The evolution of social cohesion

From Santa Fe Institute Events Wiki


  • Andrew Bell
  • Simon Angus
  • Paul Hooper
  • Rafal Raciborski
  • Elizabeth Mullane

Close friends

  • Will Ludington
  • Alejandro Balbin
  • Ryan Chisholm


We often study the effect of certain social institutions, and sometimes look at transitions, but what effect does passed-on cultural institutions play in the evolution and formation of cohesive social institutions?

Zooming in: Economists and other practitioners of game theory generally represent variation in insitutions as different modifications of the payoff structure of a specific game. They have been successful in understanding the impact of different institutions on equilbrium game behavior when institutions are imposed exogenously by nature, the gods, or the central planner. But there have only been a few early attempts at modeling the endogenous evolution of institutions guided by the motivated actions of the agents themselves. We propose to do just this with a specific problem (e.g. commons management), hoping to develop a formal framework that may be then generalizable to other problems.


Suppose that a set of agents have choices regarding both how to behave toward each other in the current period, as well as the institutions under which they carry out this behavior.

This could occur in an alternating fashion: within an institutional period players how to behave towards each other; between institutional periods players decide which institution(s) to pass on to the next generation. There may be other ways to go about this.

Research Questions follow:

  • is there a stable (long-run) social institution that is selected?
  • does memory (cultural, historical, heritage) affects agents' long-term decisions about social institutions?
  • does this institutional structure have a path-dependance (i.e. must institutaion A then B then C preceed the selection and stabilisation of institution X)?
  • do agents operate heterogeneosly within a period (e.g. old agents who have cultural knnowledge prefer institution X, but young agents, with shorter memories, or trust in passed-on heritage select institution Y)?
  • what scaling? do small vs. large population affect these decisions?
  • what about a two-population model? do we see group selection occuring to promote a certain institution in both camps? or is there a stable complimentary institutional framework (e.g. E. vs. W. Germany)?


  • Keep things simple
  • construct a simple interaction game for the agents, causing the institutions to be also simple
    • (leave out voting methodologies, political interest etc.)

Brain-storming model structure

model 1

  • a standard hunter-gatherer scenario under scarcity
  • agents face a decision problem whether to cooperate the hunt or to act individually (stag-hunt style)
  • for the coalitions: the success of any grouping is proportional to the number in the coalition (due to division of labour within the coalition, trust-based hunting methods)
  • for the individuals: a minority game design (we are hunting) could lead to preferable outcomes (I didn't hunt where the large, noisy group hunted?)
  • institutions:
    • a 'good' institution for cooperation: a transfer system (public good provision?) to all members of society, supports free-riders in the short-term, but does this survive in the long-run (over successive generations, since some agents remember when everyone had lack)
    • a targeted (progressive) taxation system: a transfer system specifically from the well-off top half to the bottom half (ranked in terms of reward from expended gathering effort)
    • no transfer system: you get (only) what you work for, no other transfers or public good provision
    • a consumption taxation regime: distribution based on discretionary welfare (so we would need a metabolism system for basic needs, and then a discretionary part for 'leisure')
    • public good provision through taxation: but only to the 'cultural memory fund' -- i.e. for libraries, books, plays, histories, langauge etc. that capture things about how times were in previous set-ups


  • memory of each individual (how many previous instutional arrangements they remember)
  • birth/death rates (how fast we turn over the population ('physical' memory)

model 2

(this is possibly just a different discussion of the same model)

  • a small set of m state variables - water, guns, food, whatever
  • a set of n institutions that govern these state variables, of the norm and rule format (making this set is a tricky bit)
    • n > m so that there is overlap/conflict among different institutions for governance of the same state variable
  • a population of agents that are guided by a subset of these institutions (this subset is possibly parameterized by "memory")
  • stressing events on the state variable that draw out different institutions
  • some kind of network structure among agents to govern their interactions

The question over time then might be what makes different belief systems (which here are subsets of the institutional set shared by groups of agents) stable or persistent over time.

What is obviously missing from this summary is a description of what the institutions are, what the set needs to include, etc., and i think this is informed partly by some of the other discussion that simon and rafal have posted.

is this structurally too complicated?

Notes and thoughts

Paul: Andrew introduced Ostrom's typology of institutional norms--mays, musts, and must nots--where each directive is accompanied by a promise of reward or punishment for compliance or non-compliance. 'Good' institutions presumably alter the fitness landscape of individual players to be more compatable with socially desirable outcomes (contribute to the public good, don't shirk, engage in low-cost or pro-social rather than disruptive forms of status competition, etc.). I wonder if the institutions that the players choose/develop/vote on between periods could be formulated using this framework, where the institution pairs a punishment/reward to any given individual behavior. The form of the punishment/reward would have to be specified by several variables.

