The Dynamics of Web Traffic

This paper examines the distribution of Web audiences over time using a rich new data set. We show that traffic churn is highly structured. While large sites have stable audiences, variance in traffic increases exponentially as sites get smaller. Web traffic concentration overall is remarkably stable. We propose and test a simple model that reproduces the most important features of the observed data. Under a wide variety of initial conditions - including when sites start with equal traffic - the model converges to an approximate power law traffic distribution. We argue that these patterns are the result of preferential stability and discuss its implications for communication and the online public sphere.