I am grateful to all the commentators and humbled by the variety of responses and the care that went into them. Thankfully, the majority of commentators are on board with the basic thesis of the target article: that group-level traits are distinct from individual-level traits and constitute different units of analysis in evolutionary thinking. As such, a large part of this response will entail clarifying and extending the idea of group-level traits in response to various suggestions and concerns brought up by the commentators.
Although most commentators supported the thesis of the article, there were also those who were not so readily persuaded that any new perspectives or methods are necessary to understand the evolution of human social complexity. Some of these commentaries expressed concerns over ideas that were at the foundation of my argument. I believe it is crucial for any discussion to have a firm foundation. Therefore, I will begin with a response to some of the most critical commentaries. First, I will justify an evolutionary perspective that includes group organization and nongenetic inheritance. Next, I will discuss the concept of emergence and how I use it to elucidate group-level properties and behaviors.
Following that, I will transition to an exploration of a number of important ideas and concerns brought up by some of my more supportive commentators, as well as by those who accept the basic thesis but were critical of some of the details. This will include a discussion of differences among types of groups; the psychology of group-level traits; the uses and limitations of an institutional perspective; the link between transmission, adaptation, and selection in the evolution of group-level traits; current and future methodologies for the study of social behavior and evolution; and the various fields in the biological and social sciences that may benefit from a group-level traits perspective.
R1. The false allure of “standard evolutionary theory”
The most severe critique of the target article comes from Scott-Phillips & Dickins, who claim there is no need whatsoever for new methods for understanding group-level traits. Their attack is not, in fact, leveled explicitly at the group-level traits idea but is a larger attack on recent advances in evolutionary theory that incorporate methods of nongenetic inheritance; these include cultural evolution, epigenetic inheritance, and niche construction. They argue that these approaches are misguided and represent disruptive deviations from the path to an enlightened understanding of evolution, one that focuses on those methods that they view as constituting “standard evolutionary theory.” As this is a critique on the overarching framework essential to the approach presented in my target article, it is important to address their comments first.
Scott-Phillips & Dickins's critique stems from two beliefs: (1) that all heritable traits, including behaviors, are derived from genes (e.g., Dickins & Rahman Reference Dickins and Rahman2012), and (2) that culture is a strictly proximate mechanism. Neither belief withstands closer examination. First, a gene-only view of evolution ignores the role of self-organizing forces in shaping traits (Kauffman Reference Kauffman1993) and the fact that nongenetic phenotypic factors may be both heritable and subject to natural selection, including epigenetic markers (Jablonka & Lamb Reference Jablonka and Lamb2005), environmental niches (Odling-Smee et al. Reference Odling-Smee, Laland and Feldman2003), and cultural institutions (Richerson & Boyd Reference Richerson and Boyd2005). Second, treating culture as a proximate mechanism ignores both the coevolutionary interaction between culture and genes and the fact that culture itself evolves in a Darwinian process (Mesoudi Reference Mesoudi2011). Scott-Phillips & Dickins also seem confused about basic principles of multilevel selection theory. For example, they argue against the use of multilevel selection (MLS) theory by saying “there is no formal justification to consider the cultural group as a unit of selection.” But MLS theory does not, in fact, require that the group is a unit of selection, only that the circumstance of belonging to a group influences individual fitness (Okasha Reference Okasha2006). Finally, Scott-Phillips & Dickins take me to task for ignoring the work of cultural epidemiologists such as Dan Sperber. I agree that cultural epidemiological models are valuable, but their direct relevance to a conceptual discussion of group-level traits is minimal. Further, the cultural evolutionary perspective promoted by Boyd, Richerson, and colleagues (Boyd & Richerson Reference Boyd and Richerson1985; Richerson & Boyd Reference Richerson and Boyd2005; Mesoudi Reference Mesoudi2011) is completely coherent with cultural epidemiological models (Henrich et al. Reference Henrich, Boyd and Richerson2008), and because these models also treat cultural variants as individual-level traits, the discussion in the target article applies to them as well.
Evolutionary theory is a relatively young field, one that continues to develop as researchers identify new principles and incorporate more of the real world's daunting complexity into their models. It is, of course, important to study established methods and frameworks, and to work within them when appropriate to avoid reinventing the wheel. But it is not pragmatic to cling to a restricted body of theory deemed “standard” while ignoring important advances that will help us to better understand biological evolution, as well the processes of change in other systems, such as human culture. The issue here seems to be largely a problem of perspective. Sometimes adopting different perspectives, metaphors, or frameworks helps us to visualize a system, and related problems, in different ways. As Kauffman (Reference Kauffman, Buck and Cohen1971) eloquently elucidates, there is often not a single “correct” model for conceptualizing a complex system. Rather, the model one adopts – and the subsequent articulation of the parts of the system – depends on the particular questions one asks. The gene-centric perspective of “standard evolutionary theory” is not wrong, but insisting on it as the only valid model limits one's vision, and thereby limits the questions one can ask, let alone answer.
