Evaluate complementary methods beyond ML and EV for classifying subsampled heavy-tailed distributions

Assess whether additional statistical approaches—including the maximum entropy test of Bee et al., Wilk’s test as used by Malevergne et al., the finite size scaling method of Serafino et al., and the methods proposed by Zhang, Corral, and Artico—can fruitfully complement the maximum likelihood framework of Clauset–Broido and the extreme value approach of Voitalov in distinguishing power-law distributions from heavy-tailed alternatives under subsampling.

Background

After evaluating the maximum likelihood (ML) and extreme value (EV) methods under incident subgraph sampling, the authors find that each has limitations, including false rejections for power laws (ML strong evidence criteria) and potential false acceptances of heavy-tailed alternatives (EV).

They therefore point to the need for investigating complementary methods from the literature—maximum entropy tests, Wilk’s test, finite-size scaling, and other approaches—to see whether they can mitigate these issues in the subsampled, heavy-tailed setting.