1. Introduction

Selection among groups rather than individuals is not a straightforward idea, especially not ontologically; yet in order to explain the evolution of altruism and the evolution of sociality the notion of group selection is often used. The theory of group selection is typically evoked with the aim of explaining the evolutionary proliferation of a trait that reduces one's own fitness while enhancing the fitness of other group members. Since the 1960s heated debates have captured the interest of many biologists and philosophers of biology with regard to the notion of ‘group selection’.

These debates involve three different, albeit often overlapping, issues: the first involves selection (Sober and Lewontin Reference Sober and Lewontin1982), the second adaptation (Brandon Reference Brandon1990), and the third evolutionary transitions (Maynard Smith and Szathmáry Reference Maynard Smith and Szathmáry1995). First, it is necessary to determine if selection for displaying the altruistic trait has occurred mostly between groups or between individuals, that is, whether variations in fitness and in trait frequency between groups exceed those variations within groups. Second, we have to determine whether groups reproduce in a manner that ensures group-heritability of altruism. The third issue is how the evolutionary transition from solitary organisms to social groups has occurred: that is, how various cooperative adaptations have combined to bring about systematic altruism, so that individuals have lost their independent reproduction and mate only within the larger encompassing whole, i.e., a social group.Footnote ¹

This paper concentrates on the issues of selection and adaptation. I argue that specific narrow readings of the notion of ‘group’ predispose leading researchers to undermine criteria for testing group selection; and I suggest, instead, a wide-ranging image of ‘group’.

The passionate discussions over group selection concern conceptual, methodological, and empirical dimensions. I will focus on empirical tests for group selection, and not on its conceptual ambiguities (Lloyd Reference Lloyd, Singh, Krimbas, Paul and Beatty2001). The following section briefly delineates the semantic disputes from the empirical ones. Subsequently, I illustrate the empirical deficiencies involving the debates over group selection by examining two well-known experiments: Rissing et al.'s notable research on leaf-cutting ants, published in Nature in Reference Rissing, Pollock, Higgins, Hagen and Smith1989; and Wade's study on flour beetles, published in Proceedings of the National Academy of Sciences in Reference Wade1976. The last section argues that extreme images of ‘group’, as a loose aggregate or a cohesive superorganism, frame some of the empirical deficiencies regarding the debates over group selection. I suggest, instead, a wide and multifaceted meaning for ‘group’.

2. Semantical Disputes and Empirical Criteria for Group Selection

The notions of ‘group selection’ and ‘group adaptation’ both rely upon the meaning of a ‘unit of selection’. A biological level of organization, e.g., genotype, organism, or group, is also a unit of selection as long as it shows phenotypic variance, fitness variance, and heritability of traits relating to fitness (Lewontin Reference Lewontin1970). Such an entity is represented by some as a contextual background for its individual parts and as an emergent collective by others.Footnote ² The function of such a unit is still under conceptual dispute as either direct replication or interaction in a way that causes differential reproduction (Hull Reference Hull1980).

The function of replication implies that the gene is the major unit of selection, since an organism that reproduces sexually replicates only one half of its traits on average, and a group that splits into daughter groups has an even smaller average chance to replicate its trait, e.g., its frequency of altruists, to the next generation of groups. The alternative understanding of the notion ‘unit of selection’, focuses on the function of interaction with the environment, and mainly considers whole genomes (i.e., individuals) and groups as units of selection. My own conviction on this conceptual matter follows Wilson and Sober's (Reference Wilson and Sober1994) influential paper, which understands a ‘unit of selection’ as an interactor.

Supporters and opponents of group selection generally agree that one can switch between Hull's ‘replicator’ and ‘interactor’ perspectives and between Kerr and Godfrey-Smith's ‘contextual’ or ‘collective’ mathematical representations in order to explain the evolution of altruism, yet this does not end the debate.Footnote ³ On the contrary, the anticipation for empirical resolution continues as heatedly as before. For example, Wilson (Reference Wilson1983, 171), Maynard Smith (Reference Maynard Smith and Dupré1987a, 123–124, 131), and Dawkins (Reference Dawkins1982) declare that their disagreements are not about empirical facts, but about semantics or methodological convenience. Nonetheless, only a few pages later, Maynard Smith (Reference Maynard Smith and Dupré1987b, 148) claims that a fact confirms one model over another; and, on the very same page, Wilson (Reference Wilson1983, 171) claims that he uses semantic terms in a true manner and that there is empirical evidence for his own use (Reference Wilson1983, 172–173). According to Dawkins (Reference Dawkins1982), although the debate “is really an argument about what we ought to mean [sic] when we talk about a unit of natural selection” (Reference Dawkins1982, 82), in that same page multilevel selection is false and incorrect [sic], not merely a different perspective (Reference Dawkins1982. 82). In short, these well-known researchers do not live up to their claim that the debate is not about facts, hence not about true or false statement but about the utility or fruitfulness of interchangeable perspectives.

