Effect of incident subgraph sampling on separability of heavy-tailed alternatives from power laws

Determine how incident subgraph (edge-based) sampling influences the statistical separability of non-power-law heavy-tailed distributions—specifically lognormal and stretched exponential distributions—from power-law distributions when analyzing subsampled frequency/degree data.

Background

The paper studies how subsampling, particularly incident subgraph sampling where each edge is included with probability π along with its endpoints, affects the identification of power-law distributions versus other heavy-tailed distributions. Prior work has highlighted difficulties in distinguishing power laws from alternatives such as lognormal and stretched exponential, and subsampling can distort observed distributions, complicating inference.

The authors frame a key uncertainty about whether and how this subsampling scheme changes the ability to separate heavy-tailed alternatives from true power laws—a question central to interpreting empirical data drawn from partial observations of large systems.