error in these analyses, but can be overcome with a probabilistic model of species co-occurrence. The model uses combinatorics to analytically derive the exact probability that two species would co-occur at a frequency greater than what is observed. Using the model, I reanalyzed pairwise co-occurrence patterns in some classic datasets that included Diamond’s New Guinean birds, Brown’s North American desert rodents, and Haefner’s Carribean Anolis lizards.
Results/Conclusions
Overall, in most datasets, there were more species pairs that were positively associated (10 – 75% of species pairs) than were negatively associated (2 – 10%). Negative associations were least common in New Guinean bird assemblages (3% of species pairs) and most common in desert rodent assemblages (5 – 16%). Also, most datasets had a relatively high percentage of random species associations, often 50% or greater. The preponderance of positive and random associations points to the prevailing importance of dispersal and habitat in determining co-occurrence and a lesser role for competition. The probabilistic model of species co-occurrence could become widely useful as a practical tool for analyzing co-occurrence as well as a general framework for viewing community assembly and distributional patterns.
