|
Activity Number:
|
354
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #304955 |
|
Title:
|
A Steplength Algorithm for Fitting ERGMs
|
|
Author(s):
|
Ruth M. Hummel*+ and Mark S. Handcock and David R. Hunter
|
|
Companies:
|
Penn State University and University of Washington and Penn State University
|
|
Address:
|
, University Park, PA, 16802,
|
|
Keywords:
|
Erdos-Renyi ; exponential random graph model ; Markov chain Monte Carlo ; maximum likelihood estimation ; maximum pseudolikelihood estimation ; partial stepping
|
|
Abstract:
|
Because of the intractability of directly calculating the maximum likelihood estimate in many exponential family random graph models, a Markov chain Monte Carlo method can be used for obtaining approximate maximum likelihood estimates. This approximation to the likelihood ratio relies heavily on the starting value being close to the MLE, while providing no intuition into how we might choose this starting value. In this paper we describe the difficulties surrounding maximum likelihood estimation in ERG Models, and we introduce as a promising improvement in MCMC ML estimation an iterative method of moving toward the MLE by jumping between the original parameter space and a reparameterized space in which we know the value of the MLE. We illustrate this method with an simple example using an Erdos-Renyi model.
|