Griesemer (Reference Griesemer1998, Reference Griesemer2000a, Reference Griesemer2000b) clarified the importance of empirical information in the debates over group selection: in order to theoretically consider the existence or absence of a group selection process, in each specific case one must first empirically test if the population is, in fact, divided into clusters and if these social entities actually show heritable variance in fitness. One may then contemplate on the different advantages for different representations for this data. However, I argue that extreme images of ‘group’ tend to undermine this much needed focus on population structure and population dynamics.

Moreover, precisely these empirical details may decide between certain processes of individual selection versus group selection. Brandon (Reference Brandon1990, 98–116) reviews the empirical criteria: when there is no variance in group fitness,Footnote ⁴ or when the variance in group fitness does not depend on group structure (for example, group differential reproduction is independent of the relative frequency of altruists in the group, but depends on shelter availability in the physical environment), a process of selection between groups cannot occur. When both individual selection and group selection processes affect an altruistic trait, selection within groups is more effective when variance in the fitness of individuals within each group exceeds variance in the mean fitness between groups or when the variance in group-trait is not heritable.Footnote ⁵

Thus, if different models of individual selection attempt to identify different causal processes, then processes such as reciprocity (Trivers Reference Trivers1971; Nowak and Sigmund Reference Nowak and Sigmund1998), policing (Boyd and Richerson Reference Boyd and Richerson1992), and byproduct mutualism (Dugatkin and Mesterton-Gibbons Reference Dugatkin and Mesterton-Gibbons1996)—can be tested against group selection on a case-by-case basis. In addition, kin selection relies upon different model-assumptions about population structure than intergroup selection (Hamilton Reference Hamilton and Fox1975; Grafen Reference Grafen, Krebs and Davis1984; Wade Reference Wade1980, Reference Wade1985); hence, it may be possible to determine which explanation is most likely by accurately describing the population structure (Lloyd Reference Lloyd1994, 156–159; for a general criticism of pluralism in the group selection debates see Lloyd Reference Lloyd2005).

3. Experimental Tests of Group Selection

This section will reexamine two well-known experiments in the history of the controversy, in order to illustrate some of the common procedures and practices for testing selection between groups. Not many such experiments exist. Dugatkin (Reference Dugatkin1997) reports of nearly one thousand experiments in support of various models of cooperation: reciprocal altruism, kin selection, byproduct mutualism, and, also, group selection. He singles out an experiment performed in Reference Rissing, Pollock, Higgins, Hagen and Smith1989, by Rissing et al. that discusses the specialized foraging behavior of the leaf-cutting ant Acromyrmex versicolor. Dugatkin and others consider their work as “one of the strongest cases of trait-group selected cooperation …” (Dugatkin Reference Dugatkin1997, 161; see also Sober and Wilson Reference Sober and Wilson1998, 115).

The ant A. versicolor lives in communal nests, where all cofounders reproduce. It is not uncommon for a species of the Hymenoptera to establish their nest with several unrelated queens (Kukuk and Sage Reference Kukuk and Sage1994), but it is rare for the cofounders to keep cooperating even after the workers emerge. In this species, cofounders live together nonaggressively before and after workers have emerged. One queen specializes in building the nest, and usually the same individual specializes in foraging outside the nest (Rissing, Pollock, and Higgins Reference Rissing, Pollock and Higgins1996).

