Degree distributions in sexual networks: a framework for evaluating evidence

Area of Science:

• Epidemiology
• Network Science
• Statistical Modeling

Background:

• Understanding the structure of sexual networks is crucial for predicting sexually transmitted infection (STI) transmission.
• Power-law models are frequently used to describe the degree distribution of social networks, but their empirical fit and predictive accuracy for epidemics remain debated.

Purpose of the Study:

• To develop and apply a statistical framework for evaluating the fit of power-law and alternative social process models to sexual network data.
• To assess the accuracy of STI epidemic predictions derived from these different network models.

Main Methods:

• A likelihood-based statistical framework was employed to compare model fit.
• Five US-based surveys of sexual networks were analyzed.
• Model performance was assessed using Akaike Information Criterion, Bayesian Information Criterion, and prediction of generalized STI epidemics.

Main Results:

• Formal goodness-of-fit tests did not consistently favor any single model across all datasets.
• Power-law models predicted a generalized STI epidemic in the US.
• Most alternative models failed to predict a generalized epidemic.

Conclusions:

• Power-law models do not offer a superior fit to US sexual network data compared to alternative models.
• The epidemic predictions derived from power-law models were found to be inaccurate.
• Development of more sophisticated models is necessary to accurately represent the behavioral underpinnings of sexual networks for effective disease transmission modeling.