Mapping the Spiral of Silence: Surveying Unspoken Opinions in Online Communities

Paper Title

Mapping the Spiral of Silence: Surveying Unspoken Opinions in Online Communities

Publication Info

• Topic area: Examining self-silencing behavior and its implications in online political communities.
• Keywords: Spiral of silence, self-silencing, online communities, Reddit, political discourse, opinion expression, community design, moderation, diversity, social media.

Background and Problem

• Problem / challenge: Social media is often treated as a reflection of public opinion, but discrepancies exist between publicly shared and privately held views. The "spiral of silence" theory suggests that individuals refrain from expressing minority opinions due to fear of social isolation, leading to distorted representation of viewpoints online.
• Significance: Understanding and mitigating self-silencing is crucial for fostering inclusive online communities and accurately representing public opinion, especially on contentious political and social issues.
• Motivation and related work: Prior studies have explored the spiral of silence in offline and platform-level contexts, but the role of community-specific factors (e.g., diversity, moderation) in shaping self-silencing on platforms like Reddit remains underexplored. This paper addresses this gap by focusing on community-level dynamics.

Solution

• Proposed approach: A hybrid human-AI method to study self-silencing on Reddit, combining large language models (LLMs) to generate community-specific controversial topics and surveys to measure opinion expression.
• Novelty:
  1. Development of a human-AI pipeline for generating and validating controversial topics specific to online communities.
  2. Empirical measurement of self-silencing across diverse political subreddits using survey data.
  3. Identification of community design factors (e.g., diversity, moderation) that influence self-silencing behavior.
  4. Recommendations for community-level interventions to mitigate self-silencing.
• Procedure and key techniques:
  1. Topic generation: LLMs propose controversial topics for subreddits based on community descriptions and rules. Topics are validated by active subreddit members.
  2. Survey design: Participants indicate their likelihood of sharing viewpoints on selected topics, their perception of majority opinions, and their perceived community values (inclusion, diversity).
  3. Analysis: Mixed-effects models assess the relationship between opinion incongruency, community factors, and likelihood of opinion expression.

Results

• Concrete findings:
  • Participants are less likely to share minority (incongruent) viewpoints (27.9%) compared to majority (congruent) viewpoints (47.2%).
  • Community diversity positively correlates with sharing incongruent viewpoints (β = 0.29, p = 0.03).
  • Higher content removal rates negatively correlate with sharing incongruent viewpoints (β = −0.31, p = 0.03).
  • Upvoting, an anonymous form of expression, is more common than commenting but still shows bias toward congruent viewpoints (83.0% vs. 65.4%).
• Advantage over baselines: The study provides a nuanced understanding of self-silencing by focusing on community-level factors, unlike prior work that primarily examines platform-level dynamics.
• Experiments / evaluation:
  • Surveyed 439 responses from 58 participants across 12 political subreddits and 11 topics.
  • Mixed-effects models analyzed the impact of opinion incongruency, community diversity, inclusion, and moderation on sharing behavior.
• Limitations and future work:
  • Reliance on self-reported measures may lack ecological validity.
  • Focus on politically-oriented subreddits limits generalizability to other types of communities.
  • Future work could explore experimental methods, additional platforms, and alternative operationalizations of self-silencing.

Summary

This study investigates the spiral of silence in online political communities, finding that individuals are less likely to share minority opinions due to perceived incongruency with majority views. Community diversity is associated with reduced self-silencing, while higher content removal rates exacerbate it. The authors propose actionable interventions, such as fostering diversity and increasing moderation transparency, to encourage opinion-sharing. By combining LLM-driven topic generation with survey-based measurements, the study provides a scalable method for analyzing self-silencing across diverse online communities. These findings have implications for designing more inclusive and representative online platforms.