[Paper Review] A Model of Competing Narratives

Kfir Eliaz, Ran Spiegler|arXiv (Cornell University)|Nov 10, 2018

Electoral Systems and Political Participation8 references4 citations

TL;DR

This paper formalizes political narratives as causal models using Bayesian Networks (DAGs), showing that competing narratives—especially 'opportunity narratives'—can sustain extreme policies by distorting beliefs about policy consequences. In equilibrium, narratives that maximize anticipatory utility favor hopeful, biased beliefs, leading to polarization even when narratives use the same variables but different causal structures.

ABSTRACT

We formalize the argument that political disagreements can be traced to a "clash of narratives". Drawing on the "Bayesian Networks" literature, we model a narrative as a causal model that maps actions into consequences, weaving a selection of other random variables into the story. An equilibrium is defined as a probability distribution over narrative-policy pairs that maximizes a representative agent's anticipatory utility, capturing the idea that public opinion favors hopeful narratives. Our equilibrium analysis sheds light on the structure of prevailing narratives, the variables they involve, the policies they sustain and their contribution to political polarization.

Motivation & Objective

• To formalize the role of competing narratives in shaping public opinion and political polarization.
• To model how narratives function as causal models that map policy actions to perceived consequences.
• To analyze how public opinion selects between narratives based on anticipatory utility, favoring 'hopeful' stories.
• To investigate the structural conditions under which narratives sustain extreme policies despite shared variables.
• To explore the equilibrium dynamics of narrative competition in policy debates, especially when narratives involve confounding variables.

Proposed method

• Represents narratives as directed acyclic graphs (DAGs) with actions, intermediate variables, and consequences as nodes.
• Models beliefs as conditional probabilities derived from DAGs, capturing perceived causal effects without specifying effect magnitudes.
• Defines equilibrium as a probability distribution over narrative-policy pairs that maximizes a representative agent’s anticipatory utility.
• Analyzes limiting cases (ε, δ → 0) to identify dominant narrative structures in equilibrium.
• Compares lever narratives (where a variable is endogenous and influenced by policy) with opportunity narratives (where a variable is exogenous and policy responds to it).
• Uses mathematical optimization to determine the maximum achievable belief in favorable outcomes (P(y=1|a)) under different narrative types.

Experimental results

Research questions

• RQ1 How do narratives shape public beliefs about policy outcomes in the absence of systematic information differences?
• RQ2 What structural features of causal narratives (e.g., lever vs. opportunity) lead to support for opposing policies?
• RQ3 Why do narratives that use the same variable but assign it different causal roles (endogenous vs. exogenous) lead to divergent policy preferences?
• RQ4 Under what conditions does equilibrium feature multiple dominant narrative-policy pairs, and what drives their coexistence?
• RQ5 How does the cost of narrative promotion affect the selection of narratives in equilibrium?

Key findings

• In the low-ε limit, the equilibrium narratives are opportunity narratives that maximize anticipatory utility, leading to extreme policy support.
• Opportunity narratives can achieve higher P(y=1|a=1) than lever narratives for any α ∈ (0,1), specifically 1−α(1−μ) > μ/(μ+(1−μ)α).
• Equilibrium requires α = 1/2 to balance net anticipatory utility between left and right extreme policies, ensuring narrative stability.
• The same variable can serve as a lever in one narrative (influenced by policy) and as an exogenous threat/opportunity in another, supporting opposite policies.
• Equilibrium narratives are not unique; multiple narrative-policy pairs can coexist due to diminishing returns in narrative popularity.
• When narratives are restricted to linear chains (a→x₂→x₃→y), the equilibrium structure favors longer chains, suggesting a preference for complex narratives with many variables.