Schedule - Parallel Session 6 - Games and Strategic Behaviour 1WMG IDL Boardroom - 15:40 - 17:10
When it Pays to Be Nice in the Prisoner's Dilemma
Frederic Moisan, Robert Tenbrincke, Ryan Murphy, Cleotilde Gonzalez
The Prisoner’s Dilemma is a classic conundrum representing many decision problems between two players: they decide whether to cooperate with each other or to act in their own interest. Past research indicates that individual incentives influence cooperation especially in repeated interactions. However, these studies typically underplay the influence of intrinsic social preferences from each member of the pair. To what extent do genuinely nice people cooperate? To what extent do nice people get exploited? And to what extent do nice people get compensated for their behavior? In an experiment using a collection of prisoner’s dilemma games and a measure of social preferences as an individual difference, we find evidence for three distinct phenomena emerging from explicit incentive structures. We identify incentive structures in which: (1) cooperation is insensitive to social preferences; (2) where nice people can be exploited; and (3) where being nice is consistently rewarding. We discuss the theoretical and practical implications of these findings.
Isolating Distributional Social Preferences in 2X2 Games
David Butler; Pavlo Blavatskyy
We design a novel experiment to isolate the existence of distributional social preferences. Using four sets of popular 2×2 games (asymmetric matching pennies, battle of the sexes, prisoner’s dilemma and chicken) we elicit choices both when payoffs to the other party are hidden and when they are revealed. This design contrasts a player’s attitude to risk over the underlying lotteries paying out to self with preferences over the same outcomes coupled with consequences to the other. In this design the ‘recipient’ is a passive player, permitting a tightly controlled investigation of Fehr & Schmidt’s inequality- aversion parameters as only attitudes to the impact of these factors can account for any differences between treatments. No other source of intentions-based social preferences, such as reciprocity or team-reasoning, are possible in this design, removing common confounds in earlier tests of their model. Our experiment employs a Holt & Laury-type method for eliciting risk preferences over the underlying lotteries then a related ‘generalised strategy method’ to do the same when the full game is revealed to subjects. We run a simulation of a random preference formulation of the Fehr & Schmidt model with risk-aversion for each of our games to see the likely parameter magnitude and distribution that can match our experimental findings. We find evidence of pro-social concerns in the prisoner’s dilemma and stag hunt games. Importantly, we also find evidence that those showing such concerns in one game also exhibit them in others. Previous tests of Fehr-Schmidt have found some support for the model in aggregate but not at the individual level. However, we find only mild evidence of pro- or anti-social preferences in the asymmetric matching pennies and battle of the sexes’ games, suggesting distributional concerns are context dependent. At most a half of our participants showed evidence of any pro- or anti-social preferences across the set of these games. The other half of the subjects is best modelled as having inequality-aversion parameters equal to zero. While offering support to the existence of Fehr-Schmidt preferences in some games by some people, the magnitude of the effects in general is not large. This suggests distributional preferences of the kind proposed by Fehr & Schmidt may be milder and less general than previously assumed. But it also shows there is an independent effect of inequality aversion for some even after controlling for intentions-based forms of social preference.
Participants in Ultimatum Bargaining Games Have Individualistic Preferences After All
Jack Stecher; Todd Kaplan
What explains behavior in ultimatum games? Experimenter demands? Insult penalties? Fairness? We argue that, instead, ultimatum game proposer and responder behavior is well explained by (noisy) expected wealth maximization. After adjusting for random noise and learning, we find that the best predictor of behavior in ultimatum games is the subgame perfect Nash equilibrium. The key is that the type of learning involved is not learning about the game. Rather, it is learning about the mean and variance in a random utility shock. Our structural estimates show that, in our experiments, the mean of responders’ utility shock is 0. That is, responders do not impose an insult penalty on low offers or demand offers that appear fair, at least not systematically. We find that our proposers systematically underestimate the variance in responder behavior, initially believing that responders are more sensitive to proposal amounts than the data suggest. Additionally, we find that proposers initially believe the mean of the utility shock is positive, further causing them to raise their proposals. Our estimated parameters fit both our initial round data and those in the original study of Guth et al. (1982). Indeed, after adjusting for currency translations and inflation, our mean initial round proposal was within a fraction of a penny of the amount predicted by fitting our parameters from the Guth et al. data. Our design has proposers and responders play ultimatum games several times, each with perfect strangers, in order to isolate the effects of learning from feedback without introducing confounds from repeated interaction. In each round, starting with a prior fitted from the Guth et al. data, our proposers’ mean offer is 95% efficient, compared with the optimal offer given Bayesian updating based on the prior rounds. In the last round, our mean offer is within a few pennies of the wealth-maximizing amount. To control for fairness norms, we manipulate the claims of each player to the pie. In one treatment, the participants bargain over whether to allow the proposer to keep his or her show-up fee. In this treatment, the proposer has a chance to transfer some of his or her show-up fee to the responder, who then decides whether to accept the transfer and allow the proposer to keep the remainder, or who can destroy the proposer’s show-up fee. In a second treatment, the participants instead bargain over the responder’s show-up fee. In this treatment, the proposer has a chance to demand a transfer of some of the responder’s show-up fee. The responder can either agree to the transfer, keeping only the remainder, or can refuse, at the cost of surrendering the show-up fee. This is similar to the dictator games of List (2007), in which dictators can take some of a citizen’s show-up fee. We find no difference in responder behavior across treatments, and no difference in mean proposals across treatments in any given round.
Voluntary Cooperation in Local Public Goods Provision an Experimental Study
Daniela Di Cagno; Daniela Grieco
In a circular neighborhood with each member having a left and a right neighbor indi-viduals choose two contribution levels, one each for the public good shared with the left, respectively right, neighbor. This allows for general free riders, who do not contribute at all and general cooperators, who contribute to both local public goods, as well as for those who contribute in a discriminatory way. Although all local two-person games are structurally independent, we mainly aim to conﬁrm intra- as well as interpersonal spillover eﬀects. Additionally we try to answer questions like: Will one diﬀerentiate between one’s two direct neighbors when they diﬀer, e.g. in game parameters or past behavior? Will there be clusters of similarly behaving types? Will one observe initial cooperation up to end game eﬀects? We hope to infer more clearly motives for voluntary cooperation like direct and indirect reciprocity concerns via analyzing individual adaptations in playing two structurally independent games. Treatments diﬀer in cooperation incentives and structural (a)symmetry of local public goods.