Diffusion/Contagion Processes on Social Networks

Abstract

This study models how new ideas, practices, or diseases spread within and between communities, the diffusion of innovations or contagion. Several factors affect diffusion such as the characteristics of the initial adopters, the seeds; the structure of the network over which diffusion occurs; and the shape of the threshold distribution, which is the proportion of prior adopting peers needed for the focal individual to adopt. In this study, seven seeding conditions are modeled: (1) three opinion leadership indicators, (2) two bridging measures, (3) marginally positioned seeds, and (4) randomly selected seeds for comparison. Three network structures are modeled: (1) random, (2) small-world, and (3) scale-free. Four threshold distributions are modeled: (1) normal; (2) uniform; (3) beta 7,14; and (4) beta 1,2; all of which have a mean threshold of 33%, with different variances. The results show that seeding with nodes high on in-degree centrality and/or inverse constraint has faster and more widespread diffusion. Random networks had faster and higher prevalence of diffusion than scale-free ones, but not different from small-world ones. Compared with the normal threshold distribution, the uniform one had faster diffusion and the beta 7,14 distribution had slower diffusion. Most significantly, the threshold distribution standard deviation was associated with rate and prevalence such that higher threshold standard deviations accelerated diffusion and increased prevalence. These results underscore factors that health educators and public health advocates should consider when developing interventions or trying to understand the potential for behavior change.

Keywords: agent-based models; contagion; diffusion of innovations; social networks.