Identification of Social Influence in Bipartite Behavior Cascades

• In Event: Advances in Probabilistic Network Data Analysis

Abstract

How can social influence be distinguished statistically from the effect of prior similarity of units in event-based cascades of behavior?
Examples in political science include diffusion versus independent adoption of policies across subnational states, the ratification of treaties by states in the international system, and publicly visible voting (by judges, legislators etc.) as a function of prior preferences vis-a-vis social influence. While relational event models for bipartite graphs offer statistical inference for time-stamped sequences of ties between units and their behavior, they cannot reliably differentiate
between the two causal mechanisms. This contribution pinpoints and rectifies this identification problem. We evaluate the effectiveness and scope conditions of a shuffle test for distinguishing contagion or
diffusion on the one hand from the effect of prior similarities among units on the other hand. The bipartite relational event model coupled with temporal randomization is effective when four criteria are met:
(i) availability of fine-grained temporal data about agents (Mode I) without extensive simultaneity in the observations;
(ii) one or more kinds of binary behavior the agents can adopt once or repeatedly (Mode II);
(iii) at least partial observability of behavior by agents;
(iv) observability of one or more time-invariant attributes or network relations that structure the behavior in a relational way ("channels" or "sources of influence").
Using Monte Carlo simulations, we examine the efficacy of the permutation test under mixtures of similarity and influence regimes, with a varying time horizon, different numbers of observations, and
variations in the number of senders (agents) and receivers (behavior types) and conclude that the technique is applicable in a wide variety
of empirical research contexts. We demonstrate its applicability using an example of the ratification of global environmental treaties as a function of social influence through shared democratic values, geographical neighborhood, and shared climatic
vulnerability.

Authors

• Sebastián Martínez Bustos

• Laurence Brandenberger

• Philip Leifeld, University of Essex