1. The quest for ‘assembly rules’, i.e. the processes shaping the species composition of communities, is a central issue in community ecology. Nevertheless, so far there is no general agreement on a framework to detect assembly rules in real life data: several key elements are still missing or heavily disputed, including the choice of the appropriate test statistic (e.g. functional diversity index) and randomization strategy for each major assembly process.
2. Simulation studies based on artificial communities can help to explore the usefulness of different approaches in detecting assembly rules. Nevertheless, the currently dominant approach to simulate artificial communities (i.e. selecting species from a pool based solely on trait values) oversimplifies the complex processes involved in community assembly and thus fails to produce realistic patterns. Consequently, its value for testing methodologies is seriously limited.
3. In this study we implemented a flexible, individual-based algorithm simulating real-life community processes (individuals are born, survive, compete for resources, reproduce and die), to generate artificial species composition data. With the help of this algorithm, we estimated the type I error rates and the statistical power of five different diversity indices (FRic, Rao’s quadratic entropy, FEve, the variance of functional distances, and the variance of nearest neighbor distances) in combination with three randomization strategies (randomization of trait values in the whole dataset, within plots and within the range of trait values occurring in each plot) for detecting two underlying assembly processes (habitat filtering and limiting similarity). We also tested the influence of all adjustable simulation parameters on the simulation results in a sensitivity analysis framework.
4. The results of the sensitivity analysis show that the individual-based simulation framework proposed here can be used for creating artificial community data with realistic pattern of trait values. Based on the results, Rao’s quadratic entropy performed best for detecting both habitat filtering (trait convergence) and limiting similarity (trait divergence). Functional richness may also be suitable for detect traiting convergence. Functional evenness and variance of nearest neighbor distances, however, should not be used for finding assembly rules.