Designs for greenhouse studies of interactions between plants

1. Designs for greenhouse studies of interactions between plants are reviewed and recommendations for their use are provided. 2. Papers published over a 10-year period showed the replacement series design to be the most popular, especially in studying crop weed interactions. Fifty per cent of the studies involved only two species, although studies testing the interaction between different genotypes of only a few species were also popular. 3. Limitations imposed by the choice of design, the variables measured, and the analysis used on the range of inferences that may be validly drawn from the experiment are frequently not well understood or appropriate for the questions that appear to be addressed. One example is the failure to distinguish the outcome of competition (the long-term outcome of interaction) and the effects of species on each other. 4. Studies in which only final yield is measured are severely limited as to the inferences which may be drawn. Effects due to interspecific interaction during the course of the experiment cannot then be separated from pre-existing differences, and interpretation may be biased towards species whose individuals were initially larger. In addition, measurements at several times are necessary to understand the changing dynamics of species interaction. 5. Simple pair-wise mixtures can assess the effect of treatment factors on the outcome of competition. Replacement series and related diallel designs generally produce results that may be size-biased even when initial interspecific differences are known. Additive designs (including target-neighbour designs), despite confounding density with species proportions, offer considerable scope for addressing mechanistic questions about interspecific interactions. Designs that allow response surface analysis can avoid many of the problems inherent in the other methods, but all need to be adjusted for initial interspecific differences. Designs for multiple species experiments are still largely untested, although several designs have been used. At the level of the individual plant, hexagonal fan designs permit study of the effects of varying the spatial pattern, and the densities and the relative proportions of interacting species, but suffer from lack of independence and lack of randomization.