Identification and estimation of linear social interaction models

Abstract

This paper has two parts. The first part derives the identification conditions for higher-order social interaction models. In the case where social effects depend on the distance between individuals, the upper bounds on the network diameters for non-identified models are derived. Many network properties of non-identified models in the literature can be derived from these upper bounds. This part analyzes which fixed effect elimination methods require less restrictive identification conditions. The second part considers estimation with panel data. This part develops an estimator which is computationally simple and asymptotically as efficient as the maximum likelihood estimator under normality.