Item Response Models for Rating Relational Data.

Area of Science:

• Social network analysis
• Psychometrics
• Statistical modeling

Background:

• Relational data, often from sender-receiver ratings in directed networks, presents unique analytical challenges.
• Existing models may not fully capture the nuances of reciprocity and homophily in these networks.
• Understanding network structures and individual attributes is crucial in various social science domains.

Purpose of the Study:

• To introduce novel item response models tailored for rating relational data.
• To enable one-dimensional latent scale comparisons of senders and receivers.
• To account for unobserved homophilic relationships within the network.

Main Methods:

• Development of item response models for directed networks.
• Bayesian parameter estimation.
• Markov Chain Monte Carlo (MCMC) methods for posterior distribution approximation.

Main Results:

• The proposed models effectively capture reciprocity and clustering in relational data.
• Simulation studies confirm satisfactory parameter recovery, even with small network dimensions.
• The models demonstrate utility for both complete and incomplete network data.

Conclusions:

• The introduced item response models provide a robust framework for analyzing relational network data.
• The approach facilitates nuanced comparisons of network actors and reveals underlying relationship dynamics.
• The models offer valuable insights for empirical applications in social network analysis.