What does not kill the group, makes it stronger!

I recently finished reading The Faith Instinct: How Religion Evolved and Why It Endures, a new book by Nicholas Wade, a science writer for The New York Times. Before giving it the "full treatment" I thought it behooved me to revisit some of the scientific literature which Wade relies upon to give form to his argument. One of the pillars of The Faith Instinct is group selection, and one of the scholars who Wade specifically cites is the economist Samuel Bowles. Bowles was an author on a paper I reviewed earlier this week, on the empirical assessment of the extent of heritability of wealth across generations in various societies. But in this case what is more relevant was a paper that Bowles published in Science last spring, Did Warfare Among Ancestral Hunter-Gatherers Affect the Evolution of Human Social Behaviors?:

Since Darwin, intergroup hostilities have figured prominently in explanations of the evolution of human social behavior. Yet whether ancestral humans were largely "peaceful" or "warlike" remains controversial. I ask a more precise question: If more cooperative groups were more likely to prevail in conflicts with other groups, was the level of intergroup violence sufficient to influence the evolution of human social behavior? Using a model of the evolutionary impact of between-group competition and a new data set that combines archaeological evidence on causes of death during the Late Pleistocene and early Holocene with ethnographic and historical reports on hunter-gatherer populations, I find that the estimated level of mortality in intergroup conflicts would have had substantial effects, allowing the proliferation of group-beneficial behaviors that were quite costly to the individual altruist.

The thesis presented in this paper is essential to The Faith Instinct because a central assertion in that book is that religious behavior emerged as an evolved trait in the context of intergroup competition. In short, war among hunter-gatherers. But let us set aside the relatively controversial issue of the evolution of religion, and focus on group selection among humans, and its possible role in the emergence of the trait of altruism.

David Sloan Wilson, who just recently joined ScienceBlogs, has written extensively on the topic of group selection (or, multilevel selection, though I will use the term group selection because that is what Bowles uses). It is a controversial topic in the context of evolutionary biology. The short sketch is that naive variants of group selection were in vogue until the 1960s, when a new wave of theorists tore down its theoretical basis and offered up alternative processes to explain social behavior. In more recent years David Sloan Wilson has been leading a campaign to return group selection, or more generally multilevel selection, to respectability. Others have always "kept the faith," but only begun to speak up more forcefully in its defense recently. Why only recently? In part because skeptics of selection above the level of the individual have always assented to its theoretical possibility, but diminished the likelihood of its realized probability. With the fleshing out of some empirical data which makes that probability more than speculation, the theoretical window for group selection may translate into something more substantive.

Bowles' paper has two broad sections, theory and results. The theory condenses and formalizes an important verbal concept: the heritable occurrence of altruism which decreases individual fitness can flourish as long as fitness differentials between groups are proportional to the frequency of altruism within groups. Got that? Perhaps you can say it in words concisely, but the algebra makes it more precise. On to the model.

κ = probability per generation that a group will enter into conflict with another group
λ = probability of surviving that conflict

In this model there are two outcomes: extinction and survival. If a group goes extinct its numbers naturally go to 0, but if it survives its numbers double. At which point the group experiences fission so that a subsequent generation produces two groups whose numbers are normalized again at 1. Importantly for this model &lambda often varies by the proportion of altruists in the group.

λ_A = probability of surviving conflict contingent upon proportion of altruists

Of course, not all groups enter into conflict. So:

1 - κ= probability of not entering into conflict per generation

Your expectation of the size of a group, j, in a given generation, is the probability of the likelihood of conflict, and the product of surviving that conflict conditional upon its occurrence. If the size of the group is normalized at 1 at time t, then at time t + 1 it would be:

w_j= (1 - κ) + κλ + &kappa(1 - λ)

rearrange:

w_j= 1 - κ + 2κλ

We are at this point treating groups as the atomic units of the process of extinction and reproduction. This isn't that mind-blowing, as a multicellular organism you're also a collection of groups which can be first approximated to a very high likelihood as a unit upon which evolutionarily relevant operations occur. Let's move to the level of the individual, which is more relevant in the case of human social groups than in multicellular organisms (unless you have cancers and such).

p_ij = 1 if individual i in group j is an altruist
p_ij = 0 if individual i in group j is an altruist
p_j = proportion group j's membership that are altruists

Now let's rewrite this into the Price Equation:

Δp = var(p_j)β_G + E{var(p_ij)}β_i

So, the change in proportion of altruists across the whole population equals the variance of the proportion of altruists across groups times β_G plus the summed variances of individuals times β_i. Which begs the question what the β's are.

β_G = κ2&lambda_A, basically group fitness, as λ_A ~ probability of victory in conflict proportional to altruist frequency

β_i = -c + κ2&lambda_A/n, individual fitness, where -c equals cost to altruists for being an altruist, and the second part shows the individual's increase in fitness from the group's increase in fitness (this is why you divide by n, the number of individuals). Note there are some assumptions here: no reproductive skew, no non-random migration, and, no fluctuation in population size. Also, remember that the groups are at parity of size.

Now, to figuring out how this relates to evolutionary genetics, you need to look at measures of genetic variance. For this, Bowles uses F statistics, which basically partition the variance between and within population structures. So, F_ST shows the proportion of total variance which can be explained between populations. For Δp to be positive, that is, for the frequency of altruism to increase due to group selection, there needs to be between group genetic variance. This goes back to the Price Equation, which was rewritten above. Δp > 0 where:

F_ST â(1 â F_ST) > âβ_i âβ_G

On the left-side you see the between group genetic variance divided by within group genetic variance. In a normal human population this should be a positive value less than 1 (e.g., F_ST on the racial level in humans is 0.15, so the ratio would be ~0.18). On the ride-side β_i includes the costs to the altruist (numerator) and β_G (denominator). Remember, within groups altruists invariably have lower fitness than non-altruists. That is one primary reason that group selection is viewed with skepticism. It can spread only if the presence of altruists within groups is so beneficial that on the whole selfishness is diminished.