The risks in foraging are known (Traniello, Fujita, and Bowen Reference Traniello, Fujita and Bowen1984), so there are clear individual benefits for the nonforager. Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989) have shown that the cofounders were no more closely related than randomly selected queens, thereby excluding kin selection among cofounders as an explanation for the altruistic behavior. They also found that foraging in A. versicolor was independent of physical strength (body size) and of reproductive ability (the number and length of oocytes), i.e., foraging did not recover an otherwise difficult social position of that individual. Since the explored behavior did not seem to improve individual fitness in any way, byproduct mutualism was also disproved by Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989). Foraging behavior was neither reciprocal nor physically forced; therefore, the reciprocal and policing models can also be dismissed.Footnote ⁶

In this species, newly founded colonies are highly aggregated, engaging in reciprocal brood raids where larger colonies tend to prevail. Colony reproduction depends on fungus supply, and fungus gathering improves with experience. Thus, a specializing forager improves the colony reproduction. From a cluster of new colonies only one reproduces fast enough to survive and become a mature colony.

Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989, 422) conclude: “Under these circumstances, a founder may be relatively indifferent as to whether she is the one that suffers the mortality risk associated with foraging, because the colony will either succeed as a unit or (more probably) fail as a unit.”

This is accepted as a clear example of a group selection process, which disposes altruistic behavior to take over the population and to reach a stable evolutionary equilibrium. I would like to challenge this claim. The authors have clearly shown fitness differentiation between groups, in the sense of differential nest extinction, but they have not empirically determined the group as the unit of selection. It has already been mentioned that processes of group selection and individual selection can be empirically differentiated. Let us apply these criteria to the trait of foraging specialization.

Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989) have not correlated variance in group extinction to variance in group trait; hence, they have not positively established group selection. Furthermore, they have not measured variance in fitness within each group. Hence, they have not shown individual selection to have a weaker affect than group selection in this case. Long hours of videotaping these ants revealed that all A. versicolor groups have an altruistic specialist, so there is no variance in this group trait, nonetheless, there is variance in group extinction. The reason for group extinction is not known, although larger groups tend to survive brood raids. In this experiment, only one trait—foraging specialization—that did not correlate with group size, and did not vary between groups, was considered sufficient to indicate group selection.

No group trait that may be related to group survival, such as the amount of time prior to which a specialist begins to forage, was examined; and no individual trait was examined for correlation with individual fitness in A. versicolor, such as survival or number of offspring for forager or nonforager queen. Undoubtedly, selection is intense here, causing large-scale extinctions of both groups and individuals. In order to know at what level selection occurs, the experimental plan must disclose whether variances in fitness and in trait frequency were larger between or within groups and whether the frequency of the trait at the mother group was heritable to its daughter groups. However Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989) did not directly test such group selection pressures.

The term ‘group’ is not used by Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989). These long lasting, well organized ant communes are named ‘associations’ in the paper. An association does not necessitate organized interactions among individuals, and this may explain why Rissing et al. (Reference Rissing, Pollock, Higgins, Hagen and Smith1989) did not test heritable variance in such group level features. Perhaps this extreme aggregative image of the notion of ‘group’ predisposed them—and their readers—to ignore the need for a direct test of the process of group selection.

Rissing et al.'s (Reference Rissing, Pollock, Higgins, Hagen and Smith1989) study fairly represents the widespread practice in this field. However, here and there, one finds an experiment that actually applies the empirical criteria mentioned in the pervious section. Wade in Reference Wade1976 and Reference Wade1977 and his student Craig in Reference Craig1982, compared the effectiveness of individual selection to group selection in several species of flour beetles (Tribolium). Wade examined the trait of group size as one of the factors determining group extinction, and Craig examined the tendency to emigrate as one of factors affected by group size.

Wade (Reference Wade1976) investigated nine generations of Tribolium ccastaneum and Craig (Reference Craig1982) followed 14 generations of Tribolium confusum. In each generation all groups began from the same initial size of 16 individuals, developed independently, and were divided into daughter-groups according to their size after inbreeding. The control treatment represented individual selection, by producing a daughter-group from every group, thus denying any variance in group fitness and hence any chance of selection between groups.

In Wade's experiments, the group selection treatments picked the largest and smallest groups for proliferation, and the size of their daughter-groups was compared to the control treatment. The difference was significant, even impressive.Footnote ⁷ In Craig's experiment, as well, selection for groups with a certain tendency to immigrate has produced similar daughter-groups that deviated significantly from the control treatment. The high mean dispersal remained in the lineage despite a large variance in discarded immigrants within each group, i.e., group selection for immigration is highly effective regardless of the intensity of individual selection within that same group against immigration.

