We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.AP

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Applications

Title: Stable Exponential Random Graph Models with Non-parametric Components for Large Dense Networks

Abstract: Exponential Random Graph Models (ERGM) behave peculiar in large networks with thousand(s) of actors (nodes). Standard models containing two-star or triangle counts as statistics are often unstable leading to completely full or empty networks. Moreover, numerical methods break down which makes it complicated to apply ERGMs to large networks. In this paper we propose two strategies to circumvent these obstacles. First, we fit a model to a subsampled network and secondly, we show how linear statistics (like two-stars etc.) can be replaced by smooth functional components. These two steps in combination allow to fit stable models to large network data, which is illustrated by a data example including a residual analysis.
Comments: 26 pages, 10 figures, 3 tables
Subjects: Applications (stat.AP); Methodology (stat.ME)
Cite as: arXiv:1604.04732 [stat.AP]
  (or arXiv:1604.04732v1 [stat.AP] for this version)

Submission history

From: Stephanie Thiemichen [view email]
[v1] Sat, 16 Apr 2016 12:03:51 GMT (2101kb,D)

Link back to: arXiv, form interface, contact.