R2. On emergence
Emergence is a tricky concept, and one that has long been the subject of debate among complexity theorists and philosophers of science. Okasha notes that the emergent/aggregate distinction is often imprecise. I agree, which is why in the target article I note that the degree of emergence I discuss must often be determined heuristically rather than algorithmically, and I suggest using Wimsatt's (Reference Wimsatt1997; Reference Wimsatt2006) heuristics for aggregativity to do so. A system that fails to exhibit all four of Wimsatt's aggregate properties exhibits the type of emergence I wish to discuss. Schank makes the useful suggestion of distinguishing between “strong” group-level traits that meet this criterion and “weak” group-level traits (such as flocking) that are aggregate to some greater degree. For convenience I will use the term “group-level traits” to indicate the strong subtype throughout this response.
Santana & Weisberg misinterpret me as claiming that strong group-level traits are necessarily irreducible. This is understandable, as that is indeed what some authors have meant by invoking the term “emergence.” Santana & Weisberg propose that group-level phenomena are always reducible, and can be captured by agent-based models in which only the actors and their relationships are represented. I am quite enthusiastic about the potential for agent-based modeling to address the problems I discuss in the target article, but I disagree that explicit representation of group-level phenomena is unnecessary. To untangle these issues, it will be helpful for me to clarify my position on emergence.
Some authors have proposed that “emergence” refers to a category of phenomena that can be rigorously defined, and therefore is a natural kind in the sense that it exists in the universe independent of an observer (e.g., Bedau Reference Bedau2008). It is not my intention to enter into a discussion of the metaphysical realness of emergent phenomena. A phenomenon is emergent, in the sense I am using the term, when the language of more primitive concepts and relationships fails to capture the phenomenon at hand. Whether or not the emergent phenomenon is a natural kind, there are still real patterns at higher levels of organization to be recognized by an observer, patterns that may lead to improved descriptive parsimony and/or enhanced predictive power over lower-level descriptions (Dennett Reference Dennett1991). For example, the precise mechanical behavior of a honeybee colony is of course reducible to all the individual bees, their environments, and their relationships. However, if we are to speak of the colony solving problems, such as when a swarm successfully chooses the optimal site for its nest through a democratic process of exploration and presentation (Seeley Reference Seeley2010), then we can properly speak of emergent behavior, because the individual bees do not solve the problem; only the colony as a whole does that. Similarly, the patterns of activity in the human brain are reducible to the individual neurons, glial cells, and their physical environments, but conscious attention and memory are emergent, because these terms describe the behavior of a higher-order entity beyond that of a collection of individual neurons (Hofstadter Reference Hofstadter2007). This is also, perhaps, why efforts to reduce all of the natural and social sciences to physics are so unsatisfactory (Fodor Reference Fodor1974). Economics does not reduce to physics, even though all the entities involved are surely rooted in physical phenomena. This is because the entities of economics only make sense (to the observers describing them) in terms of relationships between actors, institutions, markets, and capital. These entities are, in turn, psychological and sociological in nature, and are not appropriately described by invoking a (vast) set of physical relationships for the simple reason that they only have meaning as higher-order constructs.
Strong group-level traits are emergent in this sense. Admittedly, things may get semantically hairy, because we can sometimes speak of the behavior of a collective even when that behavior is highly flexible to the organization of the individual constituents. I agree with Ibbotson when he cautions that one must make sure to rule out individual-level explanations for social phenomena before looking to the group level. When a group-level analysis is appropriate, it is valuable to adopt Wimsatt's heuristics and focus on strong group-level traits.
Understanding groups as entities capable of emergent properties is also important because a group's organizational structure and social resonance can exert influence back onto its constituents, a phenomenon known as downward causation (Campbell Reference Campbell, Ayala and Dobzhansky1974a). Higher-level entities (groups) affect lower-level entities (individuals) through proximate psychological and sociological mechanisms, as well as via selective forces that promote or disrupt the individual-level features that facilitate particular group structures. When the mechanisms of downward causation are proximate, group-level traits may indeed be irreducible, because the group's constituent actors are influenced by higher-order constructs. For example, a team of athletes, hunters, or soldiers may be particularly effective as a result of synergistic interactions between teammates. This effectiveness creates a reputation that influences the behaviors of the team's constituents, as well as the perceptions and behaviors of the individuals and groups with whom the team interacts. In such cases it may still be possible to capture group-level phenomena with an agent-based model, as Santana & Weisberg suggest, but only if group-level traits are explicitly expressed in the model as potential causal factors (see Schank Reference Schank2001). This relates to my discussion in the target article concerning interactional complexity. When it comes to influences between levels of organization, there is not one causal arrow but a quiverful, pointing every which way.
R3. On groups and their constituents
The question “what is a group?” is complicated by the fact that there may be many different kinds of groups, each with different organizations, functions, and ontogenies. Further, the human psychology behind complex group organization is far from simple. In this section, I address the concerns related to group type, the psychology of group-level traits, and the cultural milieu that produces group-level organization.
