Aims The limitations of classical Lotka-Volterra models for analyzing and interpreting competitive interactions among plant species have become increasingly clear in recent years. Three of the problems that have been identified are (i) the absence of frequency-dependence, which is important for long-term coexistence of species, (ii) the need to take unmeasured (often unmeasurable) variables influencing individual performance into account (e.g. spatial variation in soil nutrients or pathogens) and (iii) the need to separate measurement error from biological variation.

Methods We modified the classical Lotka-Volterra competition models to address these limitations. We fitted eight alternative models to pin-point cover data on Festuca ovina and Agrostis capillaris over 3 years in an herbaceous plant community in Denmark. A Bayesian modeling framework was used to ascertain whether the model amendments improve the performance of the models and increase their ability to predict community dynamics and to test hypotheses.

Important Findings Inclusion of frequency-dependence and measurement error, but not unmeasured variables, improved model performance greatly. Our results emphasize the importance of comparing alternative models in quantitative studies of plant community dynamics. Only by considering possible alternative models can we identify the forces driving community assembly and change, and improve our ability to predict the behavior of plant communities.