Evolution for Everyone

The haystack model (see T&R VIII) includes many assumptions but one was especially biased. Recall that each haystack is colonized by a single fertilized female bearing four genes coding for docility or aggressiveness–two of her own and two from her mate. Maynard Smith assumed that if even one of these genes codes for aggressiveness, then the aggressive gene entirely replaces the docile gene by the time the mice disperse from the haystack. The docile gene is not just at a selective disadvantage within groups. It is as disadvantageous as it can possibly be.

Hamilton’s theory, which Maynard Smith dubbed kin selection, made a different set of assumptions. Hamilton modeled altruism as an interaction between two individuals. Like the good Samaritan helping someone in need, an altruist increases the fitness of the recipient by an amount b and decreases its own fitness by an amount c. Selfish individuals happily receive but do not give.

Since Maynard Smith was trying to compare kin selection with group selection, it seems only fair to use Hamilton’s definition of altruism in the haystack model. This can be easily done. As before, we assume that haystacks are colonized by a single fertilized female, only now the genes code for altruism and selfishness as defined by Hamilton. Within each group containing both genes, the selfish gene has the advantage and starts to replace the altruistic gene. The rate that this happens depends upon the particular values of the b and c terms. For example, if a group is initiated by one altruistic and three selfish genes, and if the mice disperse after ten generations, then the altruistic gene might decline from an initial frequency of 25% to a frequency of 8%, but there is no reason why it must necessarily decline to zero.

Similarly, groups that start with more altruistic genes grow faster than groups starting with more selfish genes, and the rate that this happens depends upon the particular values of the b and c terms. For example, after ten generations, groups initiated by one altruistic and three selfish genes might be 40% more productive than groups initiated by four selfish genes.

The modified haystack model captures the essence of what I call the original problem (see T&R II), just like the original haystack model. In both cases, the trait that is “for the good of the group” is selectively disadvantageous within groups and requires a process of between-group selection to evolve. In the modified model, however, the b’s and c’s are allowed to determine the relative importance of within- and between-group selection, rather than arbitrarily assuming that within-group selection is as strong as it can possibly be.

What is the result of the modified haystack model? It turns out that altruism can evolve by group selection, using reasonable values of b and c, even when the altruistic gene is initially rare in the total population. The model that led to the rejection of group selection is favorable for group selection after all.

What was the impact of the modified model? Did it cause the entire field to reconsider the rejection of group selection? Not in the least. Nobody even thought to modify the original model until 1986, when I published an article titled “The haystack model revisited” in the journal Evolution. By then, group selection was thoroughly taboo and the article had no noticeable impact.

It gets worse. In 1970, George Price published a model that divided evolution into within- and between-group components and clearly indicated a role for between-group selection. The Price equation is regarded as a thing of beauty by theoretical biologists today, but at the time it had virtually no impact on the triumphant march of individual selection theory that had begun only a few years earlier. In 1975, Hamilton reformulated his theory on the basis of the Price equation, as I will recount in a future installment. According to Hamilton’s new interpretation, kin selection is a kind of group selection rather than an alternative to group selection. During the same year, I published my first model demonstrating the plausibility of group selection–but the individual selection bandwagon rolled on.

So much for blustery claims by Dawkins in 1982 and Alexander in 1987 that the search for plausible models of group selection had been exhausted. When we focus on the original problem, there is near universal agreement among theoretical biologists that between-group selection can successfully counter within-group selection. The recent Nature article on group selection (see T&R VII) quotes the theoretical biologist Andy Gardner as saying “Everyone agrees that group selection occurs.”

The fact that Gardner remains one of the most severe critics of multilevel selection theory will be explained in a future installment. Moreover, his statement accurately applies only to theoretical biologists knowledgeable about the subject. The vast majority of evolutionists receive their knowledge of theory secondhand, starting from textbooks when they are students. For them, the claim that group selection remains theoretically unsupported still rolls on. The situation is even worse for people from other fields interested in evolution and for the general public, who receive their knowledge of theory third, fourth and fifth hand.

