The third day of the EHBEA conference in Amsterdam brought several highly interesting talk on the evolution of cooperation, altruism, and punishment, my main areas of interest. Unfortunately, I am a bit strapped of time, so I will only summarise the talks most important to me.
Altruistic Punishment in Public Goods Games
The day commenced with a keynote by Simon Gächter of the University of Nottingham. Gächter, together with Ernst Fehr, pioneered the use of punishment in Public Goods Games1 and has been a major proponent of altruistic punishment as a solution to free-rider problems.
Gächter recapped his research program of the last ten years, showing that altruistic punishment induces cooperation, occurs widely,2 and increases total pay-offs from cooperation in the long term.3 In particular, he argued that, though abstract, Public Goods Games are psychologically rich, indicated by the anger trigger by free-riding.
A major challenge to the ecological validity of Gächter’s experiments comes from the fact that they are conducted under conditions of full anonymity, which some have argued opens the door to unrealistically harsh punishment. New data from ongoing experiments which he presented counter this criticism. When participants meet each other before playing the games, even for a short moment, this increases cooperation even in the absence of punishment (but does not stop its gradual decline).
When identifiability and punishment are combined, cooperation increases to near-perfection. At the same time, punishment becomes less frequent than expected; nevertheless it successfully sustains cooperation that would otherwise decline (or so does the threat; some groups never actually punish). This is quite fascinating, because it indicates that in prehistoric small-scale societies, in which members of small communities frequently interacted, punishment would have been a highly cost-effective way of enforcing cooperation.
Social and Individual Information; Prospect Theory
Ulf Tölch of the Humboldt University of Berlin presented findings from experiments on the integration of social and individual information. In a two-phase experiment, individuals first learned about their own accuracy in indicating a target location on a circle. Subsequently, they were presented with a combination of their own guess and either more or less accurate social information (or a combination of the latter two).
Tölch found that when integrating two bits of social information, players made bayes-optimal decisions, i.e. weighted the integration of information for source reliability. When integrating their own information with other sources, however, failed to do this. In particular, it appeared that people who were very accurate themselves overestimated their own accuracy. Using fMRI scans, the researchers found evidence that some people – who acted Bayes-optimal – were able to overwrite individual information.
Dave Mallpress presented a model for the evolution of the fourfold pattern of risk preference described by prospect theory4 In a variable, but autocorrelated environment, agents dependent on energy levels were offered the choice between a (more or less risky) gamble and a safe option. Whether agents in the model chose to gamble depended on the state of the environment in a fashion similar to the fourfold pattern of prospect theory: gamble in extremely bad environments, but play it safe in extremely good ones; mostly gamble in quite good environments, but mostly play it safe in quite bad ones.
Antonio Silva of the University College London presented two experiments out of a larger research program on parochial altruism and inter-group conflict in Northern Ireland. This research program is particularly interesting because it aims to maximise the ecological validity of experiments. The methods Silva described look very promising to me.
Parochial altruism is the idea that inter-group conflict gives rise to increased in-group altruism and decreased out-group altruism. Northern Ireland with its long-lasting conflict between Catholics and Protestants naturally lends itself to studying this phenomenon. Silva and his colleagues used several methods, including donations (of endowments to neutral, Catholic, or Protestant charities) and a lost-letter paradigm, in which letters were addressed to either Catholic or Protestant neighborhoods and ‘lost’.
The lost-letter paradigm found evidence for reduced out-group altruism – Catholic letters ‘lost’ in Protestant neighbourhoods were returned less often than Protestant and neutral letters (and vice-versa). Donations to in-group, out-group, and neutral charities were predicted mostly by socio-economic variables; a sectarian threat variable was only negatively correlated with out-group donations. Hence both measures found evidence for reduced out-group altruism, but not increased in-group altruism, and thus not for parochial altruism.
From Small-Scale to Large-Scale Societies
Simon Powers of the University of Lausanne presented a model for the evolution of punishment institutions. He argued that while most theories (such as Gächter’s) assume that social interactions are uncoordinated, “in real groups [they] tend to be regulated by institutions.” Powers explicitly bases his models for the bottom-up creation of institutions on the work of Elinor Ostrom, who also pioneered research into altruistic punishment.5
Power’s model is based on a modified Public Goods Game. Instead of making punishment decisions individually, agents first decide on the share of the public good they would like to see used for punishment (vs. investment), and then play the PGG. Institutional rules are formed by taking the mean preference of cooperators and defectors for sanctioning. The dynamics of the model are thus governed by individual preference for punishment and propensity to cooperate, defect, or not participate in the PGG.
When most of the public good is used for investment, cooperators can invade, but when investment gets too high, asocials take over, thus leading to cycling dynamics. When spatial structure is introduced in the model, however, where migration is dependent on the level of cooperation within a group, cooperators can take over a group. The group size then expands and cooperation as well as institutional sanctions stabilise at high levels.
Power’s model is interesting for multiple reasons. I was particularly intrigued that it considers migration rate as a variable dependent on cooperation levels (rather than as a constant, which I’ve seen in many group-level selection models). I’d also be curious to see how such institutional sanctions would fare in behavioural experiments (while being aware that as an evolutionary model, this does not make predictions about contemporary behaviour).
- Fehr, E. & Gächter, S. (2000). Cooperation and Punishment in Public Goods Experiments. The American Economic Review, 90(4), 980-94. [↩]
- Herrmann, B., Gächter, S., & Thöni, C. (2008). Antisocial punishment across societies. Science, 319(5868), 1362-7. DOI: 10.1126/science.1153808. [↩]
- Gächter, S., Renner, E., & Sefton, M. (2008). The long-run benefits of punishment. Science, 322(5907), 1510. DOI: 10.1126/science.1153808 [↩]
- Kahneman, D. & Tversky, A. (1979). Prospect Theory: An Analysis of Decision Under Risk. Econometrica, 47, 263-91. [↩]
- Ostrom, E., Walker, J., & Gardner, R. (1992). Covenants With and Without a Sword: Self-Governance is Possible. The American Political Science Review, 86(2), 404-17. [↩]