  • Let's say the players are considering a punishment for not contributing to a public good. Is the size of the punishment inversely proportional to the amount contributed? What is the shape of that function? Or is there a fixed fine for contributions below a certain level?

Any collective punishment or reward system will require resources for enforcement.

  • If players institute a punishment for shriking on a public good, would each be willing to contribute resources to fund the police that enforce it?

The source of new institutions:

  • Can we somewhat randomly generate institutions, and see whether they're taken up by the players? Some institutions would be dumb (e.g. the more you contribute to the public good, the more you are punished), and others favorable compared to the original intitutionless setting.
  • OR, because there are so many ways an institution could be specified, we could generate a fixed number of institutions that we introduce and allow the players to consider.

Rafal: Is it possible for a "bad" institution to thrive even if the majority of the population prefer to abolish it? Take, for example, the norm of corruption. If an agent breaks a "good" norm and is caught, he receives punishment (P1). However, the agent can propose a bribe to avoid P1. There is some probability that the bribe will be accepted or if not, a harsher punishment (P2) will be administered. It would be nice to show that under certain conditions, there may be some stable population of agents that always proposes a bribe. However, if too many agents propose a bribe, the "good" institution ceases to exist. That could cast some light on why corruption persists in some counties. A related question would be how corruption arises in the first place. Are some "good" institutions more conductive to the emergence of corruption than others? For example, if the government tolerates free riding on a good norm for a while, over time agents may internalize that norm and cooperate because this is the right thing to do. However, if P1 is harsh from the very start, it may encourage corruption because the difference between P1 and P2 will be small so it pays to offer a bribe. Thus, ironically, newly-created good institutions with a strong enforcement mechanism may be self-destructive.

Andrew: Something that is cool to think about with Rafal's bribe ideas is the difference between a system with an exogenous group (like the government) giving the punishment, versus self-governing groups where accountability and punishment arise from the agents sharing a belief in the norm itself. it would be interesting to look at whether in a self-governing system, a "good" norm and a bribe norm can both be stable, like Rafal is discussing.

Paul: I definitely favor throwing out the government and allowing the agents to endogenously determine their own institutions. An amazing transition to show would be when the agents elect to create a government, and this might ultimately be within the scope of a model of this type.

One important issue that is not yet clear to me is how new institutions should 'arrive'. Does it have to invade, with agents adopting it one at a time, or does it get applied to everyone all at once (maybe after some collective voting process)?

I think we might want to start (like Ostrom does with the PD) with a specific game (the institutionless starting point), and think through how we could formally operationalize a few institutions that might change the payoff structure of the base game. [Andrew proposed a commons-management game, which I think is a great idea.] We can mine the existing literature for possible insitutions, and translate them into the same language (define their place in them in multi-dimensional institution-space).

After we've got this down, we can see whether novel institutions could be generated and introduced, maybe using random assignment of the key variables that define the institution. Once we've done with with one base game, maybe we could do it with another, and maybe, just maybe, be able to throw nearly any base game into the model and see what institutions agents come up with.

Institutions must meet some plausibility constraints: e.g. enforcement must be financed in some way, agents cannot simply choose to pump up the subjective rewards (delta oi in Crawford-Ostrom grammar) for any old behavior. Defining the form of these (exogenous) contraints will be a key contribution of the model.


Axelrod 1986 An evolutionary approach to norms

Crawford and Ostrom 1995 A Grammar of Institutions

This is the reading I mentioned that might be a good, consistent framework to codify institutions (Andrew)

Paul: I really favor this approach. It gives us a basic structure by which we can associate rewards/punishments with specific actions undertaken by the agents.

Sabatier 1991 Toward better theories of the policy process

A quick review of some influential theories on how different group and individual actors, combine with different events/stresses to bring issues to the policy agenda

Finnemore and Sikkink 1998 International Norm Dynamics and Political Change

This article is about the emergence of norms on an international level so no need to read it closely. However, see pp.895-896 and 901-902 on the life cycle of a norm. Bottom line: It is hard to create/promote a norm but once a norm reaches a tipping point, it cascades through society. We could apply it to the domestic level: Once a certain number of agents adopt a norm, the norm is automatically adopted by the remaining agents.

