Model selection for network data based on spectral information

TL;DR

This paper presents a new non-parametric spectral method for selecting the best network data model among various classes, based on the graph Laplacian spectrum, with demonstrated effectiveness through simulations and real applications.

Contribution

It introduces a novel spectral-based, non-parametric approach for model selection in network data, addressing challenges of comparing diverse model classes.

Findings

Effective in distinguishing between different network models

Performs well in simulation studies

Useful in real-world network data applications

Abstract

We introduce a new methodology for model selection in the context of modeling network data. The statistical network analysis literature has developed many different classes of network data models, with notable model classes including stochastic block models, latent position models, and exponential families of random graph models. A persistent question in the statistical network analysis literature lies in understanding how to compare different models for the purpose of model selection and evaluating goodness-of-fit, especially when models have different mathematical foundations. In this work, we develop a novel non-parametric method for model selection in network data settings which exploits the information contained in the spectrum of the graph Laplacian in order to obtain a measure of goodness-of-fit for a defined set of network data models. We explore the performance of our proposed methodology to popular classes of network data models through numerous simulation studies, demonstrating the practical utility of our method through two applications.