I turn now to an important ambiguity in the concept of multi-level selection, discussed by authors including Arnold and Fristrup (1982), Sober (1984), Mayo and Gilinsky (1987), Damuth and Heisler (l988), Okasha (2001) and others. The ambiguity arises because there are two things that multi-level selection can mean, that is, two ways that the basic Darwinian principles can be extended to a hierarchical setting. The ambiguity has usually been discussed in relation to individual and group selection, but it generalizes to any multi-level scenario, for it stems from the distinction between collective fitness! and fitness2.
Consider again the scenario depicted in Figure 2.3, where a number of particles are nested within each collective. The key issue is whether the particles or the collectives (or both) constitute the 'focal' level.17 Are we interested in the frequency of different particle-types in the overall population of particles, which so happens to be subdivided into collectives? If so, then the particles are the focal units; the collectives are in effect part of the environment. Alternatively, we may be interested in the collectives as evolving units in their own right, not just as part of the particles' environment. If so, we will wish to track the changing frequency of different particle-types and collective-types. Following Damuth and Heisler (1988), I refer to the first approach as multi-level selection 1 (MLS1), the second as multi-level selection 2 (MLS2).
The distinction between MLS1 and MLS2 dovetails with the distinction between collective fitness1 and fitness2. To see this, consider an example ofMLS1: D.S. Wilson's (1975) 'trait-group' model for the evolution of altruism. Organisms are of two types in this model: selfish and altruist. They assort in groups for part of their life cycle, during which fitness-affecting interactions take place, before blending into the global population and reproducing. Within each group, altruists have lower fitness than selfish types. But groups containing a high proportion
17 In Sober's (1984) terminology, the issue is whether the particles or the collectives are the 'benchmarks of selection'.
of altruists have a higher group fitness!, that is, contribute more individual offspring to the global population, than groups containing a lower proportion. So within-group selection favours selfishness, while between-group selection favours altruism; the overall outcome depends on the balance between the two selective forces. Wilson's model is thus designed to explain the changing frequency of an individual trait—altruism—in the overall population. Although the explanation makes essential appeal to group structure, and treats groups as fitness-bearing entities, it permits no inference about the frequency of different types of group. Both levels of selection contribute to a change in a single evolutionary parameter.
By contrast, consider D. Jablonski's hypothesis that the average geographic range of late-Cretaceous mollusc species increased as a result of species selection (Jablonski 1987). Here the explanandum is the fact that species with large geographic ranges became more common, in a particular mollusc clade, than those with smaller ranges. The suggested explanation is that species with larger geographic ranges had greater fitness2, that is, left more offspring species, and that geographic range was heritable. Note that this hypothesis permits no inference about the frequency of different types of organism, even though the species character in question—geographic range—presumably depends on organismic characters, such as motility and dispersal. Within each species, these characters can evolve by selection at the level of the individual organism, so there is a potential interplay between the two levels of selection. But the key point is that fitnesses at each level are independently defined; so selection at each level leads to a different type of evolutionary change, measured in different units. This is the hallmark of MLS2.
Note that in Jablonski's example, the collective character subject to selection—geographic range—is emergent, but in Wilson's model it is aggregate—proportion of altruists in a group. Does it follow that the MLS1/MLS2 distinction always lines up with the aggregate/emergent distinction? The answer seems to be no. Variation between collectives with respect to emergent characters could influence the number of offspring particles they leave—in which case MLS1 would operate on an emergent character. Conversely, variation between collectives with respect to aggregate characters could influence the number of offspring collectives they leave—in which case MLS2 would operate on an aggregate character. This suggests that the MLS1/MLS2 distinction crosscuts the aggregate/emergent distinction; the issue is probed further in Chapter 4.
One might think that the MLS1/MLS2 distinction dovetails with the distinction between collectives that persist for one or a few particle generations, and those that persist for many. But this is not quite right. It is certainly most natural to define collective fitness as fitness2 when particle and collective generations are non-synchronized, as in the species selection example, and as fitnessi in cases where generations are synchronized, as in the trait-group example; but naturalness is not logical necessity. Even if a collective persists for many particle generations, one could still define collective fitness as average particle fitness, that is, in the MLS1 way. Conversely, even if generations are synchronized, one could still define a collective's fitness as number of offspring collectives, that is, in the MLS2 way. The essence of the MLS1/MLS2 distinction concerns the units whose demography we wish to track, which is orthogonal to the issue of their respective generation times.
One important difference between MLS1 and MLS2 is this. In MLS2, collective fitness is defined as number of offspring collectives. For this notion to apply, it is essential that the collectives reproduce in the ordinary sense, that is, that they 'make more' collectives; otherwise, determinate parent—offspring relations at the collective level will not exist. But in MLS1 this is inessential. The role ofthe collectives in MLS1 is to generate population structure for the particles, which affects their fitnesses. For MLS1 to produce sustained evolutionary consequences, collectives must 'reappear' regularly down the generations, but there is no reason why the collectives themselves must stand in parent—offspring relations. This is significant because, as the slime mould example illustrates, where collectives are formed by aggregation of many particles of different ancestry, the notion of collective reproduction in the sense of collectives 'making more' collectives becomes strained. In such cases it is difficult to apply MLS2 concepts, but straightforward to apply MLS1 concepts.
There is a recurring tendency in the literature to argue that MLS1 does not constitute 'real' multi-level selection at all (Maynard Smith 1976; Vrba 1989). Proponents of this view often claim that MLS1 isjust lower-level selection with frequency-dependent fitnesses, so involves only one level of selection. Two points about this view deserve mention. First, it has the unwelcome consequence that most models ofgroup selection do not deal with real group selection at all. For as Damuth and Heisler (1988) note, such models have typically been of the MLS1 variety, though they are often informally glossed using MLS2 language. This consideration is not decisive, but it does suggest that MLS1 and MLS2 should both be classified as multi-level selection, albeit of different logical types.
Secondly, although MLS1 treats the particles as the focal units, it can nonetheless shed light on collective-level phenomena. For since collectives are composed of particles, explaining the evolution of a particle character could help explain salient features of collectives too. This is especially so given that MLS1 models often focus on the particles' social behaviour, for example, their tendency to behave altruistically towards other particles, or to aggregate with them in collectives, or to police their selfish tendencies. A theory that explained how particles evolved such traits could be the first step in explaining the existence of cohesive collectives, whose constituent particles work for the good of the whole. So although the explanatory target of an MLS1 model is the change in frequency of a particle character, not a collective character, this does not mean that MLS1 can tell us nothing about the collectives themselves.
Describing MLS1 and MLS2 as different 'approaches' to multilevel selection may invite a conventionalist interpretation—as if the choice between them were a matter of taste. But this interpretation should be resisted. In a single-level scenario, whether a given trait evolves by natural selection is a matter of objective fact. Of course, our explanatory interests determine which traits we are interested in, and thus the sorts of evolutionary model we construct, but that is a different matter. The same is true of multi-level selection. MLS1 and MLS2 are distinct processes that can occur in nature; whether either occurs in a particular case is a matter of objective fact. But our explanatory interests may determine which process we wish to model, and thus which definition of collective fitness we choose.18 Any conventionalism here is of the innocuous sort that arises because all scientific investigations must focus on some aspects of nature at the expense of others.
Was this article helpful?