R3.1. The varieties of group experience
In the target article, I concentrated on describing the group-level trait as something categorically and qualitatively distinct from an individual-level trait. To do this in the broadest possible sense, I adapted Wilson's (Reference Wilson1975) definition of the trait group, which led to the inclusion of everything from mother-infant relationships to the interactions of a vast army under the same categorical heading. Highlighting the group-level trait as an overarching category is important. However, I also agree with Fuentes that dyadic and large-scale interactions should not be assumed to be functionally equivalent entities in an evolutionary model. For more precise analysis, a better taxonomy of groups – and group-level traits – may be useful.
Gerkey & Cronk correctly point out that there are many types of groups, and they suggest several ways of distinguishing them. I am most drawn to their distinction between “corporate” and “categorical” groups, as the former type seems most likely to exhibit emergent (strong) group-level traits, while the latter is the type most commonly referred to in cMLS models. In general, Gerkey & Cronk's designations of group types, which are based on a study of real human groups and social networks, strike me as more useful than the more traditional distinction in MLS theory between interdemic and intrademic group selection (Boyd & Richerson Reference Boyd and Richerson1985; Wade Reference Wade1978). It is possible that there may even be systematic differences in the types of organizational structures – and hence the types of group-level traits – that are prevalent or even possible in each of Gerkey & Cronk's suggested group types, and that cultural evolution may operate differently on each.
Caporael & Garvey take a slightly different approach, proposing a model of “core configurations,” in which different “core” group sizes point to different types of organizational structure and are associated with different types of tasks. In this case, group type is inexorably linked with group size. Caporael & Garvey suggest that group size at each of their four proposed core configurations (dyad, task group, deme, macrodeme) corresponds to different types of group tasks and therefore different key group-level traits. There appears to be a good deal of validity to this approach. What is most important about this framework is that it links group-level traits with both individual psychology and the evolution at the population level of traits related to social coordination. More will be said on both those topics in subsequent sections of this response.
R3.2. The psychology of group-level traits
Human groups exhibit and transmit group-level traits in a manner unique in the animal kingdom. This fact demands an inquiry into the unique features of human psychology that make such an organizational feat possible. I therefore agree with Muthukrishna & Schaller that understanding the proximate processes involved in group coordination and the emergence of group-level traits is essential. They are right to highlight the small but steadily growing trend in social psychology to embrace dynamical systems theory (Vallacher & Nowak Reference Vallacher, Nowak, Kruglanski and Higgins2007), which seems like as good a framework as any with which to study the psychology of group-level traits. In this section I will briefly discuss several avenues of psychological research that I (along with several commentators) believe are particularly pertinent to the study of group-level traits.
As noted by Caporael & Garvey, humans may possess different mental systems for dealing with different types of groups, organized by size and functionality. However, it is also important not to be too quick to assume the presence of specialized mental functions. The environment, including the social environment, shapes mental processes and creates affordances that constrain choice and guide behavior (Gibson Reference Gibson1979; Smaldino & Richerson Reference Smaldino and Richerson2012). Even reliably emerging features of social interactions may be the result of affordances from underlying social structure rather than reflections of explicit psychological hardwiring.
Theiner & Sutton focus on group problem-solving and distributed cognition, in which individuals collaborate to solve problems in complex ways. I agree with them that it is vital not to compare too directly group performance on a task with individual performance, as the goals and payoffs may not be the same. I must admit that I found their presentation of distributed cognition somewhat vague. A potentially clarifying idea was recently presented by Gallotti and Frith (Reference Gallotti and Frith2013), who propose that collaborating minds engage in an irreducibly collective mode called the “we-mode.” An example is that the presence of a known collaborator introduces new affordances for the decision maker. Imagine that an important object is out of reach for me, but not for my teammate or social partner. I can make collective plans that involve the object, because even though I cannot reach it, we can. The “we-mode” seems crucial for many (though not all) group-level traits.
A number of previously identified psychological features may have importance for the study of group-level traits, and vice versa. Two mentioned by O'Gorman are transactive memory (Liang et al. Reference Liang, Moreland and Argote1995) and the immense human propensity for group identity. With regard to group identity, Roccas and Brewer (Reference Roccas and Brewer2002) have noted that individuals often identify with several groups simultaneously. These group identities can be hierarchically nested or practically orthogonal, and circumstance can evoke stronger associations with one group over another. This ability to identify with multiple groups is likely important for the facilitation of group-level traits, as individuals can participate in a number of distinct group endeavors and feel strong ties to each organizational unit. Abrams posits that not only do social roles facilitate group-level organization, but also that “different social roles may encourage ‘cohesive’ cognitive subnetworks, reinforcing particular patterns of thought in individuals who fill similar social roles.” This suggests that the environmental effect of participating in a group-level endeavor may lead to a number of cognitive and behavioral similarities among individuals occupying similar social roles, above and beyond those necessary for performing those roles. A related phenomenon is that persistent influence (e.g., via physical proximity) can evoke synchrony in systems that exhibit periodicity, such as neurons and other oscillators (Strogatz & Stewart Reference Strogatz and Stewart1993). It seems possible that more complex psychological functions could work similarly, at least on some level. I find this idea quite intriguing.
