Prestige hierarchies

I wrote earlier about a paper by Eduardo Ottoni and colleagues examining social learning and expertise in nutcracking capuchins. That paper referenced another paper by Joseph Henrich and Francisco Gil-White introducing and discussing the relevance of "prestige hierarchies" in the context of the evolution of culture.

In the introduction, Henrich and Gil-White contrast the behaviors that maintain dominance hierarchies in mammals from those that maintain prestige hierarchies in humans:

Although nonhuman status is still poorly understood, a single process appears at least strongly predominant: agonism (aggression, intimidation, violence, etc. -- that is, force or force threat) The resulting social asymmetries are referred to as "dominance hierarchies" in the ethological and behavioral ecology literatures. The privileges that accrue to dominant individuals are (1) in males, preferential reproductive access to females, food, and spaces, as well as a disproportionate amount of grooming from others; (2) in females, preferential access to food and spaces, and disproportionate grooming. Despite some controversy, the evidence suggests that dominance correlates with fitness (Cowlishaw and Ellis). The stability of dominance is often reinforced through "reminders": submissive behaviors (e.g., grooming, submissive displays, yielding space, etc.) from subordinate to superior, whether or not induced through intimidation by the latter.
In humans, in contrast, status and its perquisites often come from nonagonistic sources -- in particular, from excellence in valued domains of activity, even without any credible claim to superior force. For example, paraplegic physicist Stephen Hawking -- widely regarded as Einstein's heir, and current occupant of Newton's chair at Cambridge University -- certainly enjoys very high status throughout the world. Those who, like Hawking, achieve status by excelling in valued domains are often said to have "prestige."

In human societies, both kinds of hierarchies may be in evidence:

In the Amazon, several researchers have observed two avenues to status and leadership in small-scale societies: "force" and "persuasion" (Krackle, 1978). "Forceful" leaders are domineering headmen who maintain their position through fear, threat, and compulsion (see also Maybury and Maybury). "Persuasive" leaders depend on their influence and the consent of their followers and lack the force to obligate (see also Arvelo de Jimnez, 1971; Clastres; Goldman; Huxley and Levi). These two styles of leadership, involving either persuasion or force, correspond to our two types of status: prestige and dominance.

The connection of this paper to the capuchin learning paper is the idea that learning can be made more efficient by preferentially copying individuals who have "better-than-average information." The argument is that selection should favor an ability to assess who has information relevant to fitness, and to copy the behavior of individuals who have this information.

Culture, information, and behavior

I am a bit hesitant about the form of this argument, because in a sense it deliberately confuses two distinct bases for the definition of culture: culture as information and culture as behavior.

Individuals can have information, but this information cannot be observed directly -- at least not without a mind-reading device. Information can only be inferred from its effects on behavior. But those effects are indirect and possibly inconsistent -- an individual need not behave in ways that optimally reflect the information the individual has.

For humans, at least it is possible to directly observe certain kinds of information transfer. In particular, we can keep track of what somebody is told, or we can ask subjects to tell us what they observed. But even this is imperfect -- an individual may not have ostensive knowledge about what he or she has seen, or what aspects of an observation are important and will ultimately shape his or her own behavior. For nonhuman animals, there is no direct way to poll information at all; indirect polls are possible only in experimental settings, and wild animals are observed mainly by way of their natural behaviors.

So operationally, many ethologists define culture as transmitted behaviors. Now, this is problematic because there is actually no way to "transmit" a behavior without transmitting information about the behavior. And there are many aspects of human culture that are purely informational and have only the most indirect of links to any behaviors.

For instance, most readers of this sentence can clearly understand and remember its information content, but it is rather doubtful that the sentence will alter the behavior of most of its readers in any perceivable way.

I guess the bottom line of this problem is this: If individuals have information about a prestige hierarchy, then they must infer the presence of information that contributes to fitness-enhancing behaviors from the observation of valued behaviors. If an individual could simply alter his or her behavior pattern to obtain the valuable information, then there would be no need to maintain a prestige hierarchy -- everyone would quickly share the valuable information. So a prestige hierarchy presupposes that:

(a) Information is time-consuming to obtain by observation. If information were trivial to obtain by observation, then there would be no need to keep track of who had the information. Consider students cheating on a test. If there were no disincentive to cheat – that is, if they could just ask the entire class when they didn’t know an answer, or if they could browse a large sample of exams, then there would be no need to sit next to a student likely to know the answer. The difficulty in the exam is the proctor, who tries to prevent information transfer. The difficulty in most circumstances in nature will be that observation is time-consuming. </p>

(b) Information is even more difficult to obtain without observation. If there were a way to easily find the right answer to exam questions by guessing, no one would ever cheat. In natural settings, the problem will be that behaviors are highly context-dependent and may require careful sequencing or orienting substrates to work. </p>

