Behavioral Experiments on Competitive Contagion

We describe behavioral experiments based on the model of networked competitive contagion developed by Goyal and Kearns in [2012], which built upon the work of [Bharathi et al. 2007; Borodin et al. 2010; Chasparis and Shamma 2010]. In this model, there are two distinct “infections” competing for maximum spread through a social network. For instance, these infections might represent the products of two rival firms, or the ideas of two opposing political parties. Each individual in the network must choose only one of the two options, and this decision is stochastically influenced by the choices of his neighbors. We imagine that each competing firm can “seed” the initial adoption of its product by selecting certain individuals to receive promotions or give-aways. The goal of each firm is to choose these seeds to maximize eventual adoption throughout the network at the expense of its competitor. In our experiments, a player takes the perspective of a firm, and chooses seed vertices for a series of networks. We evaluate a player on the performance of his seed choice against the distribution over the seed choices of all other players. Our goal is to study the collective performance of the population as a whole, which is measured by the population’s distance to equilibrium on each network. Our use of behavioral experiments to study strategic interaction in networks is inspired by the long line of work summarized in [Kearns 2012]. In these previous experiments, each participant was given control of a single vertex in a network, and provided with a local view of his immediate neighbors only. The experiment was performed simultaneously, with each player present in the same room. Our setting is notably different. We allow each subject a global view of the network when running simulations and making seed choices. Additionally, participation is asynchronous, with each participant playing on a web-based simulator at his convenience. The new setup allows for greater flexibility in experiment design, and makes participation easier for the subjects. Our motivation for shifting from local to global information stems from the increased awareness of networks in everyday life; individuals no longer simply act within networks, but have begun to reason about them as well. Thus, there is great value in discovering how people understand these mathematical models.