(Re-) Inventing the Traffic Light: Designing Recommendation Devices for Play of Strategic Games
How to cite
Duffy, John; Anufriev, Mikhail; Panchenko, Valentyn; Young, Benjamin. 2022. "(Re-) Inventing the Traffic Light: Designing Recommendation Devices for Play of Strategic Games." AEA RCT Registry. AEARCTR-0009706. https://doi.org/10.1257/rct.9706-1.1
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
We explore whether and how individuals can design recommendations for the play of a variety of different two action, two player games. The recommendations are elicited in the form of probability distributions for each player over the possible action profiles of each game. The games are then played by programmed robot players who consider the recommendations made and decide, on a basis of best-response analysis, whether or not to follow the recommendations. The individuals providing the recommendations (the subjects) earn a baseline payment per game and they receive further rewards if their recommendations are followed by the robot players. This experimental design enables insights into how well the designers (subjects) understand the strategic incentives at play in various two-player, two action games including Prisoner's Dilemma, Matching Pennies, Battle of the Sexes and Chicken. In some of these games, there are multiple recommendation systems that the robot players will agree to follow, that is there, are multiple correlated equilibria of these games that can be selected by the designer. One aim of this research is to determine whether subjects can learn to implement the most efficient of these correlated equilibria in their choice of recommendation systems.
This experimental design differs from the traditional method of collecting data on the play of two player, two action games in that we use only a single player who designs a recommendation system for the play of the game by robot players. Nevertheless, we view this design as complementary to evidence from the traditional method and may yield evidence of greater strategic sophistication by our single, recommendation designing subjects since our experimental design reduces strategic uncertainty.
Intervention
Subjects will be confronted with various two player, two action games that differ in their payoffs. In a sequence of rounds, they will first be shown the payoff matrix for each new game. Then, they will be asked to design recommendation devices for that game. A recommendation device involves the specification of probabilities with which the two players 1, and 2 will play either of their two available actions, labeled Red or Blue. Probabilities are elicited via the specification of the colors of balls in a container of 24 balls. Each ball is split in two halves, with one half labeled 1 and the other half labeled 2. Players designate the color of each half of all balls in the container. This decision comprises the recommendation device. Robot players are then presented with this recommendation device. The robot players then decide whether or not to follow the recommendations, based on a best response analysis from a simulation of playing the games a large number of periods. If one or more of the 2 robot players does not follow the recommendation device, the subject gets a baseline payoff. They also get some feedback as to why recommendations were not followed. If the recommendations are followed in all periods, then the subject earns, in addition to the baseline payoff, the minimum average payoff earned by either player 1 or 2, which ever is lower. Each game is played for several rounds and at the start of each round following the first, subjects get feedback on whether their recommendations were followed and have the opportunity to re-design the recommendation device or keep it unchanged. After all games are played, one game is chosen randomly for payoff purposes.
Experimental design
Subjects face a variety of different 2 player, 2 action games represented by payoff matrices. For each game they are asked to design a recommendation device specifying for each game which action each player 1 or 2 should play in each round of the 2x2 game. This recommendation device involves the specification of probabilities with which each player will be recommended to choose either of their two actions. Robot players are then presented with this recommendation device. The robot players then decide whether or not to follow the recommendations, based on a best response analysis from a simulation of playing each game a large number of periods under the given recommendation device. If one or more of the 2 robot players chooses not follow the recommendation device, the subject gets a baseline payoff. They also get some feedback as to why recommendations were not followed. If the recommendations are followed in all periods, then the subject earns, in addition to the baseline payoff, the minimum average payoff earned by either player 1 or 2, which ever is lower. Each game is played for several rounds and at the start of each round following the first, subjects get feedback on whether their recommendations were followed and have the opportunity to re-design the recommendation device or keep it unchanged. After all games are played, one game is chosen randomly for payoff purposes.
Design details
Cited by
No citation data available for this registration yet.