Social Influence: The Usage History Heuristic

Bary PradelskiSSRN 2022 · DOI

TL;DR

This paper introduces and analyzes a new model of social influence, "Usage History," where individuals' decisions are influenced by the recent frequency of others' actions, contrasting it with the traditional "Adoption Ratio" model that uses the proportion of adopters. The authors demonstrate that these two models can lead to significantly different long-term outcomes, with Usage History favoring more extreme societal behaviors.

What Problem Does This Paper Address?

Understanding how aggregate behavior emerges from individual decisions is a fundamental question in many fields. Social influence, where individuals are swayed by the actions of others, is a key driver of phenomena like fashion trends, product adoption, and opinion shifts. However, existing models often simplify the information individuals use to gauge social influence. This paper addresses the gap by distinguishing between two distinct ways people might perceive social influence: by looking at the overall proportion of people who have adopted a behavior (Adoption Ratio) versus by observing how frequently a behavior has been performed recently (Usage History). The authors argue that this distinction is crucial because it can lead to different predictions about societal outcomes and inform the design of interventions.

Methods

The paper develops a formal model of social influence in a heterogeneous population where agents update their actions asynchronously. The core of the methodology involves:

  • Utility Functions: Defining agent utility based on an inherent preference and a social conformity component, with the latter depending on an aggregate statistic of others' actions.
  • Two Information Regimes:
    • Adoption Ratio: Agents observe the proportion of players who have adopted action 1 (e.g., the proportion of iPhone users).
  • Usage History: Agents observe the average of the last k actions taken by the population.
  • Dynamic Analysis: Studying the stability of equilibria and long-run behavior of these processes under both unperturbed (best-response) dynamics and perturbed dynamics (introducing a small probability of random action choice, or "tremble").
  • Stochastic Stability: Employing tools from evolutionary game theory, specifically stochastic stability, to analyze the long-term behavior of the perturbed dynamics and identify the most likely states.
  • Empirical Discrimination: Deriving characteristic "footprints" in the distribution of the aggregate statistic for each model, which can be used to empirically distinguish between them.

Datasets

Not discussed in the paper. The paper focuses on theoretical modeling and analysis, using hypothetical examples to illustrate its concepts rather than empirical datasets.

Key Insights and Ideas

  • Usage History vs. Adoption Ratio: The distinction between observing the proportion of adopters and the frequency of recent actions is not just a theoretical nuance; it fundamentally alters the dynamics and potential outcomes of social influence.
  • Extreme Outcomes: The Usage History model, by focusing on recent actions, can lead to more extreme societal outcomes (e.g., near-universal adoption or rejection) compared to the Adoption Ratio model, which tends to favor more moderate equilibria.
  • Intervention Design: Understanding which influence process is at play is critical for designing effective interventions. For Adoption Ratio, promoting adoption (e.g., purchase incentives) is key, while for Usage History, encouraging the act of performing the behavior (e.g., usage incentives) is more appropriate.
  • Characteristic Footprints: The paper provides a framework for empirically distinguishing between the two models by analyzing the statistical properties (e.g., distribution, variance) of observed aggregate behavior.

Key Results

  • Equilibria Identification: The paper identifies the stable equilibria for both the Adoption Ratio and Usage History models.
  • Stability Analysis: It demonstrates that under perturbed dynamics, the Usage History model favors more extreme equilibria (less mixed states), while the Adoption Ratio model's stability is less dependent on how mixed the state is.
  • Empirical Distinguishability: The paper shows that the aggregate observed behavior (e.g., the proportion of action 1) will have different statistical properties (e.g., binomial distribution for Usage History, non-binomial for Adoption Ratio) under the two models, allowing for empirical discrimination.

Limitations and Future Work

  • Model Simplifications: The model assumes a binary action space and linear influence. More complex action spaces or non-linear influence mechanisms could lead to different results.
  • Heterogeneity: While heterogeneity in tipping values is considered, other forms of heterogeneity (e.g., in conformity strength or information processing) are not explicitly modeled.
  • Network Structure: The analysis assumes a well-mixed population where everyone can observe everyone else. Future work could explore how network structures influence these dynamics.
  • Empirical Validation: While the paper provides tools for empirical discrimination, it does not conduct such an analysis itself. Future work could apply these tests to real-world data.

Summary updated

BibTeX
@misc{pradelski2022social,
  author = {Bary Pradelski},
  title = {{Social Influence: The Usage History Heuristic}},
  year = {2022},
  doi = {10.2139/ssrn.4226526},
  howpublished = {SSRN},
}
Figure