Discerning media bias within a network of political allies: an analytic condition for disruption by partisans

TL;DR

This paper develops an analytic condition to identify when social networks with partisans fail to learn true media bias, revealing how network properties influence opinion turbulence and false belief formation.

Contribution

It introduces an analytic instability criterion for media bias learning in social networks, extending previous models by incorporating probabilistic opinions and network effects.

Findings

Derived an instability condition for turbulent nonconvergence.

Verified the condition through Monte Carlo simulations.

Linked results to social science theories of structural balance.

Abstract

An individual's opinion concerning political bias in the media is shaped by exogenous factors (independent analysis of media outputs) and endogenous factors (social activity, e.g. peer pressure by political allies and opponents in a network). Previous numerical studies show, that persuadable agents in allies-only networks are disrupted from asymptotically learning the intrinsic bias of a media organization, when the network is populated by one or more obdurate agents (partisans), who are not persuadable themselves but exert peer pressure on other agents. Some persuadable agents asymptotically learn a false bias, while others vacillate indefinitely between a false bias and the true bias, a phenomenon called turbulent nonconvergence which also emerges in opponents-only and mixed networks without partisans. Here we derive an analytic instability condition, which demarcates turbulent nonconvergence from asymptotic learning as a function of key network properties, for an idealized model of media bias featuring a biased coin. The condition is verified with Monte Carlo simulations as a function of network size, sparsity, and partisan fraction. It is derived in a probabilistic framework, where an agent's opinion is uncertain and is described by a probability density function, which is multimodal in general, generalizing previous studies which assume that an agent's opinion is certain (i.e. described by one number). The results and their social implications are interpreted briefly in terms of the social science theory of structural balance.