Berkes 2006 From Community-Based Resource Management to Complex Systems: The Scale Issue and Marine Commons

A contrast of a few case studies in local marine fisheries as well as international fisheries. One of the points made is that local commons management and international commons management are fundamentally different - at the local level we can expect norms to emerge that lead to stable management, whereas at the international level the added complexity leads us to expect resilience over regime change, rather than stability

Ostrom 2000 Collective Action and the Evolution of Social norms

At the risk of being Ostrom-heavy, this has a ton of gems in it, including an 8-point scheme for effective self-governance systems

Cardenas and Ostrom 2004 What do people bring into the game? Experiments in the field about cooperation in the commons

Some field experiments highlighting some of the theoretical points in Ostrom 2000

Singleton 1999 Commons Problems, Collective Action and Efficiency: Past and Present Institutions of Governance in Pacific Northwest Salmon Fisheries

A contrast between pre-contact first nations management of salmon fisheries and modern state-first nations co-management, from an institutions perspective

Faysse 2005 Coping with the Tragedy of the Commons: Game Structure and Design of Rules

A review paper that outlines prior CPR studies and presents room for future research (and we are covering most of those recommendations!).

Apesteguia 2006 Does Information Matter in the Commons? This paper presents experimental evidence to determine if there is a disjunct between CPR games with known payoffs and real world CPR situations with imprecise payoff data.

Andrew: I've ordered an older (1987) book by Taylor that Ostrom cites as exploring the different kinds of games that can be applied to collective action problems: [The Possibility of Cooperation ]. I think it might help us think about the structure of the n-player game occurring in the resource arena

Paul: The following is a scrap from a somewhat recent paper by folks from my group. It's a complicated case to consider representing more formally, but gives some contextual insight into how this stuff might play out in the real world:

Imagine the following scenario. A woman returns from collecting berries and pounding palm fiber with a bawling infant. A wingless wasp stung her baby while she had put him down to pound the fiber, and is in great pain. She is frustrated and says to the other women in camp, “This is crazy for me to go out and pound fiber when I have such a young baby. I would gladly work twice as hard when he is a little older if I could concentrate on watching him now.” A few days later when the wound is infected and the child has a fever, another woman, remembering a similar incident she experienced a few years ago, says, “You know, Singing Deer is right. We should work hard when we have no baby on the breast and allow those with a young one to care for it well.” Another woman, who has had no child in the last 10 years, says, “Why should we work to feed other people’s babies? If you have a baby, you must feed it.” Other men and women consider their own situation and the situation of their children and present their opinions. Eventually a consensus (or at least, an agreement) is reached, with those in the minority either agreeing to go along with the new norm or leaving to live with less foolish people. However, one woman, who is not nursing, hardly pounds fiber at all. Other women begin to gossip about her, remarking upon how lazy she is, because she has no child to care for. She notices that the shares she receives in food distributions start to become less generous and begins to suspect that others are talking about her behind her back. She leaves and pounds a large quantity of fiber, which she shares generously. She can feel the warmth return and has learned her lesson.

We consider another similar scenario. A fifty year old man exclaims, “Look at these lazy young men! They come back to camp at mid-day and play around. Here I am, and here you and you are with lots of children to feed and no food to give them. What will those boys do when they have big families to feed.” An age-mate agrees, adding “How do I know if that lazy one is good enough for my daughter? How do I know if he will get enough food to keep her children healthy? He should come to my fire and bring me lots of meat, then I will know.” The young men are not so enthusiastic, because they do not like hunting all day long, but they are reluctant to anger the men whose daughters they favor. One young man, who is a good hunter for his age, thinking that he could take advantage of such a system, starts to hunt longer hours, giving the older men generous shares. The other young men, afraid of being outdone, also begin to hunt longer hours, sharing the fruits of their labor.

While admittedly hackneyed, these scenarios are meant to reflect the ongoing discussions and commentaries about sharing, work effort and laziness that are so pervasive in foraging societies. We do not mean to suggest that all social norms are explicitly negotiated with words or that norms solidify over a short period as a result of a few conversations. In some circumstances, lack of compliance and ‘voting with one’s feet’ are almost surely involved in those negotiations. In fact, we know virtually nothing about how standards for appropriate behavior emerge and change in small-scale societies without official means of enforcement. It is likely that majority-rule voting arrangements are not adhered to, in the strict sense, since some individuals exercise undue influence [e.g. kombeti among Aka, kapita among Efe (Hewlett and Walker 1990), Mbuti, chiefs among Yuqui (Stearman 1989)]. Nevertheless, we propose that such multi-individual negotiations, partly verbal and partly nonverbal, do result in social norms and that the weight of opinion, based upon the individual costs and benefits of norms in given contexts, determines accepted patterns of behavior. [1]

Focused Research Question

Andrew: My take is this - given ecological constraints, what types of conditions seem to lead groups to better outcomes?

we will construct a model that treats agents as a set of binary traits. each trait has its own payoff, which in turn can be tied to other traits as well as the number of agents using them. we can interpret the traits as behaviors, and the set of traits with the highest overall payoff can be thought of as the best use of resources (or something like that). we can look at what causes particular behaviors to be stable over time, etc.