A quick illustration will help.

On an individual level the selfish gain at the expense of altruists when in the same groups. But, groups which have altruists do much better than groups without. Though the proportion of altruists in deme 1 has decreased, as predicted by the reality that altruism often does not pay on the individual level, the flourishing of deme 1 at the expense of deme 2 is such as that altruism does pay when viewed in the aggregate. This toy example is neither necessary nor sufficient to convince anyone that group selection does happen, but it shows the sort of processes which might allow it to occur.

When benefits to groups are high enough due to the presence of altruism, and when genetic differences between groups are large enough, then altruism may emerge due to group selection. Breaking apart the above equation into its constituent parts (the β's), Bowles constructs this equation which illustrates the maximum cost which altruism can extract from an individual all other conditions held equal for it to persist:

c* = κ2λ_A {FST â(1 â F_ST) + 1/n}

It is clear that as the numerators increase (more intergroup violence, more benefits from altruists to the whole group, more between group genetic difference) the altruist can incur more cost and yet remain evolutionarily viable. Figure 2b shows "curves" for groups given particular wartime mortality rates, δ, and probability of success in conflict conditional upon the gains due to altruism, λ_A, with c fixed at 0.03. The F_ST are from real groups draw from data:

The area above and to the right of the curve is viable for the spread of altruism. It fits our intuition, more between group violence, more between group genetic difference, and/or more benefits accrued from altruism, result in the perpetuation of altruism among human groups.

There was some empirical data to test. To the left are results generated using a data set drawn from aboriginal populations of Arnhem Land. Note that the c* values are modest indeed. I'll let Bowles conclude:

The evidence that intergroup conflict may have contributed significantly to the proliferation of a genetic predisposition to behave altruistically does not mean that it did, or that the mechanism I have described explains the evolution of human altruism. The model applies with even greater force to behaviors transmitted culturally rather than genetically, in part because between-group differentiation is considerably greater and hence the evolutionary impact of differential group success in contests is stronger.

One cannot say with certainty which of these data should be the basis for our conclusions concerning the evolutionary impact of lethal intergroup competition during the Late Pleistocene and early Holocene. Even though periods of climatic volatility would bring even quite distant groups into contact during migrations, the farflung settlements of the circumpolar regions, desert Southern Africa, and Western Australia would be far less likely to be in contact--either conflictual or beneficial--than groups living in closer proximity such as those in coastal Arnhem Land and lowland New Guinea. Moreover, the more populated coastal and riverine areas contributed disproportionately to the gene pool of subsequent generations. But taking all of the evidence into account, it seems likely that, for many groups and for substantial periods of human prehistory, lethal group conflict may have been frequent enough to support the proliferation of quite costly forms of altruism.

I put the highlight on this paper because it tries to plug some real data into the model. Too often group selection models are theory. Interesting, plausible, but how probable? In trying to understand human behavior evolutionary biologists are wont to appeal to reciprocal altruism or inclusive fitness, or if nothing else is at hand, the deus ex machina of free will, choice and culture. It seems that the time is appropriate to begin considering the other options, though with caution and attention to the problems which pushing into less parsimonious theoretical territory inevitably entails. Additionally, this paper focuses on group selection as biology. I am modestly skeptical of this primarily because I do not believe in total group extinction.

The barbarian says that the best in life is "to crush your enemies, to see them fall at your feet -- to take their horses and goods and hear the lamentation of their women." Not to crush their women. Conquered groups are genetically assimilated. This reduces between group variance between neighbors. Bowles may have found something in Arnhem Land, but I would like to see more diverse granular analyses of between population variance. It isn't as if human populations engage in 1,000 mile treks to battle other groups, rather, the enemies are quite often genetic cousins. The exception to this rule is likely on the linguistic frontier, but how large are these frontiers? True, in some regions such as the highlands of Papua New Guinea the frontiers are expansive due to linguistic fragmentation, but this might be a function of the local topography and ecology. Rather, I would look to cultural group selection, because there are many cases of women being assimilated into a dominant culture, and their offspring speaking the language, and expressing the values, in totality of their fathers. One inherits 50% of one's genes from one's mother and one's father, but inheritance of cultural traits which are distinctive between parents may show very strong biases. Partitioning variance between and within groups on cultural traits often shows far greater between group differences; consider variance in speech, within a tribe there are slight variations, but between tribes the accent variation may be strong enough to accurately assign any individual to the correct tribe by speech alone.

A final more general point may be the dominance of transience and flux in human cultural dynamics. There may never be very many long term equilibria states in our societies, and the adaptive landscape of behavioral fitness may shift very quickly, quite often because of the feedback loops between culture and the adaptive landscape itself. The history of China shows successive periods of dynastic genesis, periods of cohesive polity expansion and consolidation, and then the slow long wind down toward collapse and dissolution ushering in an age of chaos. At which point, the process recapitulates. When talking about the relevance of group selection, inclusive fitness, reciprocal altruism, etc., to human behavior I think the reality of the flux and its ever changing demands may prove elegant solutions to be seductive deceptions untrue to the world as it is.

Citation: Did Warfare Among Ancestral Hunter-Gatherers Affect the Evolution of Human Social Behaviors? Samuel Bowles (5 June 2009), Science 324 (5932), 1293. [DOI: 10.1126/science.1168112]