Richerson et al. have done a tremendous job of establishing the plausibility of cultural group selection (CGS) as a process of social evolution. However, they have not shown that it is “essential” to human cooperation, or even that it has operated on any single trait. Instead, they show that fitness can be partitioned so as to satisfy the conditions for CGS. This is an altogether different – and much weaker – claim.
The total fitness of a population can be carved up in multiple ways and attributed to any number of traits, so long as fitness and trait values covary. This is one of the more pleasing results derived from the Price equation (e.g., Price Reference Price1972; see also Hamilton Reference Hamilton and Fox1975; Marshall Reference Marshall2011; Queller Reference Queller1992), and it should bury the notion that individual- and multi-level selection accounts are predictively different. An allele that is favored in inclusive fitness or neighbor-modulated models will also be favored in group selection models (and vice versa), because the frequency of that allele is positively correlated with fitness whether we conceive of it as affecting individuals or groups. If CGS favors the same fitness effects as selection on genes (not a given, of course), then the findings in the target article should not be news. The average fitness of individuals is necessarily correlated with the fitness of their groups.
But just because we could view selection as taking place at the group level does not mean we should. Although individual-level and multi-level fitness partitions are predictively equivalent, they are not causally so (Birch & Okasha Reference Birch and Okasha2015; Okasha Reference Okasha2015; see also Okasha & Paternotte Reference Okasha and Paternotte2012). Sometimes, group traits directly cause group fitness, but at other times the relationship between the two is merely correlational. Using the wrong fitness partition will lead one to infer causality where none exists and, consequently, mischaracterize the adaptation. If there is individual-level variation that directly causes variation in individual fitness (Fig. 1A), then an individual-level partition is best. Of course, individual variation in traits will also directly affect trait variation at the group level; this is what Williams (Reference Williams1966) meant in saying that “a herd of fleet deer” will simultaneously appear as a “fleet herd.” In this instance, the target of selection is the fleetness of individuals, rather than that of the groups they make up. Conversely, trait variance at the group level that directly causes variance in group fitness (Fig. 1B) is best handled by a group- or multi-level partition.
Figure 1. Causality in individual and group selection. Arrows show the direction of direct causal effects. (A) An individual-level partition is justified when the individual trait directly affects individual fitness and there is no direct relationship between the group trait and group fitness. This results in an individual adaptation. (B) A group-level partition is justified when only the group trait directly affects group fitness and there is no direct relationship between the individual trait and individual fitness. This results in a group adaptation. Adapted from Okasha (Reference Okasha2015).
Consider warfare as an illustration. In typical battles where some live and die on each side, individual fitnesses cause group fitness through simple summation. In fights where the loser is annihilated (e.g., the losing battleship is sunk), however, individual fitness is directly caused by group fitness – individuals succeed or fail because their group succeeds or fails. A multi-level partition, such as CGS, is only warranted in the latter sort of case.
The burden of proof for a CGS account, then, is considerably heavier than what the target article presents. First, CGS needs to be tested on a case-by-case basis; it cannot be accepted wholesale because it cannot be distinguished from alternative fitness partitions in this way. A hypothesis for a particular group trait must be set up against alternative hypotheses. Second, the evidence must demonstrate the direct causality of the link between the group trait and group fitness (Fig. 1B), beyond mere correlation. Third, the evidence must show that the group trait has evolved and is maintained specifically by cultural transmission. While Richerson et al. do not ignore these issues, their “sketch of the evidence” falls short of making their case compelling.
Many alleged examples of group selection likely reflect individual-level adaptations. The requirements for group, as opposed to individual, adaptation are rather imposing: It is rare for individual human fates to be so intertwined with their group's fate that group success directly causes (rather than merely correlates with) individual success, such as all group members succeeding or failing together as a unit because of their group's composition. Generally, we can expect group traits to directly cause group adaptation only when partners are clonal or when there is extreme repression of competition (Gardner & Grafen Reference Gardner and Grafen2009; Okasha & Paternotte Reference Okasha and Paternotte2012), neither of which describes the human condition well. The examples given by Richerson et al. do not come close to meeting this criterion. Conflicts that lead to the success or failure of entire groups would, but such circumstances are rare.
