The recent availability of massive amounts of networked data generated by email, instant messaging, mobile phone communications, micro blogs, and online social networks is enabling studies of population-level human interaction on scales orders of magnitude greater than what was previ ously possible.1'2 One important goal of applying statistical inference techniques to large networked datasets is to understand how behavioral conta gions spread in human social networks. More pre cisely, understanding how people influence or are influenced by their peers can help us understand the ebb and flow of market trends, product adoption and diffusion, the spread of health behaviors such as smoking and exercise, the productivity of information workers, and whether particular indi viduals in a social network have a disproportion ate amount of influence on the system.