Distinguishing influence-based contagion from homophily-driven diffusion in dynamic networks

Abstract

Node characteristics and behaviors are often correlated with the structure of social networks over time. While evidence of this type of assortative mixing and temporal clustering of behaviors among linked nodes is used to support claims of peer influence and social contagion in networks, homophily may also explain such evidence. Here we develop a dynamic matched sample estimation framework to distinguish influence and homophily effects in dynamic networks, and we apply this framework to a global instant messaging network of 27.4 million users, using data on the day-by-day adoption of a mobile service application and users' longitudinal behavioral, demographic, and geographic data. We find that previous methods overestimate peer influence in product adoption decisions in this network by 300 -700%, and that homophily explains >50% of the perceived behavioral contagion. These findings and methods are essential to both our understanding of the mechanisms that drive contagions in networks and our knowledge of how to propagate or combat them in domains as diverse as epidemiology, marketing, development economics, and public health.dynamic matching estimation ͉ peer influence ͉ social networks ͉ identification T he recent availability of massive networked data sets has enabled studies of population-level human interaction at unprecedented scale (1-3). Such studies document the persistent structural properties of networks (4), how they form, evolve, and dissolve (5), and how their structure is correlated with social interaction (1, 6, 7), individual and collaborative team performance (8-11), health outcomes (12-14), and global product demand patterns (15). Networks of interactions among individuals also provide the primary pathways along which viral contagions spread in social, biological, technological, and economic systems (16-18), which may explain why network structure is correlated with such a variety of outcomes. Yet although many studies model the dynamics of viral spreading by using assumptions about susceptibility rates, transition probabilities, and their relationships to network structure, few large-scale empirical observations of networked contagions exist to validate these assumptions (16)(17)(18).We analyze a new, large scale dataset which comprehensively captures the diffusion of a mobile service product over a social network for 5 months after its launch date. A key challenge in identifying true contagions in such data is to distinguish peer-topeer influence, in which a node influences or causes outcomes in its neighbors, from homophily, in which dyadic similarities between nodes create correlated outcome patterns among neighbors that merely mimic viral contagions without direct causal influence (19). Although the diffusion patterns created by peer influence-driven contagions and homophilous diffusion are similar, they are likely to result in significantly different dynamics. Influence-driven contagions are self-reinforcing and display rapid, exponential, and less predictable diffusion as they evolve (18...

Social networks and environmental outcomes

“…It has been observed for various types of social relations, including friendships, marriage, and information sharing (12). Existing research on homophily shows that it can heavily influence the structure of social networks and their effects on people's lives (15)(16)(17)(18)(19)(20)(21)(22)(23). One of the most pervasive effects of homophily is that it can cause social networks to become highly clustered (13,21,24).…”

Evaluating reproductive decisions as discrete choices under social influence

“…All too often, however, observational data alone fail to distinguish not only between social learning and individual learning, but also between social learning and homophily-the tendency for people with similar traits to co-associate [19,20]. Sorting this out may require explicit temporal ordering of socially influenced events in observational data [21,22].…”

Integrating social networks and human social motives to achieve social influence at scale

“…Aral and coworkers offer a critical review of some of the novel approaches being attempted, including attempts to leverage big data to conduct "randomized trials" to isolate the causal claims of influence (39) and dynamic matched sample estimation techniques (40). A social network approach highlights the structured nature of influence.…”