Large-scale study of social network structure and team performance in a multiplayer online game

A question of interest in both theory and practice is if and how familiarity between members of a team, expressed in terms of social network structure, relates to the success of the team in a given task. In this paper we revisit this important question in a novel manner by employing game outcome statistics from Dota 2, a popular team-based multiplayer online game, combined with network data from Steam Community, a social networking service for gamers. We conduct a large-scale analysis of 4168 teams to study how network density, and the minimum and maximum degree of the within-team social network are associated with team performance, and determine how this association is moderated by team skill. We observe that minimum degree is strongly associated with good performance, especially in teams with lower skill. Together with previous results on network density that we corroborate in this paper, our findings suggest that a successful team is not only moderately connected overall, but its members should also individually have not too few nor too many within team connections.