Decomposing coordination failure in stag hunt games

There are many ways to define a 2x2 Stag Hunt game using different payoff components. This laboratory experiment studies how subjects make different choices in Stag Hunt game with different payoff components.

2018-10-25

2019-06-10

First, I aim to show a disconnection between our current models and human behaviour. Second, using these findings, a model of weighted incentives better predicts behaviour in Stag Hunt games.

The goal of the previous literature in this direction is to understand the payoff factors that influence humans converging to either equilibria. I share this goal and hope to improve upon our current understanding of this equilibrium selection criteria.

The experiment design follows the protocol associated with experimental economics. More particularly, my experiment design follows as closely as possible the design of previous papers investigating this phenomena. The novelty of my approach is entirely in the payoff-construction of the games.

Previous work has explored the payoff characteristics of Stag Hunt games that drive human behavior to either the payoff dominant or risk dominant equilibrium. These experiments use a between-subjects design where cohorts of eight subjects play one Stag Hunt game for seventy-five rounds. In each round, subjects were randomly re-matched and were aware of the outcome of their choices in previous rounds. Subjects accumulated points over the rounds which were converted to dollars at the end of the experiment. Subjects were paid for all seventy-five rounds. I plan to follow this design with 4 different experimental games.

My experiment design proceeds with two phases. First, using the paper's theoretical results, two games are created which are equivalent when analyzed by a class of models (Game 1 and 2). Therefore, these concepts have the same prediction about human behavior in Games 1 and 2. The first set of hypotheses test whether this equivalence accurately describes human choices. Are human choices constant across Games 1 and 2?

Second, the previous models claim empirical relationships between the direction of their concepts and human behavior. That is, the payoff-dominant choice is more likely to emerge in games with smaller or larger parameters. Games 3 and 4 are created so that all of these models predict the same behavioral change across games. Our second set of hypotheses test whether these relationships accurately describe human choices. Are human choices in agreement with these models in Games 3 and 4? I anticipate that human choices will follow the opposite prediction of these previous models.

Randomization is carried out by the computer.

Individual

No

Following from the previous literature, I plan to have 8 sessions each with 8 subjects for each of my 4 treatments. This leads to a total of 256 subjects.

Each of the 256 subjects will make 75 choices. So there will be a total number of 19,200 choices observed.

There will be 8 Game 1 sessions, 8 Game 2 sessions, 8 Game 3 sessions, and 8 Game 4 sessions.