Ising Attitude Network Model

Description

A model and analysis that treats an attitude as a network of binary feelings, beliefs and behaviours whose alignment depends on attention to the topic, fitted to survey data to reveal attitude structure and estimate its consistency.

In this model each node is an evaluative reaction toward an object (a belief such as 'is honest', a feeling such as 'angry', a behaviour such as 'eats meat'). Nodes influence each other through weighted links and have their own thresholds, and an external field such as a campaign can push all nodes. Attention or involvement works like inverse temperature: when people attend to the object, the network 'freezes' into a consistent positive or negative configuration and becomes hard to move (hysteresis); when attention is low, reactions are loosely connected and responsive to outside pressure but do not stick. Empirically, binary or dichotomised items are fitted with IsingFit, communities of items are detected (for example warmth and competence beliefs), and Gibbs entropy can quantify how consistent an attitude network is. A learning variant shows how repeated high attention strengthens links so that later consistency needs less attention. For UX and product research, the model gives a structured way to analyse attitudes to brands, features or technologies, to judge how settled they are, and to reason about why persuasion fails with highly involved users while low-involvement users are easily swayed but quickly revert.

Attributes

Loss type
Latent
Effort
Weeks

Steps

  1. Write items for distinct beliefs, feelings and behaviours toward one attitude object, scored as binary or dichotomised.
  2. Collect a large sample, ideally with a measure of involvement or attention to the topic.
  3. Fit an Ising network with IsingFit and plot it, adding community detection to group related reactions.
  4. Compare networks or connectivity between high- and low-involvement groups.
  5. Estimate attitude consistency, for example with Gibbs entropy on small networks.
  6. Use simulation of the model to reason about how involvement and messaging would change the attitude, and test those predictions.

Sources

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

Also known as

  • Causal attitude network model
  • CAN model
  • Ising attitude model
  • Attitude network entropy

Categories

  • Analytics & Behaviour
  • Behavioral Science & Economics