Convolution of individual and group identity: self-reliance increases polarisation in basic opinion model

The study addresses how individual decisions, shaped by social interactions, can produce large-scale societal outcomes such as electoral choices and market trends. Motivated by observed increases in political polarisation globally and during crises, the authors investigate whether two opposing human desires—belonging to a group and pursuing individual distinctiveness—can drive opinion polarisation. Grounded in optimal distinctiveness theory, they develop an agent-based model (ABM) in which agents balance social conformity and personal conviction without invoking homophily or bounded confidence. The model also incorporates a final binary decision to mimic real-world choices and introduces an opinion spread metric as a proxy for social cohesion. The core research questions are whether self-reliance (individualism) drives polarisation, how it shapes convergence to opinion clusters, and how it affects the trade-off between self-fulfilment (alignment with intrinsic conviction) and societal cohesion.

The paper situates its contribution within ABM research on opinion dynamics and polarisation. Prior work has emphasized homophily—tendency to interact with similar others—which explains group formation and various social structures. Extensions include biased assimilation, where inconclusive evidence reinforces prior beliefs, leading to extremity. Other strands incorporate repulsion (distancing from dissimilar opinions) and attraction-repulsion mechanisms to study tolerance, hysteresis, and factional dynamics. Further studies examine multilayer networks (echo chambers) and coevolving networks. Empirical findings on social media’s role in polarisation are mixed across platforms. The presented model differs by (a) not relying on homophily or explicit repulsion, (b) using a constant, intrinsic self-reliance parameter to modulate social influence versus adherence to initial conviction, and (c) producing emergent polarisation and stable clusters from random initial conditions. It also connects to individualisation mechanisms studied previously but removes adaptivity and homophily, focusing instead on a timeless trait: self-reliance.

Agent-based model with agents on a periodic 2D grid (default 100×100, G≈10,000). Each agent i has two time-invariant attributes: initial conviction A_i ∈ [0,1] and self-reliance γ_i ∈ [0,1] (0 = fully socially dependent; 1 = fully self-reliant). Initial attitude A_i(0) is set to A_i unless otherwise specified. Agents occupy cells on a torus and interact with a Moore neighbourhood of radius 1 (Chebyshev distance ≤1), though robustness checks consider radius 2 and partially populated grids. Dynamics: At discrete time steps, agents update attitudes synchronously based on neighbourhood influence and pull toward their own initial conviction: ΔA_i/Δt = r[(1 − γ_i) N_i(t) + γ_i (A_i − A_i(t−1))], where N_i(t) = Σ_{j∈M_i} w_{ij}(A_j(t−1) − A_i(t−1)), with constant weights w_{ij}. The neighbourhood term captures average local disagreement; the self-reliance term pulls toward initial conviction, with stronger effect when local disagreement is larger. The model runs until convergence to a steady state. After convergence, agents take a binary decision d_i = 1 if A_i(final) > 0.5, else 0. Equilibria: Two qualitative equilibria are derived (in supplementary material), including N_i = 0 and a condition linking A_i − A_i(final) to 1/γ_i times the sign of N_i. Measures: Self-fulfilment is quantified via decision alignment δ_i = 1{A_i(final)>0.5} − 1{A_i>0.5}. Societal self-fulfilment is Δ = (1/G) Σ δ_i. A mean-field approximation yields an analytical expectation for average decision alignment as a function of γ_i and P(N_i≠0), assuming independence approximations; only sufficiently self-reliant agents contribute to alignment gains. Social cohesion is proxied by opinion spread w = Pct90(A_final) − Pct10(A_final). Larger w indicates lower cohesion. Parameter distributions explored include uniform and normal (means 0.5–0.9, σ = 0.1; σ varied for robustness). Robustness checks include larger neighbourhood radius, partially populated grids, and decoupling initial conviction from initial attitude.

• Emergence of clusters: From random initial conditions with uniform A and γ, the system evolves to stable, opposing opinion clusters. Final attitudes form spatial clusters with extreme cores and moderate peripheries; final binary decisions reveal multiple distinct clusters.
• Distributional shift: Final attitudes across society become approximately normal with long tails; medians of groups (split by initial conviction above/below 0.5) move closer to 0.5, but tails persist with strongly held views.
• Self-reliance drives polarisation: Increasing the mean of γ (normal distributions with σ=0.1, means 0.5, 0.7, 0.9) increases the variance of final attitudes, separation between group means, and overall opinion spread, indicating stronger polarisation.
• Tail composition: Highly self-reliant agents (larger γ) disproportionately populate the extreme tails of the final attitude distribution and thus drive polarisation.
• Trade-off: Societies with higher average γ show higher average decision alignment (greater self-fulfilment; agents’ final decisions more often match initial convictions) but also larger opinion spread w (lower social cohesion). This trade-off is visualized as a monotonic relationship between alignment and spread.
• Analytical consistency: A mean-field approximation of decision alignment as a function of γ (assuming a fraction of agents at N_i=0; e.g., P(N_i=0)=0.25) matches the simulation trend qualitatively.
• Robustness: Findings persist under varied γ standard deviations, partially populated grids, larger neighbourhood radius (with slightly more central distributions and delayed but steeper spread increases), and when initial attitude is decoupled from initial conviction.
• Reference point: Due to uniform initial A, the initial opinion spread is about 0.8; it increases with mean γ in the simulations.

The model demonstrates that a minimal mechanism—balancing social influence with pull toward intrinsic conviction via a self-reliance parameter—can generate polarisation from unstructured starting conditions, without invoking homophily or explicit repulsion. As the share of self-reliant agents rises, society exhibits opinion drift and increased polarisation, mirroring empirical trends such as recent US political polarisation where distributions of views drift apart. The dynamics align with psychological theories like optimal distinctiveness and self-determination, and relate to empirical notions such as acrophily. Highly self-reliant individuals act as drivers of polarisation by anchoring extremes that influence local clusters. The observed trade-off highlights a societal-level tension: more self-reliance increases individual decision alignment but reduces cohesion. The simplicity of the model helps isolate this mechanism and suggests it could complement other known drivers (homophily, biased assimilation, repulsion, echo chambers) in explaining real-world polarisation.

The paper introduces a reduced-form ABM in which self-reliance mediates the balance between social conformity and adherence to intrinsic conviction. From minimal assumptions, the model produces stable opinion clusters, increased polarisation with rising self-reliance, and a clear trade-off between individual self-fulfilment (decision alignment) and social cohesion (opinion spread). Analytical approximations corroborate simulation trends. As a conceptual building block, this mechanism can help explain observed polarisation dynamics. Future research could extend the framework by exploring richer network topologies, dynamic or context-dependent self-reliance, interactions with homophily or repulsion mechanisms, alternative decision thresholds, and empirical calibration/validation against longitudinal opinion data.

• Minimal mechanism: The model omits homophily, repulsion, media effects, and other behavioural biases; polarisation emerges solely from self-reliance versus social averaging.
• Network simplification: Primary analyses use a regular grid with Moore neighbourhoods and uniform interaction weights; although robustness checks exist, real social networks are heterogeneous and dynamic.
• Static traits: Self-reliance is constant over time; in reality, individual needs for distinctiveness may adapt to context.
• Analytical approximation: Mean-field derivations assume independence between intrinsic attitudes and neighbourhood influence and rely on assumptions about the fraction of agents with N_i=0; these hold on average, not at the individual level.
• Binary decision rule: Final decisions depend on a fixed threshold (0.5), which may oversimplify real decision processes and institutions.
• Parameterization and scaling: Results depend on choices like time scale r and distributional assumptions for A and γ; quantitative conclusions may vary with different settings.