Fitting Position Latent Cluster Models for Social Networks with latentnet

Area of Science:

• Network analysis
• Statistical modeling
• Computational statistics

Background:

• Social network analysis often assumes relationships depend on unobserved actor characteristics.
• Latent position models represent these characteristics in a Euclidean space, with relationships as a function of actor distances.
• Extensions allow for clustering of actor positions, revealing group structures within networks.

Purpose of the Study:

• To introduce latentnet, a software package for fitting and evaluating latent position and cluster models for networks.
• To implement Bayesian inference via Markov chain Monte Carlo (MCMC) and maximum likelihood estimation methods.
• To provide tools for assessing model fit, determining the number of clusters, and estimating probabilistic cluster memberships.

Main Methods:

• Utilizes latent space models where actor relationships are functions of distances in a Euclidean space and covariates.
• Implements Bayesian inference using MCMC algorithms for parameter estimation and model evaluation.
• Offers maximum likelihood (ML) and two-stage ML methods for latent position and cluster models, respectively.

Main Results:

• The package facilitates the estimation of actor positions in a latent space and their probabilistic cluster assignments.
• It provides Bayesian methods to assess the evidence for clustering by estimating the number of groups.
• Goodness-of-fit can be evaluated using posterior predictive checks, and networks can be simulated from fitted models.

Conclusions:

• latentnet provides a comprehensive framework for fitting and evaluating advanced network models.
• The package supports both Bayesian and maximum likelihood approaches, offering flexibility in analysis.
• It aids in uncovering hidden structures and group memberships within social networks through latent position and cluster modeling.