Algebraic Statistics for Network Models

Abstract

This project focused on the family of exponential random graph models (ERGMs) for networks, characterized by global network summary statistics. These models are especially attractive because they are built precisely around the kinds of network characteristics that analysts are concerned with in most practical applications. The team has proposed a systematic program of mathematical research into the algebraic geometric structure of parameter estimation and assessing model fit for these and related statistical models. The team has reached all three of the proposed Phase I measurable milestones and made significant progress toward future proposed work, reaching an additional milestone originally proposed for later phases. In particular, the team has: (1) created new tools for assessing the goodness of fit of models and comparison of models within the ERGM class; and (2) characterized the statistical properties of ERGMs using geometric tools and, specifically, identified when ERGMs are "nice," i.e., not exhibiting near-degeneracies of the sort described in the statistical literature.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Feb 19, 2014
Accession Number
ADA601381

Entities

People

  • Alessandro Rinaldo
  • Sonja Petrović
  • Stephen E. Fienberg

Organizations

  • Pennsylvania State University

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Air Force Research Laboratories
  • Algorithms
  • Applied Mathematics
  • Computational Science
  • Computations
  • Computer Programs
  • Data Science
  • Data Sets
  • Families (Human)
  • Machine Learning
  • Maximum Likelihood Estimation
  • Social Networks
  • Social Sciences
  • Statistics
  • Symposia
  • Universities
  • Workshops

Readers

  • Computer Networking
  • Regression Analysis.
  • Theoretical Analysis.