Opinion Dynamics Simulation

Description

Agent-based simulations in which people with opinions interact under simple rules (copying, averaging, ignoring those too different) to explore how consensus, clusters or polarisation emerge in a community.

Opinion dynamics models share three building blocks: a network of who talks to whom, an interaction rule, and a definition of opinion as discrete or continuous. In voter models an agent copies a neighbour; in majority models it adopts the local majority. In bounded confidence models, agents only move toward others whose opinions lie within a bound: wide bounds give consensus, narrow bounds give stable clusters and polarisation, and raising the bound after polarisation may not undo it. The book's own Hierarchical Ising Opinion Model gives each agent a cusp-shaped opinion driven by information and attention, with interactions that raise attention. It reproduces a persuasion paradox: highly attentive activists entering a low-attention community can raise opponents' attention and polarise them rather than convert them. The book is frank that empirical data rarely discriminate between such models, so their main value is qualitative and robust predictions, such as that zealots increase polarisation. For UX researchers, these simulations are a thinking tool for community products, review systems, social features and advocacy campaigns, letting teams explore how design levers like exposure range or prompting engagement could shift group-level outcomes.

Attributes

Loss type
Latent
Stage
Strategy, Design
Effort
Weeks

Steps

  1. Define the research question at group level, for example whether a feed setting could reduce polarisation in a community.
  2. Choose a network topology, an opinion definition (binary, continuous, or cusp with attention) and an interaction rule matching the question.
  3. Implement the model in NetLogo or R, starting from published versions such as Deffuant, Hegselmann-Krause or HIOM.
  4. Run parameter sweeps for the design-relevant levers, such as exposure bound, share of highly engaged users or noise.
  5. Record group outcomes (consensus, number of clusters, polarisation) across many runs and check for hysteresis by reversing parameter changes.
  6. Compare the qualitative patterns with whatever field data exist and present results as scenarios, not forecasts.

Sources

  • van der Maas, H., Complex Systems Research in Psychology (2024)

Also known as

  • Bounded confidence models
  • Deffuant model
  • Hegselmann-Krause model
  • Voter model
  • Hierarchical Ising Opinion Model
  • HIOM
  • Sociophysics models

Categories

  • Strategy & Collaboration
  • Systems & Complexity