Boyd and Richerson's group selection

In my review of Boyd and Richerson’s The Origin and Evolution of Cultures, I noted that I was not completely happy with their treatment of group selection. This post catalogues some of my thoughts.

Boyd and Richerson open their group selection discussion by noting that selection can be broken down into between-group and within-group selection (as per the Price equation – and given this equation can be developed for multiple levels, we refer to “multi-level selection”). But in their analysis of what they call group selection, they do not practically use this framework and there are no attempts to decompose the levels of selection despite the initial framing of the problem around the ability to do this. Part of the reason for the lack of decomposition between the levels is that Boyd and Richerson generally (but not always) have another conception of group selection in mind – differential reproduction and spread of groups. But that is part of what makes the initial framing deceiving.

This problem becomes apparent when they start to discuss situations where it is unclear at what level the selection is occurring. Take the following:

Payoff-biased imitation means people will preferentially copy individuals who get higher payoffs. The higher an individual’s payoff, the more likely that individual is to be imitated. If individuals have occasion to imitate people in neighboring groups, people from cooperative populations will be preferentially imitated by individuals in noncooperative populations because the average payoff to individuals from cooperative populations is much higher than the average payoff of individuals in noncooperative populations. Boyd and Richerson (2000) have shown that, under a wide range of conditions (and fairly quickly), this form of cultural group selection will deterministically spread group-beneficial behaviors from a single group (at a group-beneficial equilibrium) through a meta-population of other groups, which were previously stuck at a more individualistic equilibrium.

So, individuals copy people from more successful groups and the trait then spreads within those  groups. Is this actually group selection? Why does the trait spread in the new group – doesn’t this require individual advantage?

The paper referred to in this quote – Boyd and Richerson (2000) – is also contained in the book, and it describes a model with the spread of norms about drinking. Drinking has negative long-term consequences, but some people drink due to strong discounting. However, the presence of people with puritanical (rather than tolerant) norms can increase the cost of drinking due to social disapproval, meaning populations with puritan norms are better off as a whole than populations of tolerant people.

As people with tolerant and puritanical norms get on each other’s nerves, an isolated group’s members will tend to have the same norm. But given the lower number of drinkers in the puritan groups, the puritan groups will have the higher total payoff. Thus, if groups can mix, the puritan norms may spread as people copy the most successful individuals from other groups. Boyd and Richerson describe this as group selection, but the spread of the norms within groups after mixing demonstrates a degree of individual benefit. At what level are the dynamics dominant?

In other parts of the book, it is difficult to disentangle what the trait under group selection is. For example, when Boyd and Richerson write of the spread of ritualised cannibalism in New Guinea and the associated spread of the disease kuru, they describe the process as group selection. But is the relevant cultural trait eating kuru? Conforming to group rituals? Conforming to rituals concerning cannibalism? Which of these is being selected affects the assessment of the level of selection. Educated guesses can be made, but it is hard work.

These examples indicate a degree of looseness in Boyd and Richerson’s use of the term group selection. At times the term seems to be thrown at any dynamics that involve groups, with no clear definition of what group selection is and no attempt to place the observed behaviour in the context of the definition. This is, of course, a broader issue in much of the literature concerning group selection.

Having said this, as I mentioned in my review, I am not averse to the idea of examining cultural evolution in a group selection framework. I like many of Boyd and Richerson’s models and the descriptions of the dynamics, even if I consider the group selection label has been incorrectly applied. And it is possible that some of my complaints above could be dealt with through better explanation. But Boyd and Richerson use the term group selection so loosely that it is hard to agree with their use, particularly as I’m not sure what exactly I would be agreeing with. For the moment I prefer to describe their overall approach as “cultural group dynamics”.

4 comments

  1. Joe Henrich’s cultural group selection paper, perhaps, better describes the CGS framework and why it is a good way to think about cultural traits. http://www2.psych.ubc.ca/~henrich/Website/Papers/HenrichplusCommentsJebo2003.pdf

    That said, I do not think the B&R usage is that loose. Calling something a “CGS” model depends on the level of selection, not on the mechanism of selection. If you have a model (word model, computational model, math model, or whatever else) that decomposes selection to more than one level it is a multilevel selection model. If one of those levels is between different collections of individuals (groups) you have a group selection model. Selection on traits at this level may, mechanistically, be a result of differential group extinction, differential imitation, differential migration, some combination or whatever else. That’s up to the modeler and what makes sense for the types of groups and traits in question.

    N-person games and conformist transmission (which are important for understanding humans) are examples of things that are generally easier to model in a multilevel selection framework than with inclusive fitness accounting.

    For the New Guinea example, eating kuru is likely the relevant cultural trait.

    1. I like Henrich’s paper, but it demonstrates my point. In that paper, Henrich goes through the process of analysing and breaking down the levels of selection (acting in opposite directions) on a trait, whereas Boyd and Richerson don’t do that in the book. They show it can be done through the Price equation, but then describe various models as group selection without carrying out the exercise. If they did decompose the levels of selection, in many cases we would find plenty of action at the individual level (such as for eating kuru, if that is the relevant cultural trait) and the story would become a true multilevel selection analysis.

      You can’t take from their work the lesson that a multilevel selection framework provides an easier modelling environment than inclusive fitness accounting, as they don’t use such a framework. Ultimately, I am not so much complaining about the models, but rather their use of a label which their analysis does not justify.

  2. I agree with most of what you say, but I don’t think labels are the only problem with the idea. In a recent conference I attended on this topic (in Knoxville) CGS was described as an umbrella that covered quite a few models (at least 5 different classes). This was a reply to a remark I made that speakers throughout the conference often talked about *very* different processes that were all being classified as CGS. A similar feeling can arise from looking at the literature. My feeling is that CGS can certainly do with more (mathematical) rigor in order to avoid the semantics and focus on the core issues.

Comments are closed.