A psychology that facilitates group cohesion is one prone to respond to and adopt social norms, and there is good evidence that humans possess such a psychology (Chudek & Henrich Reference Chudek and Henrich2011; O'Gorman et al. Reference O'Gorman, Wilson and Miller2008b). Nevertheless, Schank rightly points out that imitation is not a sufficient psychological mechanism for learning, because imitating another's practices also involves the selection of which practices are (partially) imitated. This is related to the more general problems involved in the psychology of option generation, reviewed in Smaldino & Richerson (Reference Smaldino and Richerson2012).
MacDonald accuses me of having an “inadequate psychology of groups.” In the target article, I chose to focus on defining the concept of the group-level trait at the organizational level and discussing its role in cultural evolution. I completely agree that understanding the psychology of group-level traits is crucial, and I quite explicitly discuss how human psychology has been shaped to deal with group living, including mechanisms that fall under what MacDonald describes as “explicit processing” (e.g., leadership, teaching). For example, I discuss the importance of pedagogy for the development of social roles, as well as Durham's (Reference Durham1991) concept of TRIMs, which are the “ecological, psychological, linguistic, and cultural barriers to the blending of cultures.” I do agree with MacDonald, however, that my discussion of human psychology is somewhat cursory. I felt that there was only so much material I could cover without overly tangling the narrative thread.
MacDonald also misrepresents my position on behavioral genetics when he says that I “exclude by fiat the behavior genetic literature.” My point about repeated assembly is to contrast it with views of behavioral genetics that imply that genetic influences to behavior can be considered additive or otherwise simplistic. Genes and environments interact through a process of continuous feedback. I completely agree that genetically influenced individual differences, such as differences in personality, should be related to theories of the emergence of division of labor. Recent advances by evolutionary researchers on the emergence and stability of individual differences (e.g., Johnstone & Manica Reference Johnstone and Manica2011; McNamara & Leimar Reference McNamara and Leimar2010; Wolf & McNamara Reference Wolf and McNamara2013) will be an exciting avenue to incorporate into the study of group-level traits.
R3.3. Culture is more than individual differences
Culture is not simply a statistical description of individual differences, contra Kim, Jeong, & Park (Kim et al.). They criticize my position that culture influences cognition and claim that I characterize culture as an “autonomous, external variable that independently affects cognitive processing.” I want to be clear: I think no such thing. Rather, I think that what the cultural psychology literature shows quite clearly is (1) how much our thinking, perception, and behavior are influenced by learning, especially the learning that occurs during development, and (2) how cultural groups provide a common ecological and social environment that leads to highly correlated patterns of cognition among members of the same culture. Kim et al. claim that “culture is an aggregate of individual differences in psychological variables within and between groups.” This claim is not supported by the overwhelming evidence that individuals tend to adopt the perceptual and behavioral norms of the culture of their birth, independent of their genetics (Boyd et al. Reference Boyd, Richerson and Henrich2011; Cohen Reference Cohen2001; Harris Reference Harris2012; Hespos & Spelke Reference Hespos and Spelke2004; Kelly et al. Reference Kelly, Quinn, Slater, Lee, Ge and Pascalis2007; Kinzler et al. Reference Kinzler, Shutts, DeJesus and Spelke2009; Kuhl et al. Reference Kuhl, Williams, Lacerda, Stevens and Lindblom1992). This is not to say that there are not regional or cultural regularities in genes, or that individuals do not differ in temperament, only that cultural (e.g., socially learned) factors are also tremendously important. Moreover, cultures are defined by much more than the aggregate psychologies of their members. Cultures also involve traditions of learning and customs, of institutions and infrastructure. The fact that bride theft is prevalent in Kyrgyzstan and not in the United States (Werner Reference Werner2009) is the result of cultural history, not innate psychological differences. Similarly, the culture of honor still prevalent in the American South and the biological responses that support it (Nisbett & Cohen Reference Nisbett and Cohen1996) cannot be explained by genetic differences in psychological temperaments, but only by examining cultural traditions. Finally, to address group-level traits, I completely agree that individual differences in innate psychology will influence the opportunities and affordances for an individual's roles in group-level traits. However, it is ludicrous to explain the sailing of a ship or the rituals of a religion solely through an appeal to individual psychological differences. The emergence of such traits depends also on the cultural infrastructure that guides learning and opportunity.
