|
Activity Number:
|
390
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #310143 |
|
Title:
|
Extensions to Latent Cluster Models for Social Networks
|
|
Author(s):
|
Pavel Krivitsky*+ and Mark S. Handcock and Adrian E. Raftery and Peter D. Hoff
|
|
Companies:
|
University of Washington and University of Washington and University of Washington and University of Washington
|
|
Address:
|
CSSS, Seattle, WA, 98195-4320,
|
|
Keywords:
|
social networks ; latent variable models ; model-based clustering ; Bayesian inference
|
|
Abstract:
|
Latent space models for social networks postulate the existence of a latent ``social space', where the probability of a relation between entities depends on their relative positions within this space. The latent cluster model for social networks models groups of entities as model-based clusters on the latent space, and Bayesian fitting of this model via MCMC allows the latent space position estimation to borrow strength from the cluster process. We refine and extend this model in a number of ways, including reparametrizing to improve interpretability and reduce computational cost, modeling inhomogeneity through actor-level random effects, and generalizing the model to non-binary data. We demonstrate applications of this family of models to several datasets.
|