The events that I have recounted provide a fascinating example of stasis in science, whereby a major decision becomes set in stone and is not easily revised, even when it richly deserves to be. If they knew then what we know now, group selection would never have been rejected as theoretically implausible. Yet, the field as a whole does not spontaneously clean up its mess after the fact. That is why a deliberate effort is required. Andy Gardner and I might disagree at some level, but I think I speak for both of us when I say that group selection is theoretically well supported. That should be the new consensus view. Those who disagree should familiarize themselves with the current literature before repeating the formulaic statements of the past.

To be continued.


  1. #1 bob koepp
    October 31, 2009

    So, it looks to me like a plausible “explanation” of Maynard Smith’s hard-core individual selectionist stance invokes the common human tendency to assign the most extreme values (often, “to the limit”) to model parameters when engaging in thought experiments. Sometimes this works to our advantage, sometimes not.

  2. #2 David Sloan Wilson
    October 31, 2009

    Sure. Maynard Smith didn’t intend the Haystack model to be the final word. What’s interesting is that it became the final word. Everyone was so ready to reject group selection that the mere hint of theoretical implausibility was sufficient.

    It’s also curious in retrospect that Maynard Smith couldn’t see kin selection in his own haystack model. After all, a haystack is a group of full siblings, the only twist being that it lasts for multiple generations. Today, most people reflexively categorize the haystack model as an example of kin selection, even though it was Maynard Smith’s alternative.

    Here is my explanation: When selection is examined within and among groups, as in the haystack model and the Price equation, then group selection jumps out at you. When selection is merely calculated for the total population without examining what’s happening locally, as in Hamilton’s original formulation, it seems that group selection isn’t being invoked even when it is. That’s why Maynard Smith regarded his haystack model as an example of group selection, why he and Hamilton regarded Hamilton’s original formulation as not group selection, and why Hamilton changed his mind as soon as he saw his own theory through the lens of the Price equation.

    The question remains: Why did the rejection of group selection become set in stone, given all these developments only a few years later?

  3. #3 bob koepp
    October 31, 2009

    Yes, it is curious that Maynard Smith, and legions of those who read him, did not see that the haystack world begs for interpretation in terms of kin selection. Why did he think kin selection represented an alternative to group selection? What’s going on, cognitionwise, when we see contrasts where none exist?

  4. #4 InfuriatedSciTeacher
    October 31, 2009

    Is it possible that the political climate of the “Free world” at the time that group selection theory was introduced contributed to its dismissal? Collectivism wasn’t exactly popular in the 60’s and 70’s, from what I know of that period (I was born in ’78, so someone who lived the time period is welcome to trounce that idea for me).

  5. #5 piker
    October 31, 2009

    It’s also possible that people in the 60s and 70s realized intuitively that there was nothing there that amounted to selection in the evolutionary sense at all. It simply looked like, as noted above, a thought experiment involving humans setting up mice to breed selectively.

    And again I would expect that the mutual exclusivity paradox stuck out for many like a sore on the Mendellian thumb.

  6. #6 Shane Horan
    October 31, 2009

    I think we’ve taken the haystack model as far as it can go. We know it’s theoretically possible. The question is, is it an important factor in evolution as it actually happens.
    If we observe a seemingly altruistic group of animals who had a haystack-like life cycle (likely a parasite), it’s a live empirical question to decide if between-group selection is why the population is largely altruistic. There may be another reason, e.g. the emergence of a “policing” gene in the altruistic individuals to suppress the egoists. To decide between the two possibilities, we might divide the population into individual altruits and egoists, then introduce a founder population of egoists into an altruistic group. If the egoists always increase in frequency within their host, yet remain a minority in the population at large, we can then tentatively declare group selection to be a force.
    So, are their any real-world examples?

  7. #7 David Sloan Wilson
    November 2, 2009

    I appreciate Shane’s impatience to see the empirical data, but let me make a few observations before we leave the haystack model.

    In general (not just group selection), the plausibility of a theory is a major issue. If something is theoretically plausible but we don’t see it in the real world, our understanding is incomplete. If it is theoretically implausible but common in the real world, our understanding is again incomplete. Hamilton’s rule (b/r-c>0) is eminently plausible. We’re gratified when it is (roughly) confirmed, which makes our theoretical understanding consistent with the real world. If group selection is as theoretically plausible as Hamilton’s rule, then it would be surprising if it could not be empirically confirmed.

