Identifying Social Influence in Networks Using Randomized Experiments

Identifying causal estimates of peer-to-peer influence in networks is critical to marketing strategy, public policy and beyond. Unfortunately, separating correlation from causation in networked data is complicated. We argue that randomized experimentation in networks, made possible by the digitization of human interaction at population scale, can dramatically improve our understanding of the ebb and flow of market trends, product adoption and diffusion, the spread of health behaviors, the productivity of information workers and whether or not particular individuals in a social network have a disproportionate amount of influence on the system. We also discuss some of the complications that arise when conducting randomized experiments in networks by describing an experiment designed to test how different viral product design strategies affect peer influence and social contagion in new product diffusion