Power-Law Distributions in Empirical Data
By Aaron Clauset, Cosma Rohilla Shalizi, Mark Newman. 2007
9,837 citations recorded by the source corpus.
Power-law distributions occur in many situations of scientific interest and have significant consequences for our understanding of natural and man-made phenomena. Unfortunately, the detection and characterization of power laws is complicated by the large fluctuations that occur in the tail of the distribution—the part of the distribution representing large but rare events—and by the difficulty of identifying the range over which power-law behavior holds. Commonly used methods for analyzing power-law data, such as least-squares fitting, can produce substantially inaccurate estimates of parameters for power-law distributions, and even in cases where such methods return accurate answers they are still unsatisfactory because they give no indication of whether the data obey a power law at all. Here we present a principled statistical framework for discerning and quantifying power-law behavior in empirical data. Our approach combines maximum-likelihood fitting methods with goodness-of-fit tests based on the Kolmogorov-Smirnov (KS) statistic and likelihood ratios. We evaluate the effectiveness of the approach with tests on synthetic data and give critical comparisons to previous approaches. We also apply the proposed methods to twenty-four real-world data sets from a range of different disciplines, each of which has been conjectured to follow a power-law distribution. In some cases we find these conjectures to be consistent with the data, while in others the power law is ruled out.
Subfields: Economics and Econometrics, Statistical and Nonlinear Physics, Scaling & Learning Theory
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Economics and Econometrics
- Michael W. Mahoney
- Norbert Marwan
- Charles H. Martin
- Gangjin Wang
- Chi Xie
- Jürgen Kurths
- Yong Zou
- Reik V. Donner
In Statistical and Nonlinear Physics
- Jure Leskovec
- Mark Newman
- Santo Fortunato
- Kristina Lerman
- Olaf Sporns
- M. A. Porter
- Huan Liu
- Jon M. Kleinberg
In Scaling & Learning Theory
- Bin Jiang
- Joseph Bonneau
- Michael W. Mahoney
- Remco van der Hofstad
- Neil Zhenqiang Gong
- Jure Leskovec
- Kristina Lerman
- Matjaz Perc
Citations in the corpus
- Gradient-Based Neural DAG Learning (2019, background): “Barab´asi (2009) have stated that these types of graphs have similar properties to real-world networks which can be found in many different fields, although these claims remain controversial (Clauset et al., 2009).”
- Protection Against Reconstruction and Its Applications in Private Federated Learning (2018, methodology): “Using Zipfian models of text and discretely sampled data [21], a reasonable a priori model of the sequence X, when we assume that a user must draw at least m elements, is that independently”
- Scalable training of artificial neural networks with adaptive sparse connectivity inspired by network science (2017, neutral): “To see if the null hypothesis between the degree distribution of the hidden neurons and a power-law distribution can be rejected, we have computed the p -value 35,36 between them using a one-tailed test.”
- Adversarial Attacks on Neural Networks for Graph Data (2018, background): “While there is no exact and closed-form solution to estimate α in the case of discrete data, [10] derived an approximate expression, which for our purpose of a graph G translates to”
- What Stops Social Epidemics? (2011, background): “This distribution has a scale-free shape that is common to degree distributions of social and other complex networks (Amaral et al. 2000; Clauset, Shalizi, and New- 1The data set is available at http://www.isi.edu/˜ lerman/downloads/digg2009.html man 2009) and is well described by a power law of…”
- Griffiths phases and the stretching of criticality in brain networks (2013, methodology): “Special attention has been devoted to the choice of the lower bound Smin, as advised in Ref [59].”
- How Fat is the Top Tail of the Wealth Distribution? (2018, neutral): “Absolute number Pct of sample Sample size > 2 million > 1 million > 500TH > 2 million > 1 million > 500TH (1) (2) (3) (4) (5) (6) (7) USA 6482 965 1259 1692 0.”
- powerlaw: A Python Package for Analysis of Heavy-Tailed Distributions (2013, methodology): “A fundamental assumption of the maximum likelihood method used for fitting, as well as the loglikelihood ratio test for comparing the goodness of fit of different distributions, is that individual data points are independent [5].”
- Real-world visual statistics and infants' first-learned object names (2017, background): “Like the distribution of elements in many naturally occurring phenomena [35-37], the frequency distribution of visual objects in these corpora of”
- A framework for quantitative analysis of cascades on networks (2010, methodology): “We do not use the commonly used meth-ods such as least square minimization, because the data that spans many orders of magnitude and least square minimization can produce substantially inaccurate estimation of parameters of heavy-tailed distributions like the power-law [4].”
- Selection Problems in the Presence of Implicit Bias (2018, background): “More generally, given the ubiquity of power law distributions [6], we find it surprising how little is known about how their order statistics behave qualitatively.”
- The multi-layer network nature of systemic risk and its implications for the costs of financial crises (2015, background): “[53] Aaron Clauset, Cosma Rohilla Shalizi, and Mark EJ Newman.”
- Self-organized criticality as a fundamental property of neural systems (2014, background): “The identification of power-laws is thus delicate and demands for advanced fitting procedures because power-laws are difficult to differentiate from other heavy-tail distributions (Clauset et al., 2009; Klaus et al., 2011; Markovi´c and Gros, 2014).”
- Image sequence geolocation with human travel priors (2009, methodology): “Black curves are maximum likelihood fits of truncated Lévy flight models [5, 11].”