    Second, if kin selection no longer counts as an alternative to group selection, then every empirical confirmation of kin selection is also a confirmation of group selection. To be precise, if you look at an empirical example of genetic relatives being altruistic to each other, you’ll find that the altruists are at a selective disadvantage within their own kin groups and evolve only by virtue of kin groups with more altruists differentially contributing to the total gene pool. What kin selection does is cluster the altruists in some groups and the non-altruists in other groups, thereby increasing the importance of group selection compared to randomly formed groups. The clustering is incomplete whenever r<1, so that every case of altruism that evolves by kin selection is a case of between-group selection trumping within-group selection. The case of r=1 represents pure group selection because there is no genetic variation within groups.

    If kin selection is a kind of group selection, then interest shifts to examples of group selection that go beyond kin selection. Are there examples of group selection when the groups are composed of non-relatives, for example? The answer to this question is "yes", as we shall see. In fact a new example will appear in the next issue of Science magazine and older examples will be presented in future installments of this series. But the situation is complicated by changing definitions r within kin selection theory. Starting with the Price equation, r became interpreted not as an index of genealogical relatedness--the original interpretation--but as a correlation between the phenotype of the individual and the phenotype of its social environment based on any mechanism. For example, if altruists have a way of clustering with each other without regard to their genealogical relatedness, then r becomes high. Once r became a generalized index of clustering, every case of group selection can be represented as a case of kin selection. Welcome to the wonderful world of equivalence (for the cognoscenti, there might be cases of non-equivalence even with the generalized interpretation of r).

    Finally, one point I made in comment 2 (above) is that seeing group selection requires a comparison between local selective differentials and global selection differentials. If you merely know that something evolves in the total population, you don't have sufficient information to evaluate whether group selection is taking place. The need to compare fitnesses locally in addition to globally applies to empirical examples as much as theoretical models. When we examine empirical examples of cooperation, altruism, public good provision, etc., with this in mind, we see that the empirical evidence for group selection is very strong.

    To summarize, the rejection of group selection required a conceptual structure in which a) group selection is theoretically implausible and b) there is a more plausible theoretical alternative. If both (a) and (b) are incorrect, then the conceptual structure collapses into a pile of dust.

    In the next installment, I will examine another concept that is widely regarded as an alternative to group selection-- the gene as the "fundamental unit of selection".

    Please be patient with my examination of conceptual structures. They are truly the lenses through which we see the real world, so it is important to get them right. We're making a lot of progress in the short space demanded by the blog format.

  8. #8 piker
    November 2, 2009

    “To summarize, the rejection of group selection required a conceptual structure in which a) group selection is theoretically implausible and b) there is a more plausible theoretical alternative. If both (a) and (b) are incorrect, then the conceptual structure collapses into a pile of dust.”

    That makes absolutely no sense as a requirement for acceptance of an inadequately tested hypothesis. It’s essentially the argument that if you can’t prove I’m completely wrong, then you can’t deny me the extent of my rightness.

    No wonder you refuse to address the questions of inherent paradox in your theory. The questions haven’t come with a conceptual structure that you, which you seem to have reserved as your right, will choose to accept as a valid alternative.

    The “progress” you claim to be making would require that you be final arbiter of the nature of its impediments.

  9. #9 daedalus2u
    November 2, 2009

    piker, what criteria do you use for rejecting a hypothesis? You can’t prove a negative. If there is no conceptual structure where group selection is theoretically implausible, and there is no more plausible alternative to group selection, by what basis is group selection to be rejected?

    DSW isn’t talking about accepting the group selection hypothesis; he is talking about not rejecting the group selection hypothesis. You don’t reject a hypothesis because you can’t figure out how to make it work, you can only reject a hypothesis when you have shown that it can’t work (i.e. is worse than implausible) or that it doesn’t work (i.e. there is another hypothesis that works better). In no way has this been done for group selection.

    This is the whole point of DSW’s series of posts, group selection was rejected for insufficient reasons and that premature rejection has become written in stone to the detriment of the field.

  10. #10 piker
    November 2, 2009

    What is being rejected are the claims of its status as factual rather than plausible. The argumentation used to establish the factual nature of the plausible is seriously and critically flawed. (Which goes to the plausibility as well in retrospect.)

    And of course you can’t prove a negative. But you’d have it by that standard there are therefor no demonstrable negatives.

