This paper considers a global game with ambiguity-averse players, where the variance of noise terms in private signals may be unknown, and it presents a laboratory experiment to test comparative statics results with respect to information quality. Suppose that one of the actions is a safe action yielding a constant payoff and it is a risk dominant action. Then, low quality of information makes less players choose the safe action, whereas ambiguous quality of information makes more players choose the safe action. The experimental results show that subjects' behavior is consistent with this comparative statics results.
Ai Takeuch, Yukihiko Funaki, Mamoru Kaneko, and J. Jude Kline
日付:
Sat, 2010-03-20
Abstract:
We conduct an experimental study on behavior and underlying cognition in prisoner’s
dilemmas with/without role-switching from the viewpoint of inductive game
theory. Subjects start with no knowledge about his and the other’s payoffs, and
learn them through repeated play. In cases with no role-switching, a large proportion
of subjects adopt a dominant strategy. In the cases with role-switching, where
the subjects alternate positions (row and column players), the prediction by inductive
game theory that they will choose the pair of actions maximizing the simple
sum of payoffs is observed for various pairs. We study their behaviors and cognitions
from various different viewpoints. The study suggests a strong implication
about certain foundational assumptions of inductive game theory: In particular,
statistical hypothesis tests are not only used to analyze the experimental data but
also play more positive roles for inductive game theory.
Humans think, communicate, and behave to adapt to a particular social ecology, and by doing so they collectively create, maintain, and change the very ecology (i.e., social niche) they adapt to. The niche construction approach to culture analyzes how people induce each other to think and behave in particular ways by behaving in particular ways themselves. The best support to this approach is found when cultural differences in cognition and behavior that are regularly observed in everyday life disappears in a social vacuum – in a situation in which people are rid of any social concern about implications of their behavior and can express their thoughts and behave free of any social concern. Examples of such experimental manipulations are presented and discussed.
The centipede game is one of the most celebrated examples of the paradox of backward induction. Experiments of the centipede game have been conducted in various settings: two-person games with linearly increasing payoffs (McKelvey and Palfrey, 1992), two-person games with constant-sum payoffs (Fey, McKelvey and Palfrey, 1996) and three-person games (Rapoport et al. 2003). The deviations from the subgame-perfect equilibrium prediction observed in laboratories have so far been attributed to some kind of fairness concern or altruism of the subjects. This paper attempts to offer another explanation for the observed deviations by using level-k analysis, a non-equilibrium model of strategic thinking. We show that level-k analysis gives consistently good predictions for the results of experimental centipede games. The results suggest that experimental results of centipede games be explained without invoking fairness or altruism.
We conduct experiments of a cheap-talk game with incomplete information in which one sender type has an incentive to misrepresent her type. Although that Sender type mostly lies in the experiments, the Receiver tends to believe the Sender's messages. This confirms "truth bias" reported in communication theory in a one-shot, anonymous environment without nonverbal cues. These results cannot be explained by existing refinement theories, while a bounded rationality model explains them under certain conditions. We claim that the theory for the evolution of language should address why truthful communication survives in the environment in which lying succeeds.
We conducted a field experiment on the Internet and investigated the participants' belief updating in an individual learning environment where they observe a sequence of private signals and in a social learning environment where they observe a sequence of other people's actions. We observed that participants do not update their posterior beliefs as efficiently as Bayesian, and that participants rely more on private signals than on other people's actions even when the informativeness of both is identical. Furthermore, we confirmed that participant's trust in other people's actions and their conformity to other people's actions are affected by their demographic characteristics.
This paper investigates face-to-face lying and beliefs associated with it. In experiments in Sweden and Japan, subjects answer questions about personal characteristics, play a face-to-face sender-receiver game and participate in an elicitation of lie-detection beliefs. The previous finding of too much truth-telling (compared to the equilibrium prediction) also holds in the face-to-face setting. A new result is that although many people claim that they are good at lie-detection, few reveal belief in this ability when money is at stake. Correlations between the subjects' characteristics and their behavior and performances in the game are also explored.
In this paper, a variant of Kiyotaki and Wright model of emergence of money is investigated. In the model, each good has different durability rather than storage cost assumed in Kiyotaki and Wright model. Two goods are infinitely durable but one is not durable. With certain condition, non durable good can be money as a medium of exchange. But equilibrium condition may be sensitive to the time evolution of the distribution of goods that each agent holds in its inventory. We test, with several learning models using different level of information, whether or not the steady state in this economy can be attainable if the distribution of goods is far from the steady state distribution. Belief learning with full information outperforms other models. The steady state never be attained by belief learning with partial information. Agents learn not to use non durable good as money by reinforcement learning which does use no information of the distribution of goods. It is surprising that providing partial information of the distribution of goods is rather detrimental for attaining emergence of a non-durable good money.
We present the experimental results of cheap-talk games with private information. We systematically compare various equilibrium refinement theories and bounded rationality models such as level-k analysis in explaining our experimental data. As in the previous literature, we find that when interests between sender and receiver are aligned, informative communication arises frequently. While babbling equilibrium play is observed more frequently in conflictive interest cases, a substantial number of players show a tendency to choose truth-telling and credulous play. We also find that level-k analysis outperforms equilibrium refinement theories in explaining this phenomenon. Our results also confirm the existence of the “truth bias” and “truth-detection bias” reported in communication theory.