(c) An individual has the opportunity to observe prestigious individuals. Again, there is the exam proctor. But moreover, students who know the right answers on an exam may closely monitor those around them to make sure there is no cheating off them – otherwise, the curve may be blown! In natural settings, the problem will be aggression – it may be impossible to approach other individuals for a sufficient time to observe their behavior. </p>

Henrich and Gil-Stein note this opportunity as the most important difference between dominance and prestige hierarchies. In particular, they note that selection should favor the evolution of "deference":

Cultural transmission is adaptive because it saves learners the costs of individual learning. Once some cultural transmission capacities exist, natural selection favors improved learning efficiencies, such as abilities to identify and preferentially copy models who are likely to possess better-than-average information. Moreover, selection will favor behaviors in the learner that lead to better learning environments, e.g., gaining greater frequency and intimacy of interaction with the model, plus his/her cooperation. Copiers thus evolve to provide all sorts of benefits (i.e., "deference") to targeted models in order to induce preferred models to grant greater access and cooperation. Such preferred models may be said to have prestige with respect to their "clients" (the copiers).

On the one hand, this makes a lot of sense. Sure, if you want access to another individual, you should defer to them -- otherwise, she may be annoyed by you and drive you away.

On the other hand, the "all sorts of benefits" seems pretty limited in practice. The capuchins don't give anything at all for the chance to learn. Children give deference to parents, but it's limited, and partly enforced by dominance mechanisms. The authors go this far:

In order to gain this kind of preferential access, infocopiers become valuable interactants by "kissing up." Infocopiers have evolved to do all sorts of things that models were already adapted to like or seek in potential interactants, such as being especially trustworthy, offering all sorts of help without expecting anything in return, deferring to the model's judgment, being nice and helpful to the model's children, exempting the model from certain obligations vis-à-vis the copier, etc.

I'm on the fence about this. The "benefits" conferred on a good teacher ought to be quite a bit lower than those given as a matter of course to genetic relatives, so maybe these kinds of activities are exactly in that range.

(d) The information can be learned. Since nearly all information might be possible to learn, another way to say this is that the valued commodity is really information as opposed to something else. </p>


The predictions of the model are interesting. There is this:

The distribution of deference is a reliable and honest signal of relative model worth because such signals are costly to fake. Sycophants cannot deceive their competitors by deferring to someone they would rather not copy without increasing their total deference costs and losing some access to their preferred models. Sycophants also cannot easily conceal deference directed to the desired model, for this entails a bias for private deference and therefore a reduction in total deference, and hence in less access. Moreover, models should prefer public displays of deference in order to broadcast their prestige and attract more clients.

This leads to the hypothesis that "infocopiers" should seek out models that have large prestige followings, and thereby high status.


Some further implications of our theory arise from considering the coevolution between copiers and models. In prestige, clients choose whom they defer to, so a kind of "market" results. Like "firms," models compete for "customers" (the copiers) who shop around for the best deal. Models should be sensitive to how "profit curves" change with added clients, for these "firms" can have too many customers. A hunter's fitness initially increases as more clients raise total deference. So the hunter may prefer having 3 sycophants to 1, but would he want 20? Large hunting parties may scare off potential prey. Thus, good hunters should raise the cost of access by acting more arrogantly as clientele size approaches the optimum. On the other hand, if no practical limit on optimal clientele size exists (e.g., great storytellers), or if means other than arrogance will limit clientele-size (e.g., bodyguards), then increases in arrogance should not accompany growing prestige. Alternatively, if one's benefits do not come directly and primarily from client deference, the prestigious may learn that arrogance is not too costly (e.g., some sports stars). Finally, models should prefer above-average learners because they advertise the model's quality and provide a potential source of valuable information.
For clients, the benefits of access diminish rapidly with increasing clientele size. Competing with more clients may mean less individual attention from the model, so copiers may prefer less popular, lower-quality models with cheaper prices of access.

This does raise the question of whether "prestige" can be distinguished from "entertainment value". We may listen to great storytellers, but that doesn't mean that we want to be great storytellers. Is our listening a passive side effect of our ability to pick out models for learning?

If it isn't, then that would imply that prestige hierarchies may be maintained for other reasons besides learning value. These may range from the nebulous (like "cultural cohesion") to the particular (like "beer-selling ability"). Considering that social interactions themselves require some skill to navigate, and we learn to interact by watching other people interact, this has many possible levels of complexity.

In any event, there are lots of predictions and descriptions of possible complexities in the paper that provide much food for thought when thinking in terms of human social interactions relevant to learning and experience.


Henrich J, Gil-White FJ. 2001. The evolution of prestige: freely conferred deference as a mechanism for enhancing the benefits of cultural transmission. Evol Hum Behav 22:165-196. DOI link

Ottoni EB, de Resende BD, Izar P. 2005. Watching the best nutcrackers: What capuchin monkeys (Cebus apella) know about others' tool-using abilities. Anim Cogn 24:215-219. DOI link