Modeling Heterogeneous Peer Assortment Effects Using Finite Mixture Exponential Random Graph Models.

Area of Science:

• Social Network Analysis
• Statistical Modeling
• Psychological Science

Background:

• Exponential random graph models (ERGMs) are widely used for network analysis.
• Existing ERGM frameworks may struggle with unobserved heterogeneity in nodal covariate effects.
• This can lead to model misfit and inaccurate conclusions in social network research.

Purpose of the Study:

• To develop a novel class of models, sender/receiver finite mixture exponential random graph models (SRFM-ERGMs).
• To extend the ERGM framework to explicitly model unobserved heterogeneity without assuming a block structure.
• To evaluate the performance and utility of SRFM-ERGMs in analyzing social networks.

Main Methods:

• Development of the SRFM-ERGM framework.
• Application of the models to an empirical dataset on adolescent substance use.
• Extensive simulations under various conditions to assess model performance.

Main Results:

• Unobserved heterogeneity in nodal covariate effects is a significant source of misfit in standard network models.
• The proposed SRFM-ERGM approach effectively alleviates model misfit caused by unobserved heterogeneity.
• The models demonstrate good performance across diverse simulated network conditions.

Conclusions:

• SRFM-ERGMs provide a valuable tool for analyzing social networks with unobserved heterogeneity.
• Accounting for unobserved heterogeneity is crucial for accurate network modeling in psychological science.
• The findings have important implications for researchers studying social interactions and behavior.