Testing the hypothesis of preferential attachment in social network formation

Area of Science:

• Epidemiology
• Network Science
• Statistical Modeling

Background:

• The hypothesis of preferential attachment (PA) is debated in epidemiological networks.
• Simple PA models suggest diseases cannot be eradicated via population-level controls like random vaccination.
• Evidence for PA is typically sought through asymptotic power-law behavior.

Purpose of the Study:

• To develop a general statistical method for testing preferential attachment (PA) in count data.
• To apply this method to contact data relevant to respiratory disease transmission.
• To investigate if PA, despite its implications, allows for disease control.

Main Methods:

• Developed a flexible, numerically efficient, likelihood-based statistical model.
• Applied the model to integer count data representing social contacts.
• Utilized model selection techniques to evaluate PA evidence.

Main Results:

• Identified evidence supporting a form of preferential attachment (PA).
• Found that the best-fitting PA models indicate significant contact heterogeneity.
• Demonstrated that this heterogeneity permits the control of respiratory diseases.

Conclusions:

• Standard model selection may favor PA, but careful analysis reveals nuances.
• Contact heterogeneity within PA frameworks can enable disease control strategies.
• The developed statistical approach is applicable to other integer datasets for PA hypothesis testing.