Optimal evidence accumulation on social networks
A fundamental question in biology is how organisms integrate sensory and social evidence to make decisions. However, few models describe how both these streams of information can be combined to optimize choices. Here we develop a normative model for collective decision making in a network of agents performing a two-alternative forced choice task. We assume that rational (Bayesian) agents in this network make private measurements, and observe the decisions of their neighbors until they accumulate sufficient evidence to make an irreversible choice. As each agent communicates its decision to those observing it, the flow of social information is described by a directed graph. The decision-making process in this setting is intuitive, but can be complex. We describe when and how the absence of a decision of a neighboring agent communicates social information, and how an agent must marginalize over all unobserved decisions. We also show how decision thresholds and network connectivity affect group evidence accumulation, and describe the dynamics of decision making in social cliques. Our model provides a bridge between the abstractions used in the economics literature and the evidence accumulator models used widely in neuroscience and psychology.