Evaluation via simulation of statistical corrections for network nonindependence

Area of Science:

• Social epidemiology
• Health services research
• Network analysis

Background:

• Social processes and context significantly influence health behaviors.
• Social influence within networks, such as among healthcare providers, can shape health outcomes.
• Statistical methods often assume data independence, which is violated in social networks.

Purpose of the Study:

• To systematically compare commonly used statistical methods for addressing network non-independence due to social influence.
• To evaluate the performance of these methods regarding false positive rates, coefficient bias, and statistical power.

Main Methods:

• Simulated network data were generated to mimic social influence.
• Eight statistical methods for accounting for network non-independence were compared.
• Performance metrics included false positive rates, unbiased coefficient estimates, and statistical power.

Main Results:

• None of the tested methods achieved the nominal 0.05 false positive rate.
• Random sampling of network nodes demonstrated the lowest false positive rates.
• The random sampling method resulted in reduced statistical power.

Conclusions:

• Current methods for addressing network non-independence in health services research have limitations.
• Further methodological development is necessary to balance accuracy and power in analyzing social influence.
• Accurate analysis of health-related social influence requires robust statistical approaches.