Network effects error components models

Abstract

This paper develops a random effects error components structure for network data regression models. In particular, it allows for edge and triangle specific components, which serve as a basal model for modeling network effects. It then evaluates the potential effects of ignoring network effects in the estimation of the variance-covariance matrix. Network effects will typically imply heteroskedasticity, and as with the Moulton factor, the key role is given by the joint consideration of the intra-network correlation of the error term(s) and the covariates. Then it proposes consistent estimator of the variance components and Lagrange Multiplier tests for evaluating the appropriate model of random components in networks. Monte Carlo simulations show the tests have very good performance in finite samples.