Phase space · Opinion dynamics · Trust co-evolution
The same intervention — dialogue, trust-building, structural reform — produces opposite effects depending on when it is applied. This simulation makes that visible.
X-axis: how far apart the two camps' opinions are. Y-axis: how much trust exists between them. The moving dot is your system. The coloured zones are attractors — stable states the system tends toward. The dashed curve is the tipping point boundary. Above it: dialogue is possible. Below it: the same dialogue backfires.
Dialogue — works above the tipping line. Reduces distance, builds trust. Below the line: camps are too far apart, forced contact reduces trust further. Trust building — community events, shared projects. Raises trust regardless of opinions. The gentler, slower path. Structural reform — institutional change. Short-term disruption, potentially long-term shift. High risk, high reward.
Let the system run to polarisation, then try the interventions. The trajectory tells you what's possible, not your intentions. The same resources, the same programme, the same good faith — opposite outcomes depending on where the system is when you act. Timing is not tactical. It is structural.
What this is
This simulation models a society moving through two interacting dimensions: how far apart the dominant opinion camps are, and how much trust exists between them. It is not a prediction. It is a tool for thinking — a way of making visible the dynamics that are usually only felt after the fact.
The central argument is simple: the same intervention produces opposite effects depending on where the system is when you apply it. Dialogue programmes that work in one context destroy trust in another. Structural reforms that succeed when conditions are right make things worse when they are not. The simulation makes this visible by tracking the system's trajectory through a phase space divided into distinct zones.
The underlying opinion dynamics follow the Deffuant bounded confidence model (Deffuant et al., 2000). Agents hold opinions on a continuous spectrum from −1 to +1. At each step, two agents are selected at random. If their opinions differ by less than a confidence threshold ε (openness), they move toward each other. If they differ by more, nothing happens — they are too different to influence each other.
This single rule produces three qualitatively different outcomes depending on ε: consensus (high ε), bipolarisation (medium ε — two stable camps), and fragmentation (low ε — many isolated clusters). The transition between these regimes is abrupt. A small decrease in openness can flip the entire system. The model uses D — the mean distance between the two emerging camp centroids — as its primary output.
The second variable, inter-group trust T, is added based on empirical sociology and the adaptive network literature (Holme & Newman, 2006). Trust grows when cross-camp contact happens, and decays when it stops. But the decay rate depends on D. When opinions are close, lack of contact has little effect. When opinions are far apart and contact has broken down, trust decays rapidly — and hostile contact accelerates the decay further.
This creates a bifurcation structure: above the tipping line, trust and contact reinforce each other and the system resists perturbation. Below the tipping line, distance and distrust reinforce each other and the system accelerates toward maximum polarisation regardless of what you do to opinion distance alone.
The boundary between the two basins is the curve T_critical = 0.52 − 0.35·D. At low opinion distance, quite low trust is still sufficient to remain bridgeable. At high distance, substantially higher trust is required. The curve captures the empirical finding that trust is what determines whether distance is bridgeable — not distance alone.
The system is in a stable basin. Trust gravitates toward a natural resting level (~0.55) when cross-camp contact is occurring. Distance may be high but the social infrastructure for dialogue still exists. Interventions work here.
A second basin dominates, with a strong attractor at maximum distance and zero trust. Trust decays proportional to D² — catastrophic at high distance. Cross-camp contact here is hostile contact: it accelerates trust loss. Dialogue backfires here.
| Zone | What it means | What works | What doesn't |
|---|---|---|---|
| Bridgeable | Low distance, sufficient trust. The window of opportunity is open. | Dialogue, structural reform, trust-building — all effective. Invest now. | Nothing backfires, but delay is costly. |
| Tense | High distance but trust still holds. Crossing the tipping line is the risk. | Trust-building to hold the line. Careful structural reform. | Confrontational dialogue formats risk dropping trust below the line. |
| Frozen | Low distance, low trust. Apathy or suppression — people have stopped engaging. | Low-stakes shared activity. Reduce perceived threat before raising contact. | Premature dialogue can feel threatening. Reform may be passively resisted. |
| Polarised | High distance, low trust. The strong attractor is pulling toward maximum polarisation. | Trust-building only — slowly. Do not attempt dialogue. Wait for windows. | Dialogue actively worsens the situation. Structural reform faces high resistance. |
Where on the phase space would you locate your current context? What is your evidence for the trust level you're estimating? What is driving the trust decay rate? What intervention are you currently funding, and what zone does it assume you're in?
If dialogue backfires below the tipping line, what has to happen before dialogue becomes safe? Is trust-building possible without any contact? When do external shocks — crises, shared threats — create windows by temporarily raising trust?
No elites. Elite polarisation — the strategic use of divisive rhetoric for political gain — is one of the strongest real-world drivers. It is absent here. In practice, elite behaviour shapes the trust decay rate and the effective confidence threshold of the population.
No media or algorithms. Social media amplification and algorithmic recommendation affect contact rate and opinion salience. The trust decay slider is a rough proxy, not a model of platform dynamics.
One-dimensional opinion space. Real political opinion is multidimensional. Cross-cutting cleavages on multiple dimensions can prevent polarisation that a single-axis model predicts. Treat D as a summary, not a full description.
Symmetric camps. Both camps are treated identically. Real polarisation is often asymmetric — one camp may be more internally cohesive, more extreme, or more effectively coordinated. These asymmetries matter for intervention design.
These are features, not bugs. The model works by stripping away complexity to make one dynamic legible. Use it to develop intuitions — then complicate those intuitions with the specifics of your context.