  11. #11 piker
    November 2, 2009

    And daedalus2u, aren’t you that one that says it’s anthropomorphic to attribute human traits to other animals, when in fact we have a theory proposed that is instead “ant-thropomorphic” by its attribution of insect behavioral traits, as one example, to humans.
    But such traits are gene-specific only to the extent of their physiological components, which are the predominate element in insects but arguably the weakest element of behaviors in humans.

  12. #12 daedalus2u
    November 3, 2009

    piker, I am not sure you appreciate my objections to anthropomorphic metaphors. Perceiving ants to have human-like social traits, relabeling those traits as ant-like and then rerelabeling human-like social traits as ant-like isn’t what I consider to be useful progress.

    A large part of why humans perceive things in anthropomorphic terms is because that is how human brains are configured to perceive things, so as to better communicate with and understand other humans. There is no a priori reason for reality or other organisms to be configured in anthropomorphic terms. When non-human things are perceived to behave in anthropomorphic ways, we should strongly consider that that perception is a type 1 error, a false-positive due to our human tuned perception.

    I think the whole focus on “selfish genes” and “altruistic genes” is an anthropomorphic distraction. Is a particular polymorphism of the gene for cytochrome c “selfish” or “altruistic” and what experiments could be used to answer that question? Is the relative abundance of different cytochrome c genes in humans evidence of their “selfishness” or of their “altruism”?

  13. #13 piker
    November 3, 2009

    Then, daedalus2u, if you believe ‘the whole focus on “selfish genes” and “altruistic genes” is an anthropomorphic distraction,’ why are you so supportive of the group selection hypothesis that largely relies on that particular distractive focus?

  14. #14 daedalus2u
    November 3, 2009

    Why? Because for the vast majority of the genome, there is no understanding of what it does. There is a great deal of breathtaking naïveté in genetics and in the genetic basis of diseases and disorders.

    “Approximately two thirds of all knockouts of individual mouse genes give rise to viable fertile mice. These genes have thus been termed ‘non-essential’ in contrast to ‘essential’ genes, the knockouts of which result in death or infertility.”


    This is an interesting paper that shows that the rate of change of gene sequences does not correlate inversely with how “essential” a gene is.

    I think a large part of the rejection of group selection had to do with a naïve and anthropomorphic characterization of genes as “selfish” and “altruistic”. “Genes” don’t function in isolation. They need a whole genome to function, most of which is not genes. Simplistic models that look at genes in isolation are too simplistic to represent the real behavior of gene and non-gene DNA, or at least the interactions that I am most interested in, how those relate (or not) to complex genetic diseases, especially autism. I think if there was less naïve anthropomorphic thinking that the field would progress faster. I appreciate that thinking in non-anthropomorphic terms is very difficult for some people. They should make the effort anyway.

    There are genetic diseases that are transmitted by Mendelian genetics. These are relatively rare. The most common diseases, diabetes, obesity, hypertension, heart disease, autism, liver failure, Alzheimer’s are all called “complex genetic diseases” because they are transmitted familially but don’t have simple Mendelian genetics, and appear to be “caused” by at least dozens of genes. By what mechanism have such common “genetic” diseases become common?

    How is the observation of common “complex genetic diseases” compatible with the observation that two thirds of genes are non-essential? That two thirds of genes are non-essential implies enormous redundancy in the genome. That there are complex genetic diseases implies that there is little to no redundancy. I appreciate that because these observations are data, reality has to be compatible with both of them simultaneously. I think it is our naïve understanding of genomes and evolution that leads to this apparent incongruity. I think premature rejection of group selection has contributed to that apparent incongruity.

  15. #15 piker
    November 3, 2009

    Interesting paper that you cited. This is not the place to discuss all the implications. But consider that genes that appear non-essential may be redundant only in terms of their present utility. We may find that the evolutionary process involves anticipatory functions that take more of a long term view of their purposes (metaphorically speaking) than we have imagined.

  16. #16 Guy
    November 3, 2009

    I didn’t see the flaw in the haystack model that you found, and it is a big one. However, I’d like to add another one I forgot to mention in my previous post. This relates to the point somebody made about the importance of having groups composed completely of altruists. If there are many haystacks in the system, drift alone will ensure the existence of purely altruistic groups. This effect turns out to be very important in spatially explicit population genetic models.