Neutral Theory Starts from False Assumptions
McGill et al. (2006) aptly summarize the aversion to neutral theory as follows: “neutral theories of biodiversity assert that all individuals of all species are competitively identical. […] This contradicts 100 years of community ecology”. Evidently, neutral theory does start from assumptions that are partly false, but so do many other theories for reasons of simplicity, elegance and practicality (e.g. metapopulation theory of Levins 1969; see Etienne 2000, 2002; or the metabolic theory of ecology, of Brown et al. 2004; see Etienne et al. 2006; Apol et al. 2008). The most obvious example in the case of neutral theory is the assumption of neutrality: equivalence between individuals belonging to different species. However, other auxiliary assumptions exist and can be relaxed (Leigh 2007), for example concerning speciation (Etienne et al. 2007a; Etienne and Haegeman 2011; Rosindell et al. 2010), spatial structure (Rosindell and Cornell 2007, 2009) and the zero-sum rule (Etienne et al. 2007b; Haegeman and Etienne 2008), and this often does not affect the theory’s predictions or produces predictions that match observations better.
Although neutral theory does assert, for the sake of simplicity, that all individuals are equivalent, this should not be regarded as a weak spot of the theory. Rather, it is its strongest asset: it makes tractable models that describe the dynamics of an ecological system without adaptation. The only reason that neutrality is regarded as an extra assumption is that so many other models include non-neutral details. To say that ‘all individuals of different species are ecologically equivalent’ is arguably not an assumption, but simply a lack of detail about the specifics of ecological interactions given by competing theories. If a lack of information is generally considered as an assumption, the list of ‘assumptions’ for almost any model can readily become long and preposterous. For example imagine the phrase ‘this model assumes that no evolution occurs and no new species evolve, that organisms move unrestricted about the ecosystem, that conspecific individuals have no genetic differences, that seasonality and the weather have no influence on the ecosystem, that species never go extinct, that parasites do not exist,…’. We do not generally provide such a list with models. Instead we are more likely to see ‘seasonality is modeled by assuming…’ or simply no mention of seasonality. With the construction of such a list of assumptions any person could, with careful rhetoric, make any model appear worthless. Compare the neutral theory of community ecology with that of population genetics: examples of evolution by natural selection are not generally regarded as proof that genetic drift does not exist. Rather, models based only on genetic drift describe the dynamics of a population’s genetic diversity in absence of natural selection, and a baseline against which to compare data of cases where natural selection is present.
Neutral Theory Makes False Claims
From an instrumentalist perspective, false assumptions are not a proper reason to dismiss a theory; theories should not be evaluated based on their assumptions but on their predictions. Friedman (1966) put it even more strongly:
“Truly important and significant hypotheses will be found to have ‘assumptions‘that are wildly inaccurate descriptive representations of reality, and, in general, the more significant the theory, the more unrealistic the assumptions (in this sense). […] The reason is simple. A hypothesis is important if it ‘explains‘much by little, that is, if it abstracts the common and crucial elements from the mass of complex and detailed circumstances surrounding the phenomena to be explained and permits valid predictions on the basis of them alone.” (Friedman 1966)
Ricklefs (2003) was one of the first to cut past the assumptions, looking instead at neutral theory’s predictions, which he deemed to be not very impressive. Some of this original criticism (and the similar criticism of Nee 2005) can, however, be attributed to immature elements of the theory at the time. Rosindell et al. (2010) solved some of the problems that Ricklefs raised by making speciation in the neutral model more realistic. McGill (2003) and McGill et al. (2006), even though they consider neutral theory’s assumptions to be false, also (in line with Friedman’s philosophy) evaluated its predictions and thus made use of the theory in the proper scientific way. From an extensive literature review of tests of neutral theory vs. niche-assembly theory they concluded that there is little evidence to support neutral theory. Similarly, Dornelas et al. (2006) claimed that the data they find in coral reefs does not match the predictions made by neutral theory. Whilst some of these failures of neutral theory may again be attributed to auxiliary assumptions (in the case of the corals, Volkov et al. 2007 show that a modified neutral model can in fact produce results that fit the observed coral data well, although they require low immigration rates to achieve this, which are considered unrealistic for corals), there are of course cases where neutral theory’s predictions will not match observations. And even if they do, this does not necessarily imply that niche-based mechanisms are unimportant, because neutral pattern does not imply neutral process (see for example Du et al. 2011).
