Maynard Smith's argument is simple. The group selection question is about 'whether there are entities other then individuals with the properties of multiplication, heredity and variation, and that therefore evolve adaptations by natural selection' (1987a p. 121). In the process envisaged by Wynne-Edwards, the groups do possess these proper-ties—for they are 'sufficiently isolated for like to beget like' (p. 123). But in Wilson's trait-group model, where the groups periodically blend into the global population, this is not so. A given trait group 'does not give rise to a group of the same composition', Maynard Smith says (p. 124). Therefore trait groups do not exhibit heredity, so cannot evolve adaptations, so cannot be 'units of evolution'.
Importantly, Maynard Smith is not denying that the trait-group model involves selection at the group level (though in an earlier 1976
paper he did deny this). Rather, his claim is that the mechanism at work in the trait-group model is (a) fundamentally unlike traditional group selection, and (b) incapable of producing group adaptations.
In reply to Maynard Smith, Sober (1987) wrote: 'I agree that the process Wynne-Edwards postulated involves groups that exhibit heredity. David Wilson's. . . groups require no such thing' (p. 133). But this difference is irrelevant, Sober argued, because group heredity (heritability) is not needed for evolution by group selection. He wrote: 'for natural selection to produce evolution, heritability of some sort is essential. But for group selection to cause evolution it is not essential that the heritability be group heritability' (p. 136, emphasis in original). So group selection can lead to the evolution of altruism—a group adaptation—without group heritability, Sober claims.
In their recent book, Sober and Wilson (1998) make a different reply to Maynard Smith. They agree with him that the evolution of group adaptations requires group heritability—the point that Sober (1987) denied. They also agree that for selection at any level to cause evolution, there must be heritable variation in fitness at that level. But trait groups do satisfy the heritability requirement, they claim—despite what both Sober and Maynard Smith thoughtin their 1987 exchange. Foralthough trait groups periodically blend, we can still identify parent—offspring lineages among them, and thus consider the heritability of any group character. Blending means that trait groups have multiple parents—but this is simply the analogue of sexual reproduction. At the individual level, sexual reproduction reduces the heritability of individual characters vis-a-vis clonality, but does not make the concept inapplicable. The same is true at the group level, Sober and Wilson argue.
Therefore, there are three competing views of the role played by group heritability in models of group selection, summarized in Table 6.2.
Which of these views is correct? Despite their apparent incompatibility, I think each is partly correct. In Chapter 2, we saw that there are actually two notions of group heritability. Group heritability2 measures the resemblance between parent and offspring groups; it is relevant to multi-level selection of the MLS2 type. Group heritability! measures the resemblance between a parent group and the set of offspring individuals that it produces; it is relevant to MLS1. Recall also that in MLS1, it is possible to write the total evolutionary change in terms of the global individual heritability, without mention of group heritability.
When Sober (1987) argued that in the trait-group model, what matters is that there should be individual heritability, not group heritability,
The Group Selection Controversy Table 6.2. The role played by group heritability
Is group heritability Do trait-groups Do trait-group and necessary for groups exhibit group 'traditional' group to evolve adaptations? heritability? selection belong together?
Maynard Smith 1987a
Sober & Wilson 1998
yes no yes yes no no yes yes no he was partly right. For the trait-group model has an MLS1 structure, so group heritability2 is irrelevant. As we saw in Chapter 2, in MLS1 it is not necessary for the groups to stand in parent—offspring relations at all, nor therefore for parent—offspring resemblance at the group level. So Sober's assertion is correct, if he means group heritability2.
When Sober and Wilson (1998) argue that group heritability is in fact required in the trait-group model, they too are partially correct. In MLS1, group selection means some groups contributing more individuals than others to the next generation.15 For this to produce an evolutionary response, there must be a resemblance between a group and the set of offspring individuals that derive from it. So Sober and Wilson's assertion is correct, if they mean group heritability1.
However, Sober and Wilson (1998) do not distinguish the two types of group heritability; they appear to be talking about group heritability2. They stress that parent—offspring relations between trait groups can be discerned, permitting us to ask whether parent groups tend to resemble their offspring groups. But this question is irrelevant, given that the groups are not the focal units in the trait-group model.
It follows that in one respect, Maynard Smith (1987a) and Sober and Wilson (1998) are both mistaken. Maynard Smith argues that trait groups can't be 'units of evolution', since they do not 'give rise to other groups of the same composition' (p. 123). Sober and Wilson reply that trait groups do give rise to groups with similar composition. But since the trait-group model is of the MLS1 type, what really matters is that a trait group should give rise to a set of offspring individuals of
15 This assumes the Price approach to MLS1; the issue between the Price and contextual approaches does not matter for the moment.
similar composition. When this condition is satisfied, there may be a resemblance between parent and offspring groups too; but this is an incidental side effect.
This point is important because periodic blending affects the two types of group heritability differently. Blending greatly reduces the group heritability2 from what it would be if groups reproduced asexually, for example, by fission. So in a model of the MLS2 type, blending limits the possible response to group selection. But blending has no effect on the group heritability1. The resemblance between a parent group and the set of offspring individuals it produces will be exactly the same, whether or not those individuals blend with the progeny of other groups to form the next generation of groups.
To sum up: the Sober (1987) and Sober and Wilson (1998) positions are both correct, in different senses of group heritability. The dispute between Maynard Smith (1987a) and Sober and Wilson (1998) is misplaced; both parties miss the point that group heritability1 is the relevant quantity in the trait-group model.
Where does this leave us? Maynard Smith argued that trait-group selection cannot produce group adaptations, and should not be co-classified with 'traditional' group selection. On the latter point, he is surely wrong. While it may be true that Wynne-Edwards was imagining an MLS2 process, most traditional group selection models were of the MLS1 type. So the fact that in the traditional models the groups were isolated demes that did not blend is beside the point; for the evolutionary mechanism was the same as in the trait-group model, just as Sober and Wilson say.
What about the first point? Since trait-group selection can lead to the evolution of altruism, which is group beneficial, surely Maynard Smith is wrong here too? But this might be disputed. For the trait-group model explains the changing frequency of different types of individual, not group. So while it is true that altruism is group-beneficial, and evolves for that reason, arguably it is an individual adaptation, for it is a property of individuals, not groups. The corresponding group property is 'proportion of altruists'; but group selection of the MLS1 type does not explain the evolution of that property, as stressed previously.
This issue was not discussed in previous chapters, for our focus was on selection rather than adaptation. How should it be resolved? On the one hand, the idea that group adaptations must be properties of groups, rather than individuals, seems natural. On the other hand, this severs the link between selection and adaptation, presuming we allow that MLS1 counts as genuine multi-level selection, for it implies that traits that evolve by selection at a given level need not be adaptations of entities at that level, which is counter-intuitive. I see no conclusive way of deciding the issue. Ultimately, how we use the expression 'group adaptation' does not matter greatly; what is important is to be clear about the logic of the underlying selection process.
Was this article helpful?