Moreover, many cases of ostensible cultural transmission are explainable as strategic, individual responses to existing socio-ecological circumstances. Violent defense of one's honor – a key aspect of the “Culture of Honor” – in the Southern United States is often given as a shining example of cultural transmission (e.g., Nisbett & Cohen Reference Nisbett and Cohen1996; Richerson & Boyd Reference Richerson and Boyd2005). However, current variation in income inequality fully accounts for any cultural difference in homicide rates between the northern and southern states (Daly & Wilson Reference Daly, Wilson, Schaller, Norenzayan, Heine, Yamagishi and Kameda2010). Similarly, behavioral differences between collectivist and individualist cultures are supposedly maintained by internalized social norms and beliefs (e.g., Markus & Kitayama Reference Markus and Kitayama1991), yet these differences can disappear when the expectations of the different cultural groups are matched (Yamagishi et al. Reference Yamagishi, Hashimoto and Schug2008).
Richerson et al. wave away alternative explanations by toppling a straw-man of “evoked culture” that they readily admit no one holds and by relegating as “narrow” the interpretation of experimental research. They also ignore the fact that CGS explains genetically detrimental helping by invoking the “mistakes” of an adaptive learning mechanism (Barclay & Van Vugt Reference Barclay, Van Vugt, Schroeder and Graziano2015). More generally, the study of cooperation has long been hindered by confusion between proximate and ultimate causation (Barclay Reference Barclay2012; West et al. Reference West, El Mouden and Gardner2011), and we can now add to the list a persistent confusion over the utility of the group selection concept itself. We look forward to future work on CGS that convincingly demonstrates both the causal relationship between group traits and group fitness and the transmission of these traits by cultural means.
Richerson et al. have done a tremendous job of establishing the plausibility of cultural group selection (CGS) as a process of social evolution. However, they have not shown that it is “essential” to human cooperation, or even that it has operated on any single trait. Instead, they show that fitness can be partitioned so as to satisfy the conditions for CGS. This is an altogether different – and much weaker – claim.
The total fitness of a population can be carved up in multiple ways and attributed to any number of traits, so long as fitness and trait values covary. This is one of the more pleasing results derived from the Price equation (e.g., Price Reference Price1972; see also Hamilton Reference Hamilton and Fox1975; Marshall Reference Marshall2011; Queller Reference Queller1992), and it should bury the notion that individual- and multi-level selection accounts are predictively different. An allele that is favored in inclusive fitness or neighbor-modulated models will also be favored in group selection models (and vice versa), because the frequency of that allele is positively correlated with fitness whether we conceive of it as affecting individuals or groups. If CGS favors the same fitness effects as selection on genes (not a given, of course), then the findings in the target article should not be news. The average fitness of individuals is necessarily correlated with the fitness of their groups.
But just because we could view selection as taking place at the group level does not mean we should. Although individual-level and multi-level fitness partitions are predictively equivalent, they are not causally so (Birch & Okasha Reference Birch and Okasha2015; Okasha Reference Okasha2015; see also Okasha & Paternotte Reference Okasha and Paternotte2012). Sometimes, group traits directly cause group fitness, but at other times the relationship between the two is merely correlational. Using the wrong fitness partition will lead one to infer causality where none exists and, consequently, mischaracterize the adaptation. If there is individual-level variation that directly causes variation in individual fitness (Fig. 1A), then an individual-level partition is best. Of course, individual variation in traits will also directly affect trait variation at the group level; this is what Williams (Reference Williams1966) meant in saying that “a herd of fleet deer” will simultaneously appear as a “fleet herd.” In this instance, the target of selection is the fleetness of individuals, rather than that of the groups they make up. Conversely, trait variance at the group level that directly causes variance in group fitness (Fig. 1B) is best handled by a group- or multi-level partition.