this model requires a basic structure to represent the ecology and the payoffs, and then some other structure on top in order to represent interactions, disturbances, etc.

im attaching a short m-file that i think captures the spirit of the basic structure, if not the detail. the 'rules' i have constructed are more designed to be elegant in code than useful. the interactions i have set up have no real structure. thusly, the model doesnt have much in the way of interesting dynamics. in it, each agent has a particular payoff that is calculated for their set of traits. they each pick another agent at random and if the other agent has a higher payoff, they adopt a few traits at random. the model converges nicely and quickly to an optimum. if you make the slight modification so that agents exchange traits REGARDLESS of the payoff, you get a nice random walk in the aggregate payoff. basically, a more meaningful model would put us somewhere between these two. at any rate, i thought it might be nice to get something down in code as a conversation piece.

you will have to rename this file to 'cohesion.m'; the wiki was being a wanker about the .m extension -- First Try

Paul: Serious applause, looking forward to seeing this play out tomorrow.

So what is it our model could achieve? This is less clear-cut for me than it was a week ago.

Our new direction: create a general framework for the evolution of traits whose effect on individual fitness/utility is determined by their presence or absence in others in the population.

We had discussed the possibility of having 2 matrices that define the payoff for having a particular trait. The first says whether the payoff for a trait is positively or negatively (or neutrally) affected by the presence of some other trait within the same individual. The second says whether the payoff of that trait is affected by a greater frequency of that same trait--or other traits--in the population as a whole.

Now how to inject content into interactions besides learning from each other? Here's an idea: we change the second matrix, so that instead of making trait-specific payoffs dependent on population-level trait frequencies, we make them dependent on the presence or absence of those traits only in an individual's interaction partner(s). The model then makes individuals in the population interact in pairs, or sub-groups of some size, and calculates their payoffs for that turn. Whom they interact with next turn may be the same partner, it may be someone else, all determined by the structure of the network they're playing in (the rules of which we get to play with). This allows the structure of interaction to be important: if we were to assume that the payoffs for a trait or dependent on population-level frequencies, we'd basically be assuming random assortment (as most evolutionary game theory models do), and that's frankly no fun at all.

Thinking of an example setup, pretend the context is cooperation (ultimately though, this framework lets us abstract from a specific context). There is a population of 25 people. Folks are thrown into groups of 5 (by some assortment process which we define), and each of those players' payoffs is determined by her own traits and the traits of the 4 other group members. There are traits A through G:

A. 1 = fish responsibly, 0 = overfish B. 1 = sanction those whom you know overfished, 0 = don't C. 1 = monitor the behavior of others, 0 = don't D. 1 = wear good glasses, 0 = don't E. 1 = share what you observe monitoring with other players, 0 = don't F. 1 = punish those who did not sanction, 0 = don't G. 1 = drink wine, 0 = drink vodka

Some possible interrelations: Fishing responsibly earns you more when others do the same. Defection pays less when there are some sanctioners in the group, and at least some monitors. Not sanctioning is more costly when there are those willing to punish non-sanctioners, and someone is monitoring. The effect of monitoring is greater when monitors wear glasses. Drinking vodka makes you real happy, but decreases the effect of your monitoring... I think this kind of stuff could all get represented in a couple matrices?

Anyway, stuff we can play with, see all tomorrow.

Liz: Throwing in my two cents . . . The basic question I am interested in is seeing how norms/institutions develop in certain populations under varying conditions. There would be a null set win which agents interact in randomly selected groups of four. The players compete in a game and the winning player gets a payoff. After the first game is played, the agents may adopt one strategy from the matrix of the other agents. I am particuarly interested in manipulating thse baseline conditions to get at the environment that Andy mentioned above and what Paul said in our last meeting. I would suggest manipulating communication as a proxy for face-to-face interaction, resource availability and possibly memory to determine how these varying environments may change the pattern of norm development.

Paul: It seems like we are moving toward a fairly novel and very general way to represent social games. This is cool in itself. One more specific goal of the model may be to flesh out some general principles of behavioral evolution, defining very general classes of traits (e.g. those traits that are open to exploitation at low frequency, those traits that are strongly dependent on the pre-existence of other traits, etc.) and making some statements about the patterns of evolution of those traits on different interaction networks, given different memory constraints, group sizes, and other features of the model we can play with. These 'patterns of evolution' might include statements about the likelihood of emergence, stability, path-dependence, sensitivity to initial conditions, sensitivity to error or misperception, or dependency on network structure of a particular class of traits.

Next Meeting

Tuesday June 19 at 6:00 before th Pizza Party in the Coffee Shop?

treppenlift firmen in Germany Treppenlift