Scale-free networks are rare.

Area of Science:

• Network science
• Complex systems analysis
• Statistical physics

Background:

• Scale-free networks, characterized by power-law degree distributions (k^-α), are widely assumed to model real-world systems.
• This assumption has significant implications for understanding network structure and dynamics.
• However, the empirical prevalence and universality of scale-free networks remain debated.

Purpose of the Study:

• To rigorously test the empirical prevalence of scale-free network structures across diverse real-world networks.
• To compare the fit of power-law distributions against alternatives like log-normal distributions.
• To identify which types of networks, if any, exhibit strongly scale-free properties.

Main Methods:

• Organized various definitions of scale-free networks.
• Applied advanced statistical methods to a large dataset of nearly 1000 networks.
• Included social, biological, technological, transportation, and information networks.

Main Results:

• Found robust evidence that strongly scale-free structures are empirically rare across most networks studied.
• Log-normal distributions provided an equally good or better fit to the data compared to power laws for the majority of networks.
• Social networks showed at best weak scale-free properties, while some technological and biological networks exhibited strong scale-free characteristics.

Conclusions:

• The universality of scale-free networks is not supported by empirical data from a broad range of real-world networks.
• Log-normal distributions are a more frequent and often better fit for network degree distributions.
• The structural diversity of real-world networks necessitates new theoretical frameworks beyond the standard scale-free model.