Do We Need a New Account of Group Selection? A Reply to McLoone

Abstract

In “Some Criticism of the Contextual Approach, and a Few Proposals” in Biological Theory, Brian McLoone discusses some of the points about the contextual approach that I made in a recent paper. Besides offering a reply to McLoone’s comments on my paper, in this article I show why McLoone’s discussion of the two main frameworks for thinking about group selection—the contextual and the Price approach—is partly misguided. In particular, I show that one of McLoone’s main arguments against the contextual approach is missing the target and that one of his (and Elliott Sober’s) arguments in defense of the Price approach is flawed. Criticizing these arguments will help me present an entirely different picture than McLoone’s of the current status of multilevel selection theory. More precisely, I argue that the idea that we are dealing with “multilevel selection” in the type of multigroup cases in which the focal units are the individuals (and their traits) has recently come under threat. Finally, I discuss the ways in which this idea might be salvaged by appealing either to the contextual or to the Price approach.

Appendices

Appendix 1

Applying the Price and Contextual Approaches to the Case Used for the Reformulation of the Overstatement Argument

The fitness values of the two types within the two groups in version (a) are:

$$ \begin{aligned}&w_{xO} = 10.02 \quad w_{xP} = 10.08\\ &w_{yO} = 2.02 \quad w_{yP} = 2.08 \end{aligned}$$

If we plug these values into Eq. 1, we get the following partition of the evolutionary change:

$$ \Delta \overline{z} = \overbrace {{\frac{{E_{j} \left[ {Cov(z_{ij} ,w_{ij} )} \right]}}{{\overline{w} }}}}^{\begin{subarray}{l} individual \\ selection \end{subarray} } \,+ \, \overbrace {{\frac{Cov(Zj,Wj)}{{\overline{w} }}}}^{\begin{subarray}{l} group \\ selection \end{subarray} } = \overbrace {0.2115}^{\begin{subarray}{l} individual \\ selection \end{subarray} } + \overbrace {0.1205}^{\begin{subarray}{l} group \\ selection \end{subarray} } $$

This means that, according to the Price approach estimation, group selection is responsible for about 36 % of the total change in version (a).

On the other hand, if we plug the same fitness values into Eq. 2, we obtain:

$$ \beta_{1} = 8 \quad \beta_{2} = 0.1 \quad \alpha = 2 $$

Plugging the partial regression coefficients into Eq. 3 will show that, according to the contextual approach, the contribution of group selection to the evolutionary change in version (a) is about 0.44 %:

$$ \Delta \overline{z} = \overbrace {{\frac{{\beta_{1} Var(z_{ij} )}}{{\overline{w} }}}}^{\begin{subarray}{l} individual \\ selection \end{subarray} } \, + \, \overbrace {{\frac{{\beta_{2} Var(Z_{j} )}}{{\overline{w} }}}}^{\begin{subarray}{l} group \\ selection \end{subarray} } = \overbrace {0.3305}^{\begin{subarray}{l} individual \\ selection \end{subarray} } + \overbrace {0.0015}^{\begin{subarray}{l} group \\ selection \end{subarray} }. $$

Appendix 2

Discussion of McLoone’s Stag Hunt Case

In order to understand McLoone’s Stag Hunt case, all we need is to go back to our initial example in this article and suppose that all the altruists are clumped into one group and all the selfish individuals are in the other group. Our groups thus become “clonal groups” (Gardner and Grafen 2009). According to McLoone, the Price approach will have difficulties in spelling out which process is responsible for the resulting evolutionary change. Since individual selection is within-group selection for the Price approach, it cannot pretend that we are dealing with individual selection in this case (within each group, all individuals have the same fitness). On the other hand, a supporter of group selection sensu the Price approach apparently cannot claim that this is a case of group selection either, since the group containing only selfish individuals is not—McLoone argues—really a group. This is because, for reasons that will become obvious below, the Price approach is forced to assume an interactionist definition of groups. Indeed, as Sober and Wilson (1998, p. 93) put it, in Wilson’s (1975) classical model for the evolution of altruism, the term “trait group” emphasizes the idea that “groups must be defined on the basis of interactions with respect to particular traits.” But does this mean that the Price approach implies, as McLoone claims—following Okasha (2006)—that all groups of a given case must involve fitness-affecting interactions or else they do not count as groups? I am not convinced about this.

