A four‑week field experiment conducted on Reddit examined how online toxicity shapes participation in political discussions, revealing that polite environments prompt silent readers to post for the first time, while toxic debates drive the most active users to post more. The study appears in Science Advances and was led by Lisa Oswald of the Max Planck Institute for Human Development, with colleagues William Small Schulz and Philipp Lorenz‑Spreen.
Online spaces central to political discourse often show a participation gap: a vocal minority creates most content while the majority observe in silence. Researchers refer to these highly active individuals as power users, while the silent majority are lurkers who rarely post.
In the study, 520 participants from the United States were recruited through advertisements on Reddit and randomly assigned to one of six private Reddit communities created for the experiment. After an initial survey about political views and online habits, participants discussed a series of topics over four weeks.
Over four weeks, the groups were prompted to debate 20 political issues, covering immigration, gun control, universal basic income and artificial intelligence, among others.
To test strategies for closing the participation gap, two communities served as a control with standard platform settings. Two others offered a financial incentive: two dollars for every day a participant wrote a substantive comment.
The final two communities operated under a moderation regime, with frequent reminders to stay civil and respectful alongside automated spam filters.
Participants completed weekly surveys about their experiences and perceptions of group dynamics, and were categorised as silent (never commenting) or active (writing at least one comment). For the active users, researchers tracked the total number of comments posted.
The results indicated a mirrored psychological mechanism governing initial participation and subsequent posting intensity. Silent participants were more likely to view the group as polarised and toxic, and they preferred environments that felt respectful, knowledgeable and constructive.
By contrast, active users posted more frequently when they perceived the group as polarised, toxic and lacking in knowledge.
The