Quantifying diffusion in social networks: a Bayesian approach

The term ‘diffusion’ refers to the spread of a character, such as a novel behaviour, through a population. While such diffusions often result from social learning, there are other types of social influence, as well as non-social processes, which can account for this spread. Network-based diffusion analysis (NBDA) infers, and quantifies, the strength of social influence in a set of diffusion data by assessing the extent to which the pattern of spread follows a social network. Here the chapter illustrates the application of NBDA in a Bayesian context with the use of a simulated dataset. The chapter extends current NBDA models to incorporate random effects and facilitate model discrimination. The chapter employs a Reversible jump Markov Chain Monte Carlo algorithm to discriminate between models and determine which model provides the best fit to the data. This novel methodology is particularly useful to analyse datasets that include many covariates and thus can be fitted with a correspondingly large number of competing models.