Cross-Validation on Network Models

• Activity Number: 441
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Beau Dabbs*
• Companies: Carnegie Mellon University
• Keywords: Network ; Cross-validation ; Model Selection ; Stochastic Blockmodel

Abstract

We use cross-validation to perform model selection on network models. Our primary models of interest are those with conditionally independent dyads (CID Models). Of note, this class of models includes the stochastic blockmodel (SBM), mixed membership stochastic blockmodel (MMSBM), and the latent space model (LSM). We show preliminary results that suggest cross-validation should be a nearly unbiased estimator of risk for this class of models, and consider some factors contributing to the variance of this estimator. Using the Stochastic Block Model as an example, we examine the bias and variance of this estimator empirically. We then compare using 10-fold cross-validation as a model selection criterion to BIC and show that our cross-validation is able to more accurately select the true generative model.