This paper investigates how risk influences the way people barter. We used Minecraft to create an experimental environment in which people bartered to earn a monetary bonus. Our findings reveal that subjects exhibit risk-aversion to competitive bartering environments and deliberate over their trades longer when compared to cooperative environments. These initial experiments lay groundwork for development of agents capable of strategically trading with human counterparts in diferent environments.
CEUR ceur-ws.org
so secretive behavior is operationalized as direct (vs. global) messaging. Direct messages limit
the information given to potential competitors and, from a game theory perspective, serve a
strategy for limiting risk and exposure [
We have created a Minecraft plugin called BarterPlus [
The data recorded from these experiments include a JSON file, log file and participant feedback form. Our plugin kept record of the status and events pertaining to a given game in a JSON file. This data included each participant’s username, their profession, the items they were required to obtain, their final score at the end of the game and information about each TradeRequest. TradeRequests are events that occur when a player attempted to initiate a trade with a potential trade partner. Information in a TradeRequest include the start time, end time, initiating trader (known as the requester), the recipient trader (known as the requested), the involved traders’ scores after each trade and the items each trader ofered in an exchange. TradeRequests can have one of three diferent status: accepted; cancelled; or, denied. An accepted request is one wherein both players agreed to exchange items denoted in the trade placed in the trade menu. A cancelled trade is one wherein the requester withdrew their request. A denied trade is one wherein the requested rejected the request. The log file includes the raw output from the server console which includes timestamped logs of all global chat communications, all direct messages, and the details of each trade as they occurred.
A total of 109 subjects participated in our IRB-approved study. We conducted twenty-two runs, each involving 5-7 players. Data from 2 runs was discarded because of too few participants and a technical error. Research participants were recruited using flyers posted on the university campus and posts on university specific subreddits and Discord servers. The eligibility criteria to participate included being at least 18 years old, having Minecraft Java edition installed on a computer to enable remote connection to our server, and having prior experience playing Minecraft. Participants were paid $10 for the hour-long study and could earn an additional $5 bonus if the value of the items they obtained exceeded a prescribed threshold, disclosed to them at the start of the experiment. Each experiment was one of three types: competitive; cooperative; or, baseline. In the competitive condition, only one player, i.e., the player with the most points, earned the bonus. In the cooperative condition, all players obtained the bonus if every player collected a suficient number of the items to meet the prescribed threshold. In the baseline condition, every player obtaining number of points at or above the threshold was able to earn the bonus.
Players joined the study using a privately shared Zoom link. This ensured clear communication of the instructions and opportunity for clarifications, if necessary. At any point in the session, players could check their score. At the end of the 30 minutes, the server closed forcing everyone out of the study. Participants would then be prompted to provide feedback about the study and any strategies they used during the game.
A total of 109 players ( = 109 ) participated in the experiment. Table 1 presents statistics from the study. With respect to the hypotheses, we note that subjects used global messaging significantly more ( = 0.02 ) in the cooperative condition than in the competitive or baseline, as expected.
Likewise, results show significantly more direct messaging in the competitive condition ( = 23.52% , = 31.61% ) where the risks of not earning the bonus was higher than in the cooperative condition ( = 10.43% , = 13.11% ). This finding is supported by a two-tailed t-test ( = 0.02 ). Fig. 2 presents a box plot of all players’ percentage use of direct messages for each condition. Some of the traders from the three scenarios do not use direct messages at all. The maxima of the competitive game indicate a minority of players opting for global message.
Our results also suggest that traders experience greater workload in the competitive situation. In support of our hypotheses, we find that the trade acceptance time is longer in the competitive condition ( = 27.28 , = 68.09 ) than in the cooperative condition ( = 18.54 , = 42 ). This is supported by a two-tailed t-test ( = 0.02 ). Fig. 3 presents time to accept trades in the three game types. Small diferences in lower values indicate traders take similar minimum time to accept a trade. The upper value of acceptance time is approximately same for baseline and cooperative scenarios but traders in competitive scenario take significantly longer to accept a trade. Table 1 shows that similar behavior holds for the 25ℎ and 75ℎ percentiles, suggesting that competitive behavior is visibly diferent from other game types in the upper quartiles.
Our results show that traders’ communication strategies are significantly influenced by the condition. In the competitive environment, traders use significantly more direct communications. Presumably they recognize that global messaging in this environment could provide valuable information to competitors and reduce their chance of obtaining the bonus. In contrast, in the cooperative condition, traders broadly share the content of their inventory and the list of items they need. This behavior reflects the fact that risks associated with sharing information are low in this condition.
Our data also indicates that traders spend significantly greater time thinking about trades before accepting them in the competitive condition than in the cooperative condition. Here again, risk likely influences traders to carefully consider the consequences of accepting a deal and whether better opportunities might be available in the future. In the cooperative condition, the risk associated with a bad trade is minimal. Hence, little thought is needed.
The data we are collecting is rich. Hence this work represent only a surface level preliminary analysis how risk and trust efect trading behavior in barter situations. Subsequent experiments will examine how trust evolves and attempt to create agents capable of strategically trading in diferent environments.
Portions of this material is based on work supported by the Air Force Ofice of Scientific Research