INFERENCE FOR SOCIAL NETWORK MODELS FROM EGOCENTRICALLY SAMPLED DATA, WITH APPLICATION TO UNDERSTANDING PERSISTENT...

Area of Science:

• Social Network Analysis
• Statistical Modeling
• Epidemiology

Background:

• Egocentric network sampling is vital for studying networks where direct observation is impractical, such as sexual networks linked to HIV transmission.
• Existing methods for analyzing egocentric data lack a rigorous statistical foundation for network model estimation and inference.
• Persistent racial disparities in HIV prevalence highlight the need for advanced analytical tools to understand network influences.

Purpose of the Study:

• To develop a statistically rigorous and practical framework for estimating and inferring network models from egocentric network data.
• To apply this methodology to investigate the drivers of racial disparities in US HIV prevalence.
• To extend the framework for analyzing more complex network structures, including triadic effects.

Main Methods:

• Identification of a subclass of exponential-family random graph models (ERGMs) suitable for egocentric data.
• Application of pseudo-maximum-likelihood estimation for model parameter estimation and uncertainty quantification.
• Development of a computationally tractable approach for size-invariant ERGMs.

Main Results:

• A novel statistical methodology is established for robust network model estimation using egocentric data.
• The method provides rigorous quantification of estimation uncertainty.
• The approach is successfully applied to analyze racial disparities in HIV prevalence.

Conclusions:

• The developed framework offers a significant advancement in the statistical analysis of egocentric network data.
• This methodology can provide deeper insights into public health issues influenced by social networks.
• The framework is adaptable for analyzing more complex network structures and relationships.