Collective patterns of social diffusion are shaped by individual inertia and trend-seeking

Social conventions, like language or greeting customs, are vital aspects of society. Their value stems from widespread adoption, enabling coordination. Social diffusion is a key mechanism driving convention change, where a minority promotes an alternative, potentially replacing the status quo. Individuals play varying roles: a committed minority actively promotes the alternative, explorers test it, and non-explorers adopt it later if a critical mass is reached. While population models like the Bass model effectively describe macroscopic diffusion patterns (the S-curve of adoption), they lack explanatory power regarding underlying individual-level dynamics. Agent-based models (ABMs), offering a microscopic perspective, have become increasingly popular. They allow for hypotheses about how individual-level dynamics lead to macroscopic phenomena. Game theory provides a framework for modeling human decision-making during social diffusion, focusing on social coordination. However, existing ABMs often neglect crucial behavioral mechanisms: inertia (resistance to change) and trend-seeking (following population-level changes). This paper argues that these omissions prevent accurate modeling of real-world patterns, like the delay before take-off followed by explosive diffusion. It proposes a game-theoretic ABM incorporating these mechanisms, validated through a multi-round experiment and subsequent simulations.

Existing literature highlights the importance of social coordination in models of social diffusion. However, the ubiquitous presence of inertia and trend-seeking is often overlooked. Inertia, or status quo bias, refers to the tendency of individuals to stick with their current decisions. Trend-seeking, or sensitivity to dynamic norms, represents the tendency to follow population-level changes. Empirical evidence supports the presence of both mechanisms across a range of decision-making scenarios. Existing agent-based models, including game-theoretic models, typically do not consider these mechanisms, leading to discrepancies between model predictions and real-world observations.

To address the limitations of existing models, the researchers conducted a multi-round online experiment involving 180 recruits divided into 20 groups of 8-10 participants each. The experiment simulated a company board deciding on product investment, with 2-4 committed minority bots influencing the decision-making process. The bots helped establish a status quo (Stage I), then switched to promote an alternative (Stage II). The experiment tracked individual strategy choices across rounds, aiming to identify the influence of inertia and trend-seeking. A regression analysis with fixed effects was performed to examine the effects of the previous round's choices, the adopters in the previous round, and the overall group trend on the current round's choices. Additional tests focused separately on inertia and trend-seeking. The results of the experiment motivated the inclusion of inertia and trend-seeking in a new agent-based model. The model uses a game-theoretic framework with a log-linear learning formula to compute the probability of strategy adoption. The payoff function incorporates social coordination, inertia, and trend-seeking. Parameters for the model were calibrated using data from the experiment, differentiating between explorers (less influenced by inertia, more by trends) and non-explorers. Numerical simulations were used to explore how individual-level inertia and trend-seeking shape collective patterns of social diffusion.

The experiment revealed a heterogeneous distribution of switching rates among participants, suggesting a distinction between explorers and non-explorers. Regression analysis confirmed the significance of coordination, inertia, and trend-seeking in individual decision-making (R² = 0.46, p < 0.001). The model's simulations showed that inertia creates a delay before the diffusion process takes off, while trend-seeking results in explosive diffusion once the process begins. The length of the delay is critical in determining whether diffusion occurs, which is significantly impacted by the composition of the population (the fraction of committed minority and explorers). The model found that a committed minority comprising at least 25% of the population is sufficient to trigger social change, corroborating existing research. However, when the committed minority is below 25%, sufficient sensitivity to trends (high proportion of explorers) can still unlock social change. Monte Carlo simulations, varying population size and the fraction of explorers, confirmed that the presence of explorers is crucial for diffusion in large populations. Without explorers, the take-off time grows substantially with population size. Conversely, with sufficient explorers, the take-off time remains moderate and independent of population size. The transition time (from take-off to widespread adoption) remained consistently small, independent of population size, highlighting the inherent explosiveness of the model due to the trend-seeking mechanism. The simulations consistently produced moderate switching activity and heterogeneous distribution, mirroring the experimental data.

The findings demonstrate that incorporating inertia and trend-seeking into ABMs of social diffusion is crucial for capturing both macroscopic features (delay and explosiveness) and microscopic behavior (heterogeneous switching rates). This contrasts with existing models which, due to their omission of these mechanisms, struggle to reconcile individual and population-level observations. The model's ability to simultaneously replicate these features highlights the complex interplay between inertia and trend-seeking. The critical role of explorers in unlocking diffusion, especially when the committed minority is below the 25% threshold, is a key contribution. The model's predictive power extends beyond the specific experimental setting and is relevant to various scenarios of collective decision-making.

This research contributes a novel agent-based model of social diffusion that successfully incorporates the effects of inertia and trend-seeking, resolving inconsistencies in previous models. The model accurately captures both the macroscopic features (explosive diffusion, delayed take-off) and microscopic behavioral patterns (heterogeneous switching rates) observed in real-world social diffusion. Future research could explore the relationship between the model's parameters and the tipping point concept, extend the model to scenarios with unequal benefits between status quo and alternative, and consider diverse cultural contexts and network structures.

The study's limitations include the use of a specific experimental setup with a simplified interaction structure (all-to-all communication) and a focus on a limited cultural background (native English speakers). While the model captures key aspects of social diffusion, further investigations with other interaction mechanisms (asynchronous updates, pairwise interactions) and cultural contexts could provide a more comprehensive understanding. Future studies may also explore the influence of more nuanced concepts of inertia and trend-seeking.