Most evolutionary biologists, for or against group selection, accept that in Wade's experiments the effectiveness of selection between groups exceeds that within groups. Yet, as the next section will elaborate, opponents of group selection alongside its most firm supporters seem to agree that Wade's experiments merely confirm a rare and marginal case.

Despite the fact that such tests of group selection have been available for over twenty years, it is widely accepted that the empirical aspects of this dispute remain problematical. Since the problem is not the lack of empirical criteria (see review by Wade and Griesemer Reference Wade and Griesemer1998), I suggest that it may be worthwhile to look into the very notion of a biological group and examine its role in the present state of the debate.

4. The Notion of ‘Group’ and the Description of Group Selection Tests

Uyenoyama and Feldman (Reference Uyenoyama and Feldman1980, 395) offer an evolutionary definition of ‘group’: “A group is the smallest collection of individuals within a population defined such that genotypic fitness calculated within each group is not a (frequency dependent) function of the composition of any other group.”

However, such an explicit meaning of ‘group’ does not seem to guide research and disputes over group selection. Instead, implicit visual images of a biological group—such as a beehive, a pack of dogs, or a school of fish—often shape the implicit meaning of the notion of ‘group’. Griesemer (Reference Griesemer2000a) clarified how invoking visual images encourages specific modes of abstraction from the phenomena being investigated, and the ‘packaging’ or enframing of these phenomena express a particular theoretical perspective. I found two such underlying images of what constitutes a ‘group’: that of an aggregate and that of the superorganism. An aggregate group is typically perceived as a tentative contextual background for an individual's actions; while a superorganism is read as an emergent cohesive whole, organizing, dominating, and typically suppressing, its individual parts.

In the selection process tested by Wade (Reference Wade1976), termed intergroup selection, one finds both images intertwined. The group's trait and the group's fitness is not a special holistic predicate, but the mean of individual trait and individual fitness within that group. In that sense, the group is an aggregate of individuals. However, these beetle groups reproduce as a coordinated whole, through inbreeding and group budding and not by random individual dispersal to a common pool. Finding the mean of individual heritability in the population is irrelevant to the reproduction process by which this heritability was generated. This manner of reproduction ensures strong group heritability and pictures the group as an organized superorganism. Thus, an accurate description of Wade's experiment requires both the aggregate and the organic whole images, which differ at different aspects of the group's life history.

However, most supporters and opponents of group selection do not hold such a complex, multi-image, sense of ‘group’. Instead, they typically adhere to an extreme and monolithic image to characterize almost all aspects of ‘group selection’ and ‘group adaptation’. In other words, either they identify ‘group’ exclusively with a cohesive superorganism, or exclusively with a loose aggregate. Yet what exactly does it mean for a group to hold aggregative interactions among its parts? Wimsatt (Reference Wimsatt1997) clarified this question, and specified that aggregativity requires invariance of the system's trait to (a) rearrangement of its parts; (b) addition or subtraction of parts; (c) decomposition and reaggregation of parts; and (d) no cooperative or inhibitory interactions among its parts. Each failure of invariance indicates a specific type of emergent character of the system as a whole. Note that the monolithic notion of a group as “no more” than an aggregate tends to fade away once one specifies what type of aggregate she or he are thinking about. For example, the interactions among Wade's flour beetles are aggregative in relation to Condition (a) and superorganismic in relation to Conditions (b), (c) and (d); while the foraging ants in Rissing et al. Reference Rissing, Pollock, Higgins, Hagen and Smith1989 and Rissing, Pollock, and Higgins Reference Rissing, Pollock and Higgins1996 indicate aggregative groups in relation to Conditions (a) and (c), and organic wholes with regard to Conditions (b) and (d).

