Despite the complexity of the natural world, community ecologists have made significant progress by treating species pairs as the fundamental unit of study. Recently, however, the extent to which insights derived from two-species systems scale to multispecies communities has come under increased scrutiny. In particular, the role of higher-order interactions (HOIs) - the suite of non-additive effects that emerge when interaction strengths vary with competitor densities - has emerged as a focal point for research, and yet it is important to recognise HOIs for what they are: emergent properties of the phenomenological models we use to study competition. As such, they may provide limited insight into the underlying processes that drive them. In contrast, non-additive dynamics emerge naturally in mechanistic models (e.g., through saturating growth functions in standard consumer-resource models), without the need for higher-order terms. To improve understanding of processes underlying HOIs, we explored the emergence of non-additive dynamics (i.e., HOIs) in mechanistic consumer-resource models under a variety of different assumptions. More specifically, we: 1) generated `observed' data by sampling densities and growth rates from simulated time-series of mechanistic models; 2) made statistical fits of phenomenological models (with and without higher-order terms) to the observed data; and 3) contrasted time-series produced by the original mechanistic models with those produced by phenomenological models parametrized in the second step.
Results/Conclusions
Our main and perhaps most surprising finding is that higher-order interactions emerge in all but the most simple consumer-resource models. Of all the different model combinations we considered, only in the limiting case of two consumers a) characterised by linear functional responses, and b) competing for a pair of essential resources growing logistically, did phenomenological models lacking higher-order terms perform on a par with those including them. Any deviations from these assumptions resulted in non-additive dynamics better captured by higher-order models. Given the complexity of the natural world, we must therefore infer that higher-order dynamics are a ubiquitous feature of real systems, and that ignoring them could lead to erroneous inference and wildly inaccurate predictions. Alternatively, one could take a simpler path and just use a mechanistic model.