Imagine a biologist comes across two groupings (Q and R) of individuals of two types (X and Y): each grouping is composed exclusively of individuals of one type. By “grouping” I understand here a set of individuals whose status as a group is yet to be determined. The Xs are fitter than the Ys. But the biologist cannot know whether they are fitter because of their own properties (i.e., by simply being X) or because they cooperate in some way. In order to find an answer to this puzzle, she needs to experimentally manipulate the phenotypic distribution (Wade and Kalisz 1990). By creating multiple groupings containing various frequencies of both types, she can get a better idea about the extent to which the fitness differences between types in her initial empirical case were determined by the properties of individuals or by the cooperation between them. Imagine she finds out that, regardless of the composition of groupings, the Xs remain fitter than the Ys, and the fitness of each type remains at the level encountered in the initial population: she will justifiably conclude that groupings Q and R were not groups at all, since the group characters did not influence in any way the fitnesses of individuals. This is precisely the type of case that posed problems for the Price approach and that was meant to be excluded by the interactionist definition of groups.¹⁷ However, I think the latter is too strong a requirement and by using it we risk throwing the baby out with the bathwater. To see why, imagine the opposite scenario: the biologist finds out that, in any grouping containing both types, the Xs turn out to be less fit than the Ys, and the fitnesses of both types increase with the number of Xs in the grouping. He will have good reason to infer that the Xs are altruists, in the sense that they pay a cost (relative to another type) for a behavior that increases the fitness of the other members of the grouping. Turning back to groupings Q and R, I think he would be justified in calling them “groups,” and he would also have serious reasons to claim that group selection is acting in this scenario.

Therefore, I think McLoone is right to insist that the interactionist requirement should be amended: it should not be about groups, but about scenarios. In order for fitness-affecting interactions to have evolutionary consequences, they need not occur within each of the groups, but within at least one group of that scenario. In other words, the interactionist requirement for group selection should be rephrased in the following manner: the groupings of a scenario involving multiple groupings should only be considered groups if some of them have nonzero group character values and if these values do influence the fitnesses of their individual members. This means that a group with a zero group character value might still be considered a group as long as the scenario under consideration also involves groups whose nonzero characters do influence the fitnesses of their members. This more permissive understanding of the interactionist requirement explains why, to take just one example, Gardner and Grafen (2009) see no difficulty in applying the Price approach to clonal groups.

However, once this minor correction of the interactionist requirement is operated, we notice that McLoone’s account of group selection actually collapses into the account given by the Price approach. McLoone’s conditions for the appearance of group selection prove to be nothing more than the conditions for group selection sensu the Price approach:

• (a) There are groups in a metapopulation, and these groups vary in character and fitness;

• (b) For at least one of these groups, its composition (and, consequently, its group character) affects the fitness of its members;

• (c) Group characters do not have to affect the fitnesses of all the individuals of the group, but only the fitness of at least some of its members. This condition is granted by the fact that, unlike the contextual approach, the Price approach treats cases with linear and nonlinear effects of individual and group characters on individual fitnesses in the exact same manner. Therefore, even if the group character only affects the fitnesses of some of the individuals of the group, as long as this also alters the mean fitness of the group it can still give way to group selection sensu the Price approach (see Jeler 2015).

McLoone’s new account of group selection is therefore not new at all. Finally, I will add that the Stag Hunt case does not pose any serious problems for the contextual approach either. In cases involving clonal groups, both partial regression coefficients (β ₁ and β ₂) are undefined. However, as seen above, individual selection sensu the contextual approach is defined across groups. Therefore, in principle, not much could prevent a biologist from applying simple regression analysis to this scenario and from concluding that only individual selection is at work here. However, a more thorough investigator would also experimentally manipulate the distribution of types into groups. Experimental manipulations would allow her to calculate the partial regression coefficients in groups containing both types, and this would allow her to infer, without necessarily attempting to attach numerical values to their contributions, whether group and individual selection sensu the contextual approach were acting in the original clonal-group distribution (see Okasha and Paternotte 2012).