Richard Wrangham’s newest book, The Goodness Paradox, gets a lot right. The central thesis is that we are a self-domesticated species. We have bred ourselves for tameness, in the same way that we have bred our animal companions. The opposite of tameness is reactive aggression, or the tendency to lash out in a social confrontation. But there is another kind of aggression, the cool, calculated kind called proactive, that is also a hallmark of our species. Hence “the strange relationship between virtue and violence in human evolution”, which is the book’s subtitle.

The idea that we are a self-domesticated species has long roots but is experiencing a renaissance based on a now classic study of silver fox by the Russian scientists Dmitri Belyaev and Lyudmila Trut. The story is beautifully told in the book How To Tame a Fox and Build a Dog, co-authored by Trut and Lee Alan Dugatkin, who happens to be one of my former PhD students. Not only could tameness in silver foxes be selected in only a few generations, but a whole suite of other behavioral, physical, and life history traits also evolved as byproducts. Moreover, the same package of traits appears to evolve in all domesticated species. Thus, an important secondary theme of Wrangham’s book is that not all products of evolution require separate adaptive explanations, a point stressed by Stephen Jay Gould and Richard Lewontin in their classic “spandrels” article in 1979[1].

The existence of an entire syndrome of traits associated with domestication provides ways to test the hypothesis of self-domestication in humans and other species, such as bonobos vs. chimpanzees and species that inhabit islands compared to their continental ancestors. This allows Wrangham to be quite precise about when human self-domestication evolved. It is a hallmark of our entire species, Homo sapiens, compared to all other species in the Homo genus. Arguably, self-domestication is the reason why our species replaced those other species.

As one of the pre-eminent thinkers about primate and human evolution, Wrangham does an excellent job addressing all four questions that must be asked to fully explain any product of evolution, concerning their function (if any), phylogeny, mechanism, and development[2]. He is also one of the most lucid writers for a general audience. Hence, I warmly recommend The Goodness Paradox to experts and laypeople alike. I learned a lot from it and think that you will also.

But there is one thing that Wrangham gets wrong. He thinks that he can develop his thesis without invoking group selection, when he is invoking group selection in every way except using the words.

Group selection is the evolution of traits based on the differential survival and reproduction of groups in a multi-group population, as opposed to the differential survival and reproduction of individuals within groups. It is famously controversial. Among the scientists cited by Wrangham -- such as John Alcock, Richard Alexander, Scott Atran, David Barash, Paul Bingham, Christopher Boehm, Samuel Bowles, Elizabeth Cashdan, Timothy Cluttonbrock, Leda Cosmides, Jerry Coyne, Paul Crook, Martin Daly, Richard Dawkins, Lee Dugatkin, Frans de Waal, Andrew Gardner, Herbert Gintis, Ashley Griffin, Jonathan Haidt, Marc Hauser, Joseph Henrich, Robert Hinde, Dominic Johnson, Martin Nowak, Daijiro Okada, Steve Pinker, Anne Pusey. Matthew Ridley, Robert Sapolsky, Michael Shermer, Elliott Sober, Corina Tarnita, Ian Tattersall, Michael Tomasello, John Tooby, Carel van Schaik, Stuart West, E.O Wilson, Margo Wilson, and Robert Wright--some rely heavily upon group selection, others reject it, and still others treat group selection as equivalent to other theories of social evolution, the difference being a matter of perspective rather than the invocation of different causal processes[3].


Before critiquing Wrangham’s treatment of group selection, it is important to be precise about the definition of terms. The best way to do this is to briefly review what all theories of social evolution share in common.

Consider the evolution of a nonsocial trait such as coloration in desert living species. Individuals vary in their coloration, those that match their background are more fit, and coloration is partially heritable. The result: individuals that impressively match their background.

Now consider the evolution of a social trait such as docility and aggression. Aggressive individuals pick fights with others while docile individuals avoid fights. To model the evolution of these alternative traits, we must assign fitness values to them. But, unlike solitary traits, we cannot do this solely on the basis of their individual properties. The fitness of either type depends upon the other individuals with whom they socially interact. This makes the study of social behavior more complicated than the study of solitary traits.

Any study of social evolution must say something about the structure of social interactions.

This is true for a verbal model, a mathematical or computer simulation model, or an empirical study that aims to understand how social behaviors evolve. However, mathematical and computer simulation models have the virtue of being precise about their assumptions. For example, N-person game theory assumes that individuals socially interact in groups of size N. The fitness of any individual depends upon its trait value (such as aggressive vs. docile) and the trait values of the other members of its group. The simplest N-person game theory models assume a very large number of groups, the random distribution of individuals into groups, and the dissolution of the groups after a single round of social interactions.

Elaborated models consider non-random distributions of individuals, multiple interactions within groups, different patterns of dispersal, and so on. The details depend upon the biological details of the organism being modeled. If the real organisms interact in pairs, then only 2-person game theory will do. If they interact with genetic relatives or with partners chosen on the basis of previous experience, then the assumption of random interactions won’t do. If the groups persist indefinitely and trade a fraction of dispersers every generation, then the assumption of ephemeral groups won’t do. It is the biology of real-world organisms that decide the details of any given model!

No matter what the details, all models of social evolution share the following features.

1) The sets of socially interacting individuals that influence each other’s fitness (the N in N-person game theory) are small compared to the total evolving population. This means that most evolving populations are populations of groups in addition to populations of individuals within groups—such as fish schools, bird flocks, primate troops, and human tribes. Sometimes the groups have discrete boundaries but sometimes they are neighborhoods, such as plants that interact only with their immediate neighbors. The important common denominator is social interactions that are local compared to the size of the total evolving population.

2) Selection among individuals within each group tends to favor traits that would be called disruptively self-serving in human terms, such as aggression compared to docility. In N-person game theory, virtually all of the traits called altruistic or cooperative are selectively disadvantageous within the groups of size N. Even the tit-for-tat strategy of 2-person game theory, which starts out nice and thereafter imitates the previous play of its partner, never beats its partner and can only lose or draw.

3) If social traits that are variously called altruistic, cooperative, mutualistic, and prosocial cannot evolve by within-group selection, then they require the differential survival and productivity of groups in a multi-group population. As I put it in my 2007 article with E.O. Wilson titled “Rethinking the Theoretical Foundation of Sociobiology”, selfishness beats altruism within groups, altruistic groups beat selfish groups, and everything else is commentary[4]. In 2-person game theory, pairs of altruists do better than mixed pairs, which in turn do better than pairs of selfish individuals. This between-group advantage for altruism can override its within-group disadvantage, especially if the distribution of individuals into groups is above-random.

Notice that these three features apply to all models of social evolution, no matter what they are called. Moreover, all models of social evolution must conform to the biological details of the social traits being modeled. Otherwise they will simply arrive at the wrong answer. The definition of groups and the larger population structure whereby groups are formed and dissolve are not arbitrary. They must be tailored to each and every social trait being modelled. This is why I coined the term “trait-group” in my first article on group selection in 1975[5].

Against this background, we can define “individual selection” as “selection among individuals within groups” and “group selection” as “selection among groups in a multi-group population”. These are the definitions that are used in virtually all explicit group selection models. They also capture what Darwin meant when he famously wrote “although a high standard of morality gives but a slight or no advantage to each individual man and his children over the other men of the same tribe, yet that an advancement in the standard of morality and an increase in the number of well-endowed men will certainly give an immense advantage to one tribe over another[6].”


Now I am in a position to advance my claim that Wrangham rejects group selection in his own mind but invokes it in every way except using the words when developing his thesis. He directly discusses group selection three times in the main text. The first is when he describes a theoretical model of warfare by Jung-Kyoo Choi and Samuel Bowles (location 2410 of the kindle edition)[7]. The details need not concern us, other than to say that they conform to Darwin’s scenario and the three features of all models of social evolution just listed. Wrangham accepts it as a group selection model but rejects it for not getting the biological details of human warfare in hunter-gatherer groups right. Fair enough. I have myself stressed that the model must match the biology.

The second mention of group selection is in the following passage (3763):

Group selection theory suggests that self-sacrifice by an individual can be favored over evolutionary time if it provides sufficiently large benefits to the individual’s group, which normally means a social breeding unit such as a hunter-gatherer band. Very often, however, the group that benefits from an individual’s generosity is not a social breeding unit. As Robert Graves’s recollection of his school days reminds us, the beneficiaries might be only a subgroup of a given social network. In the group as a whole, moral behavior might benefit some individuals at the expense of others.

Robert Graves’s recollection was of his school days, where he and his friends would never cheat on each other but thought nothing of cheating on their teachers. Here, Wrangham assumes that group selection models have some fixed definition of groups, as opposed to being defined in reference to each trait. For behavior expressed among school chums, the group of chums is the salient group.

Here is Wrangham’s third mention of group selection (5002):

Group selection is commonly invoked to explain our species’s interest in nonrelatives and our occasional willingness to sacrifice our own interests on behalf of a larger good. Group selection theory, however, has never quite been able to explain how benefits at the group level override those of individuals. The theory that the moral senses evolved to protect individuals from the socially powerful suggests that group selection might be unnecessary for explaining why we are such a group-oriented species. Our deference to the coalitionary powers within our own groups leads to a reduced intensity of competition, enabling groups to thrive.

This passage contains two errors. The first is to suppose that the counterforce to within-group selection is mysterious, nebulous, or necessarily weak. This is certainly not the case for formal mathematical and computer simulation models, where between-group selection is as precisely specified as within-group selection. As we shall see, it isn’t true for verbal models or empirical studies of social behavior either.

The second error is for Wrangham to assume that his own way of thinking described in the second half of the paragraph differs from his description of group selection in the first half. Let’s consider his own account in more detail.


Wrangham is not a mathematical or computer simulation modeler, but that doesn’t matter. At the beginning of this essay I stated that any study of social evolution must say something about the structure of social interactions. Formal models have the virtue of being precise about their assumptions, but verbal models based on extensive field experience have the virtue of being realistic—the kind of realism that should be the starting point of the more formal models.

Let’s begin with the behavior of chimpanzees, which Wrangham has studied extensively in the wild. Populations are subdivided into communities of a few dozen individuals. Males remain within their natal groups while females move. Males defend the boundaries of their territories against the males of adjacent communities.

Some behaviors expressed by individuals have an impact on the whole community. However, other behaviors are more limited in their impact, such as dyadic interactions or competing cliques within the community. Thus, the community is not a one-size-fits-all group for chimpanzees, any more than a band is a one-size-fits-all group for hunter-gatherers.

Despite these complexities, some chimpanzee behaviors are easy enough to interpret from a multilevel evolutionary perspective. Take reactive aggression for example. It is far, far more common in chimps than in human groups and clearly benefits the aggressor compared to other members of the same community. It is a form of disruptive selfishness, favored by within-group selection, pure and simple.

While humans are far less reactively aggressive and far more proactively aggressive than chimps, chimps do display proactive aggression to a degree, especially in their behavior directed against members of other communities. Here is one example described by Wrangham (4149).

In a few primate species (such as chimpanzees), infanticide occurs for reasons other than sexual selection. Male chimpanzees who encounter mothers from neighboring communities tend to attack them and can severely wound or kill their small infants. In this case, the protagonists are unlikely to meet again, so there is little chance of the killer’s fathering the female’s next infant. The traditional sexual selection theory, therefore, does not apply. Possibly, the killers benefit by intimidating the female into avoiding the area, leaving more food for the killer’s community. Alternatively, the attackers might gain by killing male infants that would otherwise grow up in the neighboring community to become future opponents. Further observations will eventually test ideas.

Let’s give Wrangham the benefit of the doubt and assume that his interpretation is correct. In the traditional sexual selection theory of infanticide, males that kill infants within their own community father more offspring than males who don’t. That’s a case of disruptive within-group selection pure and simple, which is bad for the group. But harming a female from another community is a different matter. The benefits do not flow to the males inflicting the harm, but to their entire community, in the form of an expanded territory and fewer males in the adjacent community.

The same is true for killing adult males of adjacent territories, as described by Wrangham in this passage:

The attacks cost little for the attackers, but by eliminating rivals they benefit their own community. In Kibale’s Ngogo community, John Mitani and David Watt’s team recorded instances when males killed or fatally wounded eighteen members of neighboring communities during period of ten years. The Ngogo community then expanded their territory into the area where most of the kills occurred. In Gombe, Anne Pusey and her colleagues have shown that, when the territory occupied by a community increases in size, community members are better fed, breed faster, and survive better. Kill some neighbors, expand the territory, get more food, have more babies—and be safer at the same time, since there are fewer neighbors who might be able to attack you.

If Wrangham is correct in his interpretation, then by his own account he is describing a case of between-group selection. The proactively aggressive behavior provides a benefit for the whole community at a cost to the aggressors. Wrangham makes much of the fact that the individual cost of killing is not large because many are ganging up against one. Still, the cost is probably something and even if it was zero the proactively aggressive behavior would be neutral with respect to within-group selection. The benefit remains at the group level. The fact that the cost of providing a group-level benefit is low makes group selection plausible, because it is not strongly opposed by selection within groups.

Notice also that Wrangham is able to describe the group-level benefits as clearly as the individual-level costs. There is nothing mysterious, nebulous, or necessarily weak about expanding the territory of a community over a period of years.

To summarize, even before we get to human self-domestication, Wrangham is explaining reactive aggression as a product of within-group selection and proactive aggression as a product of between-group selection in chimpanzees. In the latter case, he is invoking group selection in every way except using the words.


Before proceeding to the human case, it is necessary to return to theoretical models. I have already shown that modelling the evolution of social behaviors is more complicated than modeling the evolution of solitary behaviors. Modeling the evolution of social control mechanisms is more complicated still.

Let’s begin with a standard model of altruism and selfishness in an N-person game theory model. Now let’s introduce a second trait. Some individuals punish selfish members of the group, while others allow selfishness to go unpunished. This creates four combinations of individuals in any given group: selfish punishers (SP), selfish non-punishers (SN), Altruistic punishers (AP), and altruistic non-punishers (AN). If there are enough punishers in a group, then selfishness no longer beats altruism. However, non-punishers enjoy the benefits of social control provided by the punishers without paying the cost. We haven’t solved the problem of altruism, but merely relocated it from the originally altruistic trait to the punishment trait. A rich literature exists on this topic using phrases such as “altruistic punishment” and “second-order public goods”. One fascinating result explored by another of my former PhD students, Omar Eldakar, is that of the four combinations of individuals, group selection can result in a mix of selfish punishers and altruistic non-punishers. Altruistic punishers go extinct because they pay a double cost—the cost of being an altruist and the cost of being a punisher. Selfish non-punishers are held at a low frequency by the selfish punishers. It’s as if the benefits of selfishness become a payment for the cost of being a punisher[8]!

Against this background, we can consider the central premise of Wrangham’s book; that self-domestication evolved in our species because individuals who could not control their aggressive impulses were executed. Or more precisely, execution is a necessary arrow in an entire quiver of social control mechanisms that begins with mild sanctions such as gossip and escalates as needed.

Based on what I have said about social control mechanisms as second-order public goods, it should no longer surprise the reader that the many examples provided by Wrangham invoke group selection in every way except using the words. Here is one example (2251).

Prior to Homo sapiens, Marean suggests, humans lived at low density in small societies, like chimpanzees. Then one population, which he thought might have lived on the southern African coast, developed an ability to gather and hunt so well that their food resources became far more productive. The population naturally grew to the point where there was competition over the food supply, and soon groups were fighting over the best territories. Success in war became imperative. Groups accordingly allied with one another, giving rise to large societies of the type that hunter-gatherers form today. Cooperation among warriors within groups was so vital for winning conflicts that it evolved to become the basis of humans’ exceptional propensity for mutual aid. Sociality became more complex, learning became more vital, and culture became richer.

Could there be a more clear description of group-level selection? It is little different from Darwin’s original speculation or the intent of the Choi and Bowles model of warfare. Wrangham makes much of the fact that with the advent of language and weaponry, punishing deviant behavior became so effective that groups became “a tyranny of cousins” (2777) and “like a boardroom without a chairman” (2802) in enforcing their norms. He can’t seem to see what every model of social control concludes: the lower the cost of punishment, the stronger between-group selection is relative to within-group selection.


The reason that I listed forty-one scientists by name at the beginning of this essay is to stress the magnitude of the problem that I have examined in detail for Wrangham’s book. Every one of those authors discuss the three features shared by all models of social evolution in their writing. They have no choice if they want to be biologically realistic. Some of the authors recognize between-group selection when they see it, including Christopher Boehm, who pioneered the concept of reverse dominance as an important factor in human evolution. Peter Turchin, whose 2014 book Ultrasociety: How 10,000 Years of Warfare Made Humans the Most Cooperative Species on Earth covers much of the same ground as Wrangham’s book, says this: “The central breakthrough in this new field is the theory of cultural multilevel selection.”

Other authors on the list, like Wrangham, manage to describe group selection as a failed concept even when it is staring them in the face. My respect for Wrangham and praise for The Goodness Paradox in every other respect is genuine, but someday historians will look back in wonderment as to how otherwise smart people, who were part of the same scientific community, managed to remain so divided in their own minds about the importance of group selection in human evolution.


[1] Gould, S. J., & Lewontin, R. C. (1979). The spandrels of San Marco and the panglossian paradigm: A critique of the adaptationist program. Proceedings of the Royal Society of London, B205, 581–598.

[2] These are known as “Tinbergen’s four questions” based on this article by the Nobel laureate Niko Tinbergen: Tinbergen, N. (1963). On aims and methods of ethology. Zeitschrift Für Tierpsychologie, 20, 410–433.

[3] For more on equivalence, see Chapter 3 of my book: Wilson, D. S. (2015). Does Altruism Exist? Culture, Genes, and the Welfare of Others. New Haven, CT: Yale University Press.

[4] Wilson, D. S., & Wilson, E. O. (2007). Rethinking the theoretical foundation of sociobiology. Quarterly Review of Biology, 82, 327–348.

[5] Wilson, D. S. (1975). A general theory of group selection. Proc. Natl. Acad. Sci., 72, 143–146.

[6] Darwin, C. (1871). The descent of man and selection in relation to sex (Vol. 2 vol.). London, UK: John Murray. In addition, see my conversation with Elliott Sober titled “Was Darwin a Group Selectionist?”:

[7] Choi, J.-K., & Bowles, S. (2007). The Coevolution of Parochial Altruism and War. Science, 318(5850), 636–640.

[8] Eldakar, O. T., Farrell, D. L., & Wilson, D. S. (2007). Selfish punishment: altruism can be maintained by competition among cheaters. Journal of Theoretical Biology, 2007, 198–205.