Wimsatt's Reference Wimsatt1997 paper is widely known, yet the discourse over group selection more or less ignores his heuristics just as it avoids Wade's experiments. In both cases a monolithic notion of the group—a ‘pure’ aggregate or superorganism—is preferred over a diversified notion. For example, the image of a superorganism, although needed for an accurate description of Wade Reference Wade1976, is not a part of Sober and Wilson's (Reference Sober and Wilson1998) support of group selection in general and Wade's experiment in particular (1998, 105–106). Their group's unity is achieved by the variance of its mean trait from the mean trait in other groups, not by its cohesive spatial packaging, its internal organized structure or its reproduction dynamics. Heritability of the mean trait in the group, although required by Sober and Wilson (Reference Sober and Wilson1998, 107–113), does not require any developmental links between the mother and daughter groups as a whole. Propagation or extinction of the group as a whole, required by intergroup selection (Wright Reference Wright1945; Griesemer and Wade Reference Griesemer and Wade2000), is irrelevant for Sober and Wilson's trait-group selection (Reference Sober and Wilson1998, 59–61). Although in the past they have named their groups ‘superorganisms’ (Wilson and Sober Reference Wilson and Sober1989; Wilson and Sober Reference Wilson and Sober1994), it seems that they have used this term in a unusual manner, which in most aspects fits a loose and temporary assembly.

Their all-embracing reading of ‘group’ as an aggregate brings Sober and Wilson (Reference Sober and Wilson1998) to include the more superorganism type of model tested by Wade (Reference Wade1976) within their trait-group model, where individuals do not inbreed but randomly reproduce in the population. In fact, Sober and Wilson (Reference Sober and Wilson1998, 61) describe the former as a doubtful and rare instance of group selection. Thus, Wade's experiment, although happily accepted by these notable supporters of group selection, cannot confirm their general claim for the importance of group selection. Therefore, Sober and Wilson do not consider such experiments as the way to resolve, or at least progress, in the resolution of the debate.

Not surprisingly, Wade's experiments are equally undermined by opponents of group selection theory. Current adversaries of group selection typically identify ‘group’ with an extreme superorganismic image. Within a conceptual schema that accepts only individuals or aggregates of individuals, they demand group selection to withstand tests that apply to extreme superorganisms, which, of course, rarely—if ever—happen. Some of these leading researchers define ‘group’ in a manner that more or less prevents empirical resolution. Since Wilson and Sober (Reference Wilson and Sober1994) have already argued this point with respect to Williams (Reference Williams1966), I will give an example from Okasha's (Reference Okasha2001) study.

For Wade (Reference Wade1976) and Sober and Wilson (Reference Sober and Wilson1998) the notion of ‘group fitness’ does not mean more offspring groups but more offspring individuals, and ‘group trait’ is nothing more than the average of individual traits within that group. In that sense, the group is an aggregate of individuals. Okasha (Reference Okasha2001), therefore, does not distinguish Wade's intergroup selection from Sober and Wilson's trait-group selection and from individual selection, although only the former requires whole group reproduction. Okasha's notion of ‘group selection’ requires an extreme superorganism in all aspects of group life, which leaves him no choice but to consider Wade's (Reference Wade1976) experiment as an instance of individual selection.

According to Volman (personal communication), we should recognize that a narrow sense of ‘group’ is no more than a theoretical limit. Although it is important to keep the notions of group and individual as opposite poles of a continuum, it is very difficult—if not impossible—to find an entity that actually fits a superorganism or an aggregate in all aspects. The social entities actually studied in nature or in the laboratory are superorganisms in some features and aggregates in others. Although most researchers will happily agree to the above statement, his or hers implicit reading of ‘group’ will probably remain as monolithic as before. Elsewhere (Shavit Reference Shavit2004) I have elaborated on the sociopolitical values, which partly explain the persistence of such monolithic perceptions of ‘group’, while here I illustrated some of the empirical shortcomings resulting from these monolithic images.

To conclude, the basic problem is that both supporters and opponents of group selection use an extreme image of ‘group’ in order to describe an intermediate entity. A limit-sense of ‘group’ does not fit neatly with actual groups, so it tends to disregard those phenomena related to actual population structure and dynamics. As a result, experimental studies are described without notice to empirical distinctions regarding group reproduction or group dispersal. Such framing of the phenomena allows the same “results” to support both group selection theory according to Sober and Wilson and individual selection theory according to Okasha. Instead, a notion of ‘group’, which accepts a multiplicity of group images, may direct researchers to distinguish between the different causal processes involved in the different aspects of social life. As I have shown elsewhere, noting these empirical differences improves our understanding of the evolutionary transition to sociality (Shavit Reference Shavit2001). All in all, adopting a wide-ranging reading for the notion of the group may improve the poor empirical state of the debates over group selection.