Regression Analysis of Asymmetric Pairs in Large-Scale Network Data

Abstract

In social network analysis, many estimation methods have been developed over the past three decades. Due to the computational complexity for analyzing large-scale social network data, however, those methods cannot be applied effectively. On the other hand, the structure of large-scale network data is often sparse so that the information loss by ignoring symmetric pairs is rather limited. Hence, we propose an asymmetric pairs regression (APR) approach to study the social network relationship. Accordingly, the computation of parameter estimations is simple and the theoretical properties can be obtained via the established logistic regression model. Simulation studies and an empirical example are presented to illustrate the usefulness of APR.

Keywords: Asymmetric Pair, Dyad Independence Model, Latent Space Model, Social Network, Stochastic Block Model

JEL Classification: C10, C20