Title: Latent Space Approaches to Network Analysis
Peter Hoff
Assistant Professor
Department of Statistics

Abstract:
Network data describe the presence or absence of social,
physical, and other relations between experimental units.
Analysis of such data is complicated by the
possibility of non-independence of these relations.
We take a latent variable approach to analyzing network
data, in which the relations between two units
are independent of relations with other individuals,
conditional on some unobserved  set of latent positions
in ``social space.'' This method of analysis is made
feasible using a diffuse prior for latent positions and
MCMC. We use this method to analyze datasets on
Monks, a fourth-grade classroom, and Florentine families.
In each dataset, the latent variable approach outperforms
existing methods of analysis.