A research paper by Professor Pang Xuan from our school, serving as the corresponding author and co-first author, titled A Bayesian Multifactor Spatio-Temporal Model for Estimating Time-Varying Network Interdependence, was recently published online as a FirstView article in the journal Political Science Research and Methods (PSRM). This study is part of the phased achievements of the National Social Science Foundation’s major project “Theoretical and Methodological Research on Geopolitical Risk Prediction,” for which she serves as the lead expert.
The global diffusion of behavioral patterns, policy decisions, value norms, and crisis risks is a critical issue in international relations research in the era of globalization. The research challenges primarily stem from two aspects. First, in the temporally and spatially reconfigured context of globalization, indirect (higher-order) relationships and feedback loops create multiplier effects, significantly increasing the complexity of diffusion mechanisms. Second, actors are not passively influenced by relational networks but actively construct social relationships to aid decision-making and action. The issue of endogeneity is a well-known challenge in the study of social network effects (or “peer effects”).
The paper proposes a novel solution to these challenges. First, the authors extend the traditional setup of spatio-temporal econometric models using multilevel modeling. This Bayesian model more reasonably treats peer effects as the result of both micro-level relational connections and macro-level structural evolution, effectively addressing the first challenge. Second, the authors combine multifactor analysis with machine learning as a bias-correction strategy for identification, allowing for a stricter distinction between the “birds of a feather flock together” peer effect and the “like attracts like” self-selection effect. They designed a Markov Chain Monte Carlo algorithm for parameter estimation and factor selection, and conducted three sets of computational simulation experiments to test the reliability and contribution of the new method.
Design of Computational Simulation Experiments
The paper applies the new method to specific studies in international relations. The first study seeks to answer whether international migration networks contribute to the diffusion of terrorism risk. Reliable evidence obtained through the new method shows that international migration flows did increase the global spread of terrorism risk before the 1990s, but their effect afterward lacks statistical reliability—meaning migration networks are no longer a significant channel for terrorism diffusion. This shift may be related to advancements in communication technology and the rise of social networks, which have greatly reduced the necessity and importance of physical contact for the spread of terrorism risk. In the second study, the authors examine the global diffusion of trade liberalization policies over recent decades. Findings from the new method suggest that “socialization” within the World Trade Organization framework did not promote policy convergence toward trade liberalization among nations. Instead, waves of domestic political transformation and the preferences of hegemonic states emerged as more reliable factors explaining global trade liberalization.
In the conclusion, the paper systematically compares the new method with existing major approaches and reports the simulation and empirical evidence underlying these comparisons in the online appendix. Additionally, the authors provide guidelines for applying the new method, including its limitations, and offer an open-source software package, bpNet.
The co-first author, Liu Licheng, is a Ph.D. candidate in the Department of Political Science at the Massachusetts Institute of Technology, with research interests in international political economy and quantitative methods in social sciences.
Political Science Research and Methods is the official journal of the European Political Science Association. As a peer-reviewed academic journal, PSRM publishes research across major fields of political science, categorizing submissions into four types based on length and originality: research articles, research notes, replication reports, and comments. PSRM’s latest impact factor is 5.239, ranking 12th out of 187 political science journals indexed in the SSCI.