Inference from plant competition experiments: the effect of spatial covariance

Two‐species plant competition experiments are often drawn on in an attempt to understand and possibly to predict the formation of plant community structures. Frequently, such competition experiments use a non‐random spatial design but is analysed in models where a random spatial design is implicitly assumed. Here, the effect of spatial covariance in plant competition experiments on the inferred ecological conclusions is demonstrated. It is shown that spatial covariance, which is dependent on the competition kernel and the experimental design, may affect the estimation of parameters of interest. Based on the same set of experimental data, the predicted outcome of competition between two Arabidopsis thaliana genotypes can vary between (almost certain) coexistence and (almost certain) exclusion of one genotype, depending on the assumptions made about the spatial scale of competition between the plants . The predicted outcome of competition depended strongly on the mean interaction distance of the competition kernel, whereas the functional shape of the competition kernel was less important. It is suggested that plants in manipulated competition experiments should be placed randomly, such that the design of the competition experiment is in agreement with the model used in the analysis.