From a classical instrumentalist perspective, failure of the neutral theory (even after it has matured) should lead to rejection of the theory. We agree with this procedure, but we argue that there is a second instrumentalist role to play for neutral theory. Failure to match empirical patterns does not mean that theory is useless. Instead it means that the theory has done its job in highlighting that something other than dispersal limitation or ecological drift plays an important role that is visible
in the data being studied (as opposed to a role that is not visible in the data being studied, which is probably played by many (non-neutral) factors) (Pearson and Gardner 1997; Rosindell et al. 2012). Although this may sound like we are attempting to make neutral theory invincible in some way, it is crucially the usefulness of neutral theory as a tool and not the claim that the world is neutral that we are defending here. When neutral theory fails, it does so in an informative manner: we must extend it to properly explain the system. If neutral theory succeeds, we may not need to look further for understanding the system, or we should question whether our data set is informative enough. For example, in the case of coral diversity we argue that refuting the theory should not be interpreted too negatively. Failure of neutral theory shows that there must be alternative mechanisms. Dornelas et al. (2006) hypothesize that environmental heterogeneity may be playing an important role.
Thus, neutral theory has a dual instrumentalist function: like any other theory, it can be used to predict patterns, but unlike many other theories, it is well positioned to act as a starting point, a baseline model to which one can later add more ecological mechanisms. It is exactly when it makes predictions that are not supported by empirical data that this second role is played. The following analogy illustrates our point. If one uses a gauge to test tire pressures, and finds the tire is under filled, the pressure gauge has not failed, but it has succeeded and done its job, even though the result of the test it was used for was a failure.
Attacks on neutral theory because of its inability to match some empirical data fail to see the role of the theory in a broader context; that of one theory amongst others which can be of practical importance to highlight different aspects of a particular problem that usually go unnoticed or receive less attention than they should. Niche differences have long intrigued biologists, and rightfully so, but they are not the only factors governing diversity patterns in ecological communities (Vellend 2010).
In summary, we argue that neutral theory should not be seen as just a null model in the statistical sense, one that can be rejected, but rather as a baseline model that contains necessary ingredients that more advanced models should often also contain (similar to the role of the neutral theory of population genetics that emphasized the omnipresence of genetic drift). We believe that starting from neutral theory is much easier than starting from a model that assumes niche differentiation from the outset. There might be other simple starting points as well. In this context, McGill (2010) identified three assumptions (which he calls “rules”) from which most of the (approximately) correct predictions of neutral theory and five other ‘unified’ theories can be derived: (1) intraspecifically individuals are clumped together; (2) interspecifically global or regional abundance varies according to a hollow curve distribution; and (3) interspecifically individuals are placed without regard to individuals of other species. These rules are statistical rather than mechanistic, but for the instrumentalist, this does not matter.
Neutral Theory is Not a Good Theory Because it Does Not Fit with the Philosophical Paradigm of Realism
We argue that this opinion about neutral theory is often held by critics, at least implicitly or without realizing it. This is not an argument that is often expressed explicitly, and consequently it is hard to settle. A good example of a realist plea is the opinion paper by Clark (2008), in which he argues that neutral theory does not contain a real process because its stochastic elements that are perceived as neutral forces are simply standing in for smaller scale non-neutral processes that are not modeled explicitly. It appears that Clark (2008) believes neutral theory to be a contrasting view on the real workings of the world, and hence he charts the ontological dichotomy. He goes on to argue that stochastic elements only exist in models and although he admits utility of such models for understanding and explanation, he argues that there are more promising alternative directions. This illustrates the epistemological dichotomy.
The dispute between niche-based and neutral theory is quite similar to the situation between the neurosciences and folk-psychological accounts of behavior. Niche-assembly is in itself a more realist theory than neutral theory; it focuses on detailed interactions between individuals and species and from this ‘real’ base builds up a picture of the ecological community as a whole. Neutral theory partly works top-down: local community structure is determined, to a considerable extent, by processes at the metacommunity level, for the simple reason that describing matters in that way makes them easier to understand. There is no ontological primacy given to either level.
On a more abstract level there is the question of whether we should try to choose between two seemingly conflicting theories, or try to reconcile them. This is in itself a choice between a realist and an instrumentalist approach, which explains why it is no wonder that the former option—conflict—is often preferred by those supporting niche-assembly theory (although significant reconciliatory attempts are also made by some proponents of niche-assembly; e.g. Leibold and McPeek 2006) and the latter option—reconciliation—is preferred by those tolerating neutral theory. Someone who takes a realist or instrumentalist approach in one area will likely have the same preference in another area.
The main reason why arguments about testing neutral against niche theory sound so convincing is that the realist story behind these tests is so appealing. We claim that denying the possibility of reconciliation between the two theories is ultimately counterproductive to getting a full understanding of biodiversity patterns and how ecosystems work. Even though a realist approach can advance science, so can an instrumentalist approach which should merit its application to understanding the origin and maintenance of biodiversity.
There has always been a strong realist tendency in the natural sciences and in biology in particular. There is value in this; a realist approach means that there cannot be two competing explanations for a single phenomenon and science should progress by decisive tests eliminating false alternatives (Platt 1964). Ultimately, the realist approach leads to reducing the problem into simpler sub-problems that can then be solved one by one, eventually being recombined to solve the original problem in its entirety. This approach does not work for every problem however, and over-applying it could make us miss emergent patterns that are only clearly observable when looking at the system as a whole (Quinn and Dunham 1983).