Why We Experience Society Differently: Intrinsic Dispositions as Drivers of Ideological Complexity in Adaptive Social Networks

TL;DR

The paper tackles why ideological complexity emerges in adaptive social networks by focusing on intrinsic dispositions—homophily, neophily, and conformity—and quantifying their effects with normalized Lempel-Ziv complexity ($nLZ$). It uses an adaptive, directed-weight network model with continuous opinions and coevolving edge weights, coupled with discretization and three experimental setups to reveal how internal dispositions produce distinct, robust $nLZ$ patterns that persist across environments. Crucially, the authors perform explicit causal interventions that reassign a focal node's dispositions while holding the surrounding context fixed, finding that internal rules causally shift $nLZ$ trajectories in meaningful, sometimes asymmetric ways, though some transitions imply convergence to similar macroscopic outcomes. Together, these results highlight internal behavioral identities as a dynamical fingerprint shaping individual-level experiences of ideological volatility, offering a new lens on opinion dynamics and social complexity in self-organizing systems.

Abstract

Understanding the emergence of inequality in complex systems requires attention to both structural dynamics and intrinsic heterogeneity. In the context of opinion dynamics, traditional models relied on static snapshots or assumed homogeneous agent behavior, overlooking how diverse cognitive dispositions shape belief evolution. While some recent models introduce behavioral heterogeneity, they typically focus on macro-level patterns, neglecting the unequal and individualized dynamics that unfold at the agent level. In this study, we analyze an adaptive social network model where each agent exhibits one of three behavioral tendencies-homophily, neophily (attention to novelty), or social conformity-and measure the complexity of individual opinion trajectories using normalized Lempel-Ziv complexity. We find that the dynamics are often counterintuitive-homophilic agents, despite seeking similarity, become increasingly unpredictable; neophilic agents, despite pursuing novelty, exhibit constrained exploration; and conformic agents display a two-phase trajectory, transitioning from early stability to later unpredictability. Moreover, these patterns remain similar across diverse network settings, suggesting that internal behavioral dispositions - rather than external environment alone - play a central role in shaping long-term opinion unpredictability. The broader implication is that individuals' experiences of ideological volatility, uncertainty, or stability are not merely environmental, but may be endogenously self-structured through their own cognitive tendencies. These results establish a novel individual-level lens on opinion dynamics, where the behavioral identity of agents serves as a dynamical fingerprint in the evolution of belief systems, and gives rise to persistent disparities in dynamical experience within self-organizing social systems, even in structurally similar environments.

Figures (5)

• Figure 1: Functional shapes for the behavioral functions $F_h$ (left) and $F_a$ (right) for $\theta_h = \theta_a = 0.3$ (adopted from sayama2020extreme). The cyan planes indicate zero level.
• Figure 2: nLZ vs t for the three pure networks in Scenario 1. Homophilic (left) vs Neophilic (center) vs Conformic (right) node nLZ complexities (with 95% confidence bounds) are shown in distinct colors. The nLZ value is computed over increasing time intervals to capture changes in opinion trajectorial unpredictability over time.
• Figure 3: nLZ vs t for the three 50-50 networks in Scenario 2 (with 95% confidence bounds). Homophily vs Neophily (left); Neophily vs Conformity (center); Conformity vs Homophily (right).
• Figure 4: nLZ vs t for the mixed networks in Scenario 3 (with 95% confidence bounds).
• Figure 5: Causal intervention results for Scenario 3 (heterogeneous environment) at intervention time $t^*=600$. For each transition, the nLZ complexity of the focal node's opinion trajectory is shown for the baseline (no intervention; in blue) and intervention (in orange) cases, averaged across nodes and runs with 95% confidence intervals.