Network Effects Panel Regression: A New Method Applied to Political Economy

• In Event: Political Networks and Methodology

Abstract

Many phenomena of interest in political research exhibit network dependence, whereby units influence each other through ties of one form or another. These include, for example, public policy adoption by governments, the outbreak of civil war, individual voter turnout, and the prevalence of financial risk in countries. Conventionally, researchers account for network dependence by some combination of measuring networks of interest (e.g., spatial contiguity of countries, kinship relationships between voters) and adjusting statistical methods for the general possibility of dependence among observations (e.g., using panel corrected standard errors). Recent developments in the field of network inference offer a third possibility---directly inferring and modeling the effects of unobserved networks through which units depend on each other. Unfortunately, most existing network inference algorithms would require researchers to first infer the network, and then incorporate the network into downstream analyses such as regression---analyzing the same data twice. We introduce a novel method, which we term network effects panel regression (NEPR), in which the effects of covariates, and the edges through which units depend on each other, are simultaneously inferred. We apply NEPR to several datasets from recently-published applications in the area of political economy. In nearly all applications NEPR infers latent edges that capture previously unmodeled dependence between units. In most applications, NEPR outperforms the published models in forecasting experiments. Across the applications, a number of the results change in substantively meaningful ways from incorporating network dependence. We conclude that NEPR represents a valuable, and complimentary, third option for both exploring and accounting for network dependence in panel data applications.

Authors

• Lingyu Jack Fuca, The Pennsylvania State University

• Bruce Desmarais, Pennsylvania State University

• Rebekah Fyfe, Pennsylvania State University