R4. The importance and limitations of institutional analysis
In the target article, I state that group-level traits are related, but not equivalent, to institutions. I agree with Zefferman & Richerson that an institutional perspective is important, and it is possible that I downplayed that importance. Institutions – the formal and informal rules that govern social behavior – may often be the source of group-level traits. Individuals in a group may copy another group's institutions in order to absorb a suite of norms and behaviors, as in the example of the Japanese emulation of British naval institutions. Zefferman & Richerson suggest modeling institutional evolution as a type of between-group “equilibrium selection,” as in Boyd and Richerson (Reference Boyd and Richerson1990). Although this is a useful model for some systems, “equilibrium selection” is an interdemic process that does not account for the kinds of group-level traits that compete within small groups, in which many different organizational patterns may coexist. Perhaps more importantly, some group-level traits may be the product of many interacting institutions. For example, consider the hypothetical Roman Legion I discussed in the main text. This will also help to clarify a question posed by Panchanathan, Mathew, & Perreault (Panchanathan et al.), who wondered “why a Roman Legion represents a group-level trait, while an insect colony does not.” The Legion itself is not a trait, of course. Rather, the organizational properties of the Legion and its constituents lead it to possess a number of traits. An example of a rather superficial trait might be the Legion's skill in battle that leads to them being almost undefeatable on an open plain but much more vulnerable in the close quarters of a dense forest. This trait does not stem from one institution, but from many institutions interacting with the culturally influenced psychologies of the Legion's constituents. Other example traits might include the economic resilience of a firm or the evocative tonality of a musical group.
Because not all group-level traits reduce to institutions, I feel I must reject Waring & Goff's suggestion to focus on the social organization itself and not on the phenotypic traits produced by that organization. In practice, the organization is the proximate cause of a given group-level trait and will likely be vital to its understanding. However, an explanatory trait may require a different description than that of the organization that gives rise to it. In terms of evolutionary dynamics, fitness is a property of traits, not of individual organisms or groups, and it is not explanatory to speak of the fitness of an organizational structure without appealing to the subsequent trait produced by it. The same is true of more direct influences on behavior. As an example, consider individual-level traits such as “an aggressive personality” or “being particularly gassy.” Explanations of their influences on behavior are most meaningfully detailed by invoking the traits themselves, and not through an appeal to neurophysiology or digestive system malfunction.
MacDonald claims that culture is shaped primarily by leaders in a top-down fashion, and is therefore not emergent. In this sense, he argues for the supremacy of explicit institutions enforced through leadership. I do not dispute the importance of leaders, but it seems reductionistic to an absurd degree to suggest that the coordination involved in group activities from sailing a ship to hunting a stag to the intricate dance rituals of numerous religious groups all boil down to leadership. Many processes of social coordination are not dictated by a leader, but emerge organically both through communication between individuals and from the constraints to affordances provided by group structure.
Read agrees that systems of organization that incorporate differentiated roles are important, but claims that “these systems of organization need not be emergent, but are often cultural constructions, such as the culturally formed kinship systems that provide structure and organization in human societies, especially in the small-scale societies that were the evolutionary precursors of large-scale human societies.” I think that what Read calls “cultural idea systems” are very similar to institutions. However, while institutions are generally classified as sets of rules, “idea systems” will also entail deeply entrenched cultural factors such as language, mythologies, and narratives that structure interactions in more subtle ways and are maintained through group-level organization. Cultural kinship relations represent a fascinating subset of group-level traits, and although I agree with Read that these idea systems are of enormous importance to the evolution of human social complexity, I do not agree that they are not emergent. Culture does not exist as an external, top-down force. Rather, culture is repeatedly assembled and emerges from the complex interplay between human psychology, social organization, and history.
In my target article, I focused on the institutional role of religion in enforcing and transmitting suites of social norms. I believe that it is often useful to regard religion as an institution (or set of institutions). Nevertheless, I am also in agreement with Sosis & Kiper's suggestion that a religion is much more than its institutions and is more fully described as a complex system. This is especially true when considering the evolution of religion within a population – not the fast transmission of a conversion or a conquest, but the slow change that occurs as beliefs, rituals, and institutions adapt to internal selection pressures. As Sosis & Kiper point out, religions do not simply exercise top-down control over norms, but rather permeate all aspects of social life, coevolving with the very norms they prescribe and proscribe. The complex adaptive systems perspective advocated by Sosis & Kiper is not only useful, but is probably necessary for gaining a better understanding of the complicated dynamics of group-level traits in terms of their function, their proximate causes and effects, and their evolution.
R5. Transmission, selection, and adaptation
The major idea proposed in the target article was that group-level traits are an often-unconsidered factor in human evolution. Group-level traits influence the evolutionary trajectory of individual phenotypes and also evolve in their own right. The time frame of a group-level trait does not map one-to-one with the life cycle of an individual human, which creates a problem for evolutionary modeling of group-level organization. Several commentators raised important issues with the transmission, selection, and adaptation of group-level traits. In this section, I will attempt to address these concerns.
R5.1. How are group-level traits transmitted?
