Impact of Survey Design on Estimation of Exponential-Family Random Graph Models from Egocentrically-Sampled Data

Area of Science:

• Social Network Analysis
• Statistical Modeling
• Survey Methodology

Background:

• Egocentric network sampling is a popular, cost-effective method for collecting network data.
• It involves observing a sample of nodes (egos) and their immediate connections (alters).
• Traditional analysis methods have limitations with this data structure.

Purpose of the Study:

• To investigate how design choices in egocentric network studies affect statistical estimation and inference.
• To explore the impact of measurement and sampling strategies on Exponential-family Random Graph Models (ERGMs).
• To provide guidance for optimizing egocentric network study designs.

Main Methods:

• Utilized Exponential-family Random Graph Models (ERGMs) for statistical analysis of egocentric network data.
• Conducted simulation studies to assess the impact of different sampling designs.
• Examined measurement strategies for ego/alter attributes and ties, and sampling strategies for egos/alters.

Main Results:

• Design choices significantly impact the statistical estimation and inference of ERGMs.
• Harmonizing measurement specifications across egos and alters is important for accurate modeling.
• Sampling strategies, such as stratified sampling and degree censoring, demonstrably affect statistical inference.

Conclusions:

• Principled statistical inference is achievable with egocentric network data using ERGMs.
• Careful consideration of measurement and sampling designs is essential for valid conclusions.
• Fitted ERGMs can simulate complete networks, enabling inference on whole-network properties.