Figure 1. Causality in individual and group selection. Arrows show the direction of direct causal effects. (A) An individual-level partition is justified when the individual trait directly affects individual fitness and there is no direct relationship between the group trait and group fitness. This results in an individual adaptation. (B) A group-level partition is justified when only the group trait directly affects group fitness and there is no direct relationship between the individual trait and individual fitness. This results in a group adaptation. Adapted from Okasha (Reference Okasha2015).
Consider warfare as an illustration. In typical battles where some live and die on each side, individual fitnesses cause group fitness through simple summation. In fights where the loser is annihilated (e.g., the losing battleship is sunk), however, individual fitness is directly caused by group fitness – individuals succeed or fail because their group succeeds or fails. A multi-level partition, such as CGS, is only warranted in the latter sort of case.
The burden of proof for a CGS account, then, is considerably heavier than what the target article presents. First, CGS needs to be tested on a case-by-case basis; it cannot be accepted wholesale because it cannot be distinguished from alternative fitness partitions in this way. A hypothesis for a particular group trait must be set up against alternative hypotheses. Second, the evidence must demonstrate the direct causality of the link between the group trait and group fitness (Fig. 1B), beyond mere correlation. Third, the evidence must show that the group trait has evolved and is maintained specifically by cultural transmission. While Richerson et al. do not ignore these issues, their “sketch of the evidence” falls short of making their case compelling.
Many alleged examples of group selection likely reflect individual-level adaptations. The requirements for group, as opposed to individual, adaptation are rather imposing: It is rare for individual human fates to be so intertwined with their group's fate that group success directly causes (rather than merely correlates with) individual success, such as all group members succeeding or failing together as a unit because of their group's composition. Generally, we can expect group traits to directly cause group adaptation only when partners are clonal or when there is extreme repression of competition (Gardner & Grafen Reference Gardner and Grafen2009; Okasha & Paternotte Reference Okasha and Paternotte2012), neither of which describes the human condition well. The examples given by Richerson et al. do not come close to meeting this criterion. Conflicts that lead to the success or failure of entire groups would, but such circumstances are rare.
Moreover, many cases of ostensible cultural transmission are explainable as strategic, individual responses to existing socio-ecological circumstances. Violent defense of one's honor – a key aspect of the “Culture of Honor” – in the Southern United States is often given as a shining example of cultural transmission (e.g., Nisbett & Cohen Reference Nisbett and Cohen1996; Richerson & Boyd Reference Richerson and Boyd2005). However, current variation in income inequality fully accounts for any cultural difference in homicide rates between the northern and southern states (Daly & Wilson Reference Daly, Wilson, Schaller, Norenzayan, Heine, Yamagishi and Kameda2010). Similarly, behavioral differences between collectivist and individualist cultures are supposedly maintained by internalized social norms and beliefs (e.g., Markus & Kitayama Reference Markus and Kitayama1991), yet these differences can disappear when the expectations of the different cultural groups are matched (Yamagishi et al. Reference Yamagishi, Hashimoto and Schug2008).
Richerson et al. wave away alternative explanations by toppling a straw-man of “evoked culture” that they readily admit no one holds and by relegating as “narrow” the interpretation of experimental research. They also ignore the fact that CGS explains genetically detrimental helping by invoking the “mistakes” of an adaptive learning mechanism (Barclay & Van Vugt Reference Barclay, Van Vugt, Schroeder and Graziano2015). More generally, the study of cooperation has long been hindered by confusion between proximate and ultimate causation (Barclay Reference Barclay2012; West et al. Reference West, El Mouden and Gardner2011), and we can now add to the list a persistent confusion over the utility of the group selection concept itself. We look forward to future work on CGS that convincingly demonstrates both the causal relationship between group traits and group fitness and the transmission of these traits by cultural means.