The transmission of social roles – and therefore the complex social structure that facilitates many group-level traits – is repeatedly assembled in a process of feedback and structuring. I was grateful to see that several commentators explicitly agreed with me on this point (Caporael & Garvey; Waring & Goff; Schank). I also agree with Read that an important challenge is to understand the phylogeny of this ability that, while perhaps not completely unique in the animal kingdom (O'Gorman; Wilson), is certainly hypertrophied in humans.
The transmission of group-level traits is a complex issue, and my preliminary discussion admittedly only scratches the surface. Davis & Margolis believe that my characterization of group-level traits is problematic, because “the basic aim of evolutionary theories based on population modeling” – including cMLS approaches – “is to explain frequencies in populations. Yet it makes no sense to ask, for a given group, about the frequencies of its group-level properties” (paras. 2 and 3). However, this is not a problem for my presentation, but rather a serious problem for evolutionary theorists who acknowledge the existence and importance of group-level traits, as Davis & Margolis appear to do. Classic population models are indeed based on frequencies of traits; these methods are likely insufficient to explain the evolution of group-level traits. Without even getting into emergent group behavior, simple frequency-based models have already been shown to fail when complex social or spatial organization is paramount (Goodnight et al. Reference Goodnight, Rauch, Sayama, de Aguiar, Baranger and Bar-Yam2008; Smaldino et al. Reference Smaldino, Schank and McElreath2013b). Emergent group-level traits may require new methods, as I will discuss further below.
Davis & Margolis are also skeptical that transmission of group-level traits requires group-level explanations. If a singer wants to form a band that sounds like the Beatles, they suggest, he can separately seek out a drummer who plays like Ringo and a bassist who plays like Paul. Although such a strategy could perhaps yield a serviceable band, bass/drummer synchrony in rock bands is paramount to the sound of the music. The bassist must be able to play “in the pocket” of the drummer's beat. Rock bands that sound good often do so because the styles of the constituent musicians click, which is why “supergroups” made of undeniably talented musicians from different successful bands often disappoint. Moreover, sounding like the Beatles is very different from being influential like the Beatles, or having audiences like the Beatles. These traits likely require group-level explanations.
The recipient of individual-level knowledge – information held in individual brains – must be an individual. However, Davis & Margolis err in claiming that social transmission must occur at the level of individuals. Group-level organization can emerge through processes involving the interplay of individual decisions, social affordances, and repeated assembly. Even when an organizational plan is contained within the mind of one individual, it may be the case that other participants do not possess that knowledge, and that it is only a small piece of knowledge related to skill or social roles that is transmitted to an individual. Moreover, the transmission of the social infrastructure for group behaviors – for example, religious rituals, multiperson sailing voyages, coordinated cooking – may be akin to an n-person Stag Hunt game (Pacheco et al. Reference Pacheco, Santos, Souza and Skyrms2009) in that a critical threshold of individuals must learn their correct social roles in order for the trait to persist.
R5.2. On the units of selection
The phrase “unit of selection” is a loaded one, and one I half regret using in the target article because I think it distracts from the larger points made therein. Santana & Weisberg take issue with my characterization of group-level traits as a unit of selection. I agree that a group-level trait is neither an interactor nor a replicator in the sense described by Hull (Reference Hull1980). However, neither are individual-level traits (such as eye color), since the traits themselves do not replicate. Yet traits are what selection acts upon. Fitness is a property of traits, not organisms or groups (Sober Reference Sober1984). Just as it makes sense to speak of organisms with certain traits being selected over other organisms with other traits, so it makes sense to speak of patterns of group organization being selected for. Wilson makes this point quite clearly in his commentary.
R5.3. Transmission, selection, and adaptation are entangled
Panchanathan et al. point out that group-level traits can arise from individuals maximizing utility, and not necessarily from group selection. I agree that it is necessary to separate selection on group-level traits from “group selection.” Not all group-level traits require the assumption of between-group competition. Humans evolved in a social milieu (Caporael & Garvey), and group-level traits should be able to spread through both individual-level and group-level selection. However, it may not be as easy as Panchanathan et al. imply to disentangle selection from adaptation when it comes to group-level traits. This is because trying to understand the evolution of group-level traits can lead to what Schank calls “selection-transmission thickets.” Schank writes:
Mechanisms for selecting, generating variation, and transmitting cultural traits are often intertwined. For example, imitating the practices of one group also involves the selection of which practices are partially imitated. […] To this entanglement of mechanisms of selection, variation, and transmission, it must be added that as cultures evolve and become increasingly complex, new vicarious selection, variation, and transmission mechanisms (VSVTMs) evolve and appear at new levels in culture.
Humans have long lives and a rich capacity for cultural (social) learning, as well as for complex social behavior (strong group-level traits), and as such the distinction between selection, variation, and transmission may become blurred. Additive or multiplicative models may therefore be of limited use when trying to understand the evolution of group-level traits.
This is also why the methods championed by Chudek & Henrich in their commentary are insufficient. I quite agree with them that “making sense of this complex emergent domain” is far from simple. If it were simple, a formal model in a well-established methodology would have been sufficient instead of a lengthy target article to simply introduce the problem. I also agree with Chudek & Henrich that the culture-gene coevolutionary approaches pioneered by individuals such as Boyd, Richerson, and Henrich represent some of the best attempts so far to understand the dynamics of the evolution of human social complexity. I am not in the least looking to disparage any of that work, and I agree that “starting at the start” is a key approach. However, I do not believe that the options are limited to “starting at the start” or “starting at the end,” as Chudek & Henrich imply. Indeed, studying the evolutionary dynamics surrounding group-level traits represents starting somewhere in the middle, which of course is always the most difficult part to explain in any complex story. A three-minute pop song can capture the start of a romance or its end, but it takes a novel to detail the relationship's messy middle. Similarly, it will take more complex methodologies to capture the evolutionary dynamics of group-level traits. Such approaches will add to, not subtract from, our understanding of cultural evolution.
R6. Models and methods
The evolutionary dynamics of group-level traits are still poorly understood, and making headway in that understanding represents a significant challenge. Many existing methods provide important frameworks, and they should not be ignored. Additionally, new methods and models will assuredly be needed to formalize and better conceptualize the role of group level traits in human social evolution.
I agree with Doebeli & Simon that formal models are crucial to understanding complex systems and for conceptualizing problems within them. Models can distill a system down to manageable parts and relationships, and allow us to see clearly the presence or absence of possibility in a world obscured by the fog of complexity. Doebeli & Simon propose that their modeling framework (Simon et al. Reference Simon, Fletcher and Doebeli2013) is appropriate for the evolutionary study of group-level traits. I agree that their framework has a great deal of value. Although I had not previously examined it in detail, it is strikingly convergent with some of my own recent attempts to model social evolution at multiple levels of selection (Makowsky & Smaldino Reference Makowsky and Smaldino2014; Smaldino et al. Reference Smaldino, Newson, Schank and Richerson2013a). Nevertheless, things can be become quite complicated when groups are ephemeral, or when individuals can belong simultaneously to multiple groups (Lubell Reference Lubell2013). As Doebeli & Simon acknowledge, their framework would have to be adapted for dealing with a number of aspects of human culture. These include institutions and behaviors such as marriage, property rights, resource inheritance, alloparenting, social enforcement, and division of labor.
Although a formal mathematical framework provides baseline cases and allows for rigorous analysis, the immense complexity of human social interaction may often be mathematically intractable. Computational techniques such as agent-based modeling may instead be more fruitful (Santana & Weisberg; Taylor & Bryson). Individuals can be represented explicitly, and therefore the model can incorporate essential heterogeneity related to sex, age, genotype, and group identity (Epstein Reference Epstein2006). Abrams notes that social network models may provide one way of characterizing certain key differences among social roles by capturing differences in the direction, degree, and nature of what is communicated between agents. Another important direction is the “ecology of games” framework (Lubell Reference Lubell2013), which acknowledges that individuals participate in many social organizations simultaneously, that membership can be temporary, and that co-membership with others may overlap in some but not all arenas. Agent-based modeling has investigated institutions that promote assortment of cooperators under these conditions (Smaldino & Lubell Reference Smaldino and Lubell2011; Reference Smaldino and Lubell2014), but explicit evolutionary analysis has yet to be done.
Although the development of new models is paramount, we still have much to learn from existing models of social evolution and multilevel selection. Ibbotson provides a cautionary tale when he incorrectly claims that “what is good for the group is also good for the individual.” It is clear that this is not always the case. The adoption of a strategy or institutional framework can raise the average fitness in a group while lowering the relative or even the absolute fitness of certain individuals. Traditional multilevel modeling approaches can show how a trait that increases the average group fitness can evolve even when it is detrimental to the individual. Meanwhile, game theoretic constructs must move beyond simple cooperation. McCain helpfully points out that such games exist. He presents a very interesting game that has elements of both coordination and anti-coordination. I agree that this simple game may better represent the kinds of scenarios in which organized cooperation with division of labor works best. I have not seen an evolutionary analysis of such a game, but I agree that such an analysis could be valuable in understanding the emergence of social complexity. Nevertheless, as Mattei notes, the complex psychology of group-level traits presents a challenge for game theoretic modeling. For example, empirical research has shown that individuals' behavior in public goods games is heavily influenced by cultural norms (Gerkey Reference Gerkey2013; Henrich et al. Reference Henrich, Boyd, Bowles, Camerer, Fehr, Gintis, McElreath, Alvard, Barr, Ensminger, Henrich, Hill, Gil-White, Gurven, Marlowe, Patton and Tracer2005), as well as by the opportunity for players to communicate, even when communication is anonymous and therefore cannot influence reputation (Baum et al. Reference Baum, Paciotti, Richerson, Lubell and McElreath2012).
Nonacs & Kapheim brought to my attention their very useful concept of social heterosis, which is when individuals carrying each variety of a particular allele have a higher fitness in the presence of individuals carrying different alleles than with individuals carrying like alleles. Group-level traits in which individuals must differentiate require by definition a degree of social heterosis. In fact, Henrich and Boyd's (Reference Henrich and Boyd2008) model of division of labor entails a form of cultural social heterosis. When differentiated individuals exhibit a trait by virtue of their differences, Nonacs & Kapheim's models of social heterosis may be useful for capturing the relevant fitness trade-offs and evolutionary dynamics, and they are quite right that their modeling framework is fully compatible with classic MLS theory. Nevertheless, their model contains neither social organization nor cultural transmission. The benefit of a social trait depends on both the presence of other types of individuals and on a social environment that provides the affordances for displaying the trait. Agent-based and/or network theoretic interpretations of the social heterosis idea that allow for nongenetic (cultural) inheritance will be an important avenue for future research.
Whether our data comes from empirical studies of human genetic or cultural evolution or from a computational model, how can we identify the influence of group-level traits? Okasha suggests contextual analysis, which is a powerful method for establishing the influence of traits that affect group-level fitness, and is in some ways superior to the approach centered on the Price equation that has been more popular in analyses of cultural evolution (Okasha Reference Okasha2006). However, although contextual analysis can test for the presence of group-level effects on selection, it cannot distinguish between weak and strong group-level traits (Schank), nor help us understand how the relevant group-level traits influence the fitness of other individual- and group-level traits. Still, I would love to see contextual analysis, perhaps modified if need be, applied to the study of the evolution of emergent group-level traits.
R7. A focus on group-level traits can benefit the study of social behavior in many fields
In the target article, I focused on the importance of a group-level traits perspective for the understanding of human cultural evolution. As Wilson points out, an appreciation for treating the organized group as a unit of functional analysis is important for evolutionary theory in general, and not restricted to either culture or to humans. Nevertheless, my focus has been on humans. I am certainly not the first to note the importance of social organization for human social behavior, but I agree with Mattei that an explicit evolutionary perspective on those properties and behaviors that are properly defined at the group level can lead to advances in a number of the human sciences, including social neuroscience, game theory, linguistics, and economics, as well as more applied fields like marketing and law.
Shuai & Gong highlight the study of language evolution, and point out that the language of a group is an emergent group-level trait. Language also clearly facilitates group organization and coordination and is therefore essential in many other group-level traits in humans. Moreover, Shuai & Gong are right that understanding the roles of complex organization and the competitive advantage of group-level traits is probably crucial for a better understanding of the coevolution of language and human social complexity. As an example, they highlight the importance of language in the formation of social bonds. What else would explain why teenage friends or separated lovers often spend hours on the phone discussing their feelings, their hopes, and their fears? Characterizing human social complexity will probably require more than the simple analyses of static social networks and summary statistics (such as power-law degree distributions) presented by Shuai & Gong. For example, it is clear that to whom we are tied matters (Granovetter Reference Granovetter1973; Hill et al. Reference Hill, Walker, Bozicevic, Elder, Headland, Hewlett, Hurtado, Marlowe, Wiessner and Wood2011), as do the dynamics of those ties. Nonetheless, complex models that incorporate social structure with the evolution of language represent a good direction.
R8. Conclusion
Sitting alone at our desks, it is easy to focus on the aspects of human existence that are best described at the level of the individual. Consider the very process of writing, often characterized as a very solitary activity. I write something. I struggle, alone, and finally manage something I am satisfied with. I send it to the editor. The editor reads it, and gives it his approval. The publisher prints it, and the reader reads it. These are actions with social consequences but individual-level descriptions. Yet this is a false picture of the writing-to-reading process. In reality, I write standing on the shoulders of giants, with the influence of the many articles and books I have read, filtered through my mental models and biases shaped by years of explicit learning and cultural indoctrination. I discuss my piece with friends and colleagues, who shape it further. I submit it to the editor, who solicits the advice of experts, and makes recommendations upon which I base a revision. Upon acceptance, the piece is inspected and refined by a team of proofreaders and editorial assistants. Printers, internet service providers, IT support staff, and mail carriers are all involved in the dissemination of the piece to its readers, who choose whether to read it based on a host of factors of personal history and circumstance, and who interpret it similarly. In the case of the unique structure of this particular journal, a new process starts at this point, as readers become writers, and beget an academic dialogue. The production of this very issue is not something that is produced by any individual, nor is it coordinated by a supreme leader with a host of underlings. Instead, the journal, along with many of its properties, are group-level traits, produced by the complex coordination of organized, differentiated individuals. So too are many, many other aspects of the lives of humans best described.
The role of group-level traits in human evolution is at present poorly understood. Changing that will be a vastly interdisciplinary undertaking, with contributions needed from across those fields concerned with human behavior, evolution, and the dynamics of complex systems. It is an imposing challenge, but one that is surmountable when we work together.
ACKNOWLEDGMENTS
Thanks to Bert Baumgaertner, Paul Bloom, Lesley Newson, Emily Newton, and Pete Richerson for comments.
