Taming a menagerie of heavy tails with skew path analysis

The discovery of stable, heavy-tailed distributions of activity on the web has inspired many researchers to search for simple mechanisms that can cut through the complexity of countless social interactions to yield powerful new theories about human behavior. A dominant mode of investigation involves fitting a mathematical model to an observed distribution, and then inferring the behaviors that generate the modeled distribution. Yet, distributions of activity are not always stable, and the process of fitting a mathematical model to empirical distributions can be highly uncertain, especially for smaller and highly variable datasets. In this paper, we introduce an approach called skew-path analysis, which measures how concentrated information production is along different dimensions in community-generated data. The approach scales from small to large datasets, and is suitable for investigating the dynamics of online behavior. We offer a preliminary demonstration of the approach by using it to analyze six years of data from an online health community, and show that the technique offers interesting insights into the dynamics of information production. In particular, we find evidence for two distinct point attractors within a subset of the forums analyzed, demonstrating the utility of the approach.