Major controversies in science have a way of appearing obvious in retrospect. We find it hard to understand why smart people took so long to agree that the earth revolves around the sun, that glaciers once covered the northern latitudes, that the continents drift, and that species are derived from other species.So it is with group selection, a theory that was declared dead in the 1960’s, only to come to life as an essential tool for understanding animal and human societies. Group selection theory employs the following assumptions.

1) Natural selection is based on relative fitness.

2) Selection among individuals within groups tends to favor traits that are called selfish in human terms; that is, traits that benefit individuals at the expense of other members of the group and the group as a whole.

3) The evolution of group-advantageous traits typically requires a process of selection among groups in a multi-group population.

4) Groups are defined as the individuals who influence each other’s fitness with respect to the evolving trait.

Or, as Edward O. Wilson and I put it in a 2007 article[i], “Selfishness beats altruism within groups. Altruistic groups beat selfish groups. Everything else is commentary.”

Assumption (1) is one of the bedrock assumptions of evolutionary theory. Assumption (2) is a basic matter of tradeoffs. Imagine modifying the game of monopoly so that the goal is to collectively purchase and develop the properties. That game would call for a completely different set of strategies than when the goal is to beat the other players. Most social endeavors are like the game of monopoly in this respect. Assumption (3) seems to follow inevitably from (1) and (2). If fitness differences within groups weigh in favor of selfishness, then what else can weigh in favor of group-advantageous traits but fitness differences between groups in a multi-group population? Assumption (4) is a common-sense definition of groups that is invariably employed in mathematical models of social evolution to identify the individuals that influence the fitness of any given focal individual. Taken together, these assumptions have the same kind of obviousness in retrospect that makes the three basic ingredients of natural selection (variation, selection, and heritability) so obvious and caused Huxley to exclaim “How stupid of me not to have thought of that!”

Against this background, it seems remarkable that George C. Williams could declare in 1966 that “group-level adaptations do not, in fact, exist” [ii]and that every other major theory of social evolution was self-consciously developed as a way to explain seemingly group-advantageous behaviors without invoking group selection. As an example, Richard Dawkins began his book The Selfish Gene[iii] by describing how a group of altruists invaded by a selfish individual must inevitably convert to a selfish group and how the process of between-group selection, while possible in principle, was invariably too weak to oppose the within-group advantage of selfishness. By his own account, the major import of selfish gene theory was to provide another explanation for how selfish genes can cooperate in the bodies of individuals and, occasionally, groups.Today, the four assumptions listed above can be accepted at face value and many examples have been documented of traits that evolve on the strength of between-group selection, despite being selectively disadvantageous within groups. Between-group selection can even dominate within-group selection for multiple traits, turning groups into organisms in their own right, a phenomenon known as major evolutionary transitions. But the revival of group selection theory doesn’t mean that the alternative theories are wrong. All of the major theories of social evolution can qualify as right and might deserve to coexist, as strange as this might sound. A concept of equivalence has emerged from the controversy over group selection that rivals the concept of paradigms in importance when it comes to understanding the scientific process.

To introduce the concept of equivalence, imagine an Englishman who is so pompous that he regards his language as superior to every other language. In his most insufferable moments, he claims that other languages aren’t capable of expressing concepts that can be expressed in English. In his saner moments, be begrudgingly admits that other languages can express the same concepts but not nearly as well as English. To him, it is only reasonable for everyone to speak English.

The concept of equivalence points out that some scientific theories are like languages. They examine the same causal processes but in ways that require a degree of fluency to translate from one to the other. Scientists who are well versed in only one theory can mistakenly assume that the other theories are just plain wrong or at best provide a clumsy way to describe what their own theory describes more elegantly. Scientists who don’t get the concept of equivalence are like pompous Englishmen.

Equivalent theories are different from alternative hypotheses or the concept of paradigms that was first articulated by the philosopher Thomas Kuhn in the 1970’s. Alternative hypotheses invoke different causal processes. In the game of Clue, I might hypothesize that the murder took place in the library, with the revolver, by Miss Scarlet. There is a right and wrong to the matter and more than one hypothesis can’t be correct. Kuhn observed that scientists sometimes get stuck viewing a topic a certain way. Their particular configuration of ideas is capable of a limited degree of change through hypothesis formation and testing, but cannot escape from their own assumptions in other respects, which makes the replacement of one configuration of ideas (a paradigm) by another configuration a messy and uncertain process. Still, in Kuhn’s rendering, one paradigm eventually does replace its rival. Scientists don’t talk about pre-Copernican views of the solar system any more.

The concept of equivalence therefore adds to the more familiar concepts of alternative hypotheses and paradigms in the philosophy of science, not just for the group selection controversy but all scientific controversies. The great danger is to mistake an equivalent theory, which invokes the same causal processes, for an alternative hypothesis or paradigm that invokes different causal processes. If equivalent theories are treated as if one deserves to replace another, nothing but confusion can result.Precisely this confusion has plagued evolutionary theories of social behavior for over half a century. The replicators and vehicles of selfish gene theory, the coefficient of relatedness in kin selection theory, and the N-person groups of evolutionary game theory are all so many ways of describing evolution in multi-group populations that include the four assumptions listed above. They are inter-translatable with each other and with group selection theory. This means that they can all be right in predicting what trait evolves in a given situation. But they were wrong when they claimed to explain the evolution of these traits without invoking between-group selection, since they invoke everything but the name within their own frameworks.Once the concept of equivalence is understood, the group selection controversy can be declared over.

I provide a post-resolution account in my newest book Does Altruism Exist? Culture, Genes, and the Welfare of Others [iv]. In keeping with the obviousness-in-retrospect of resolved controversies, I describe how altruism evolves and the concept of equivalence in terms that can be understood by any interested reader without requiring specialized knowledge. If you doubt my own assessment, then consider a recently published article in the Annual Review of Psychology titled “The Evolution of Altruism in Humans” by Robert Kurzban, Maxwell N. Burton-Chellew, and Stuart A. West[v], who is one of the main current critics of group selection theory. They state: “The new group selection (multilevel selection) and kin selection are just different ways of conceptualizing the same evolutionary dynamics…In all cases, where both methods have been used to look at the same problem, they yield identical results (p 10.5).” Or consider a recent article published in BioScience titled “Kin Selection and Its Critics” by Jonathan Birch and Samir Okasha[vi], who is widely regarded as one of the foremost authorities on theories of social evolution. They state: “In earlier debates, biologists tended to regard kin and multilevel selection as rival empirical hypotheses, but many contemporary biologists regard them as ultimately equivalent, on the grounds that gene frequency change can be correctly computed using either approach. Although dissenters from this equivalence claim can be found, the majority of social evolutionists appear to endorse it (p 28).” That sounds like a resolution to me. Anyone who argues the major theories of social evolution against each other runs the risk of sounding like a pompous Englishman who insists that everyone else should speak English.

Some terminological points are in order to fully make sense of the passages quoted above. Group selection has always been a bi-level theory by partitioning evolutionary change into within- and between-group components. It becomes multi-level when it is extended downward (between genes within individual organisms) and upward in a multi-tier hierarchy of units. Some authors think that multilevel selection theory has changed enough over the years to distinguish between an old and new version. The same can be said of kin selection, which no longer is restricted to genealogical relatedness, for example. Birch and Okasha distinguish between three current versions of kin selection in their article. Theory development is to be expected and is not a sign of weakness or inconsistency.With equivalence firmly in mind, a new set of issues comes to the fore. An equivalent theory must pull its weight in some way to remain in use, presumably by offering insights that are less forthcoming from the other theories by virtue of their different perspectives. Also, two theories might be equivalent in some but not all respects. For the rest of this article I will provide three examples of group selection that might not be translatable into kin selection. I challenge my kin selectionist colleagues to prove me wrong.

Kin selection theory attempts to state the criterion for the evolution of a trait in the form rb>c, where c represents the effect of the trait on the fitness of the actor (which is negative for an altruistic trait), b represents the effect of the trait on the fitness of the recipient (which is positive for an altruistic trait), and r is a coefficient of relatedness between the actor and the recipient, which can be based on genealogical relatedness or any other process that causes like to associate with like. Birch and Okasha provide an up-to-date review of the ins and outs of these terms. Here are three examples of group selection that are hard and perhaps even impossible express in this form.

Equilibrium selection: Complex social interactions often result in multiple locally stable equilibria, which means that a given trait can be favored within some groups but not others. Locally stable equilibria can vary widely in their group-level properties and between-group selection favors those that collectively survive and reproduce better than others[vii]. It is difficult to see how this process, which is straightforward to describe in terms of multilevel selection, can be expressed in the form rb>c.

Community-level selection: In the 1980’s, Charles Goodnight raised two species of flour beetles (call them A and B) in vials of flour, with each vial containing both species[viii]. After several weeks, vials with the highest density of one of the species (say, A) were selected to initiate a new set of vials. In other words, two-species communities were selected on the basis of a phenotypic trait, which was the density of one of the species. There was a response to selection, such that the density of species A increased over a number of generations. What genes were selected to produce this effect? It turns out that genes were selected in both species that interacted with each other to increase the density of species A. In essence, the species were like chromosomes inside a single vehicle of selection. How can this be expressed in the form rb>c at the level of individual beetles?

William Swenson and I continued this line of research with more complex communities[ix]. In one experiment, we grew plants in sterilized soil to which we added six grams of unsterilized soil from a well mixed slurry. In separate lines, the soil from underneath the largest and smallest plants was selected to inoculate a new generation of pots to grow plants from the same seed source as the previous generation. In other words, we were selecting soil ecosystems rather than plant genes. In another set of experiments, we selected aquatic microbial ecosystems for their ability to change the acidity of their media or to degrade a toxic compound. The communities that initiated each generation included millions of individuals and hundreds of species. The initial variation among communities at the beginning of each generation, based on sampling error, was negligible. Nevertheless, complex interactions among the species created variation among communities in the phenotypic trait under selection and there was a response to selection, which is proof of heritability at the ecosystem level. The experiment was recently replicated for soil ecosystems using the phenotypic trait of plant flowering time[x] and the method holds promise for selecting ecosystems with useful properties, such as plant growth or soil remediation. It can be easily understood as an example of higher-level selection but how can it be expressed in the form rb>c from the perspective of individual organisms within the ecosystems?

Human Cultural Evolution: Multi-level selection theory is proving to be indispensible for the study of human cultural evolution. Human groups vary widely in their practices in ways that have little to do with genetic relatedness, genealogical or otherwise. Phenotypic variation occurs at all spatial scales, from hunter-gatherer groups to nations. A given cultural trait can spread at the expense of other traits within the group or by virtue of giving the group a competitive advantage over other group. Peter Turchin uses multi-level cultural evolution to explain the rise and fall of empires. Extreme between-group conflict acts as a crucible for the selection of cooperative societies. The most cooperative expand into empires, but then disruptive cultural evolution takes place within the societies, causing their collapse. In a TVOL article titled “Blueprint for the Global Village”, Dag Olav Hessen and I show how multilevel selection can be used to formulate public policy, using Norway’s pension fund as an example. I find it difficult to see how cultural group selection operating at large spatial scales over long time periods can be translated into kin selection theory.

There is a good reason why multilevel selection theory and kin selection theory might not be equivalent in all respects. Multilevel selection theory is not a modeling method. It is a causal hypothesis based on the four assumptions listed at the beginning of this article, which can be modeled in any number of ways (e.g., analytical models, computer simulations). In contrast, kin selection theory is a modeling method for traits that can be measured in individual organisms, which requires estimation of the b, c, and r terms. This makes kin selection theory unsuitable for cases in which individual-level traits interact with other individual-level traits in a complex fashion to produce group level phenotypes. As we have seen, when groups and multi-species communities are selected in the laboratory, the entire experiment can take place by measuring variation, selection, and heritability at the level of higher-level units without requiring any knowledge of individual-level traits, just as artificial selection experiments at the level of individual organisms can take place without any knowledge of the genes and their interactions that result in variation and heritability at the individual level.It is not my purpose to disparage kin selection theory, however. As someone who appreciates equivalence, I acknowledge the usefulness of kin selection theory in some cases. My challenge to kin selectionists to explain these three cases is friendly and I will applaud them if they succeed.


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

[ii] Williams, G. C. (1966). Adaptation and Natural Selection: a critique of some current evolutionary thought. Princeton: Princeton University Press.

[iii] Dawkins, R. (1976). The Selfish gene (1st ed.). Oxford: Oxford University Press.

[iv] Wilson, D. S. (2015). Does Altruism Exist? Culture, Genes, and the Welfare of Others. New Haven: Yale University Press.

[v] Kurzban, R., Burton-Chellew, M. N., & West, S. A. (2014). The Evolution of Altruism in Humans. Annual Review of Psychology. doi:10.1146/annurev-psych-010814-015355

[vi] Birch, J., & Okasha, S. (2014). Kin Selection and Its Critics. BioScience, 65(1), 22–32. doi:10.1093/biosci/biu196

[vii] Samuelson, L. (1997). Evolutionary games and equilibrium selection. Cambridge, MA: MIT Press.

[viii] Goodnight, C. J. (1990). Experimental studies of community evolution I: The response to selection at the community level. Evolution, 44, 1614–1624.Goodnight, C. J. (1990). Experimental studies of community evolution II: The ecological basis of the response to community selection. Evolution, 44, 1625–1636.

[ix] Swenson, W., Wilson, D. S., & Elias, R. (2000). Artificial Ecosystem Selection. Proceedings of the National Academy of Sciences, 97, 9110–9114.Swenson, W., Arendt, J., & Wilson, D. S. (2000). Artificial selection of microbial ecosystems for 3-chloroaniline biodegradation. Environmental Microbiology, 2, 564–571.

[x] Panke-Buisse, K., Poole, A. C., Goodrich, J. K., Ley, R. E., & Kao-Kniffin, J. (2014). Selection on soil microbiomes reveals reproducible impacts on plant function. The ISME Journal. doi:10.1038/ismej.2014.196