Generating and testing predictions about community structure: Which theory is relevant and can it be tested with observational data?

Three issues are discussed relevant to the controversy over using null models and observational data on guild structure to test community-level predictions based on limiting similarity theory. First, I argue that most limiting similarity theory is not based on reasonable assumptions for plants and that the theory that is relevant does not generate any predictions about expected guild proportionality on a small spatial scale. Therefore, regardless of adequacy of the statistical methods, the predictions being tested by the body of literature using null models to test for niche limitation are unlikely to be relevant in most plant comunities. Second, assuming that the predictions are after all worth being tested, I argue that most tests using the guild approach do not provide adequate explanations of how the defined guilds could lead to greater competition within vs. between guilds. If this is not true for the particular guilds used, the predictions of guild proportionality or size constancy will not be valid. Third, I address the controversy over whether field experiments can provide more solid evidence than observational data about the role of competition in determining community structure by (1) suggesting methods of dealing with potential drawbacks of field experiments, and (2) suggesting alternative experimental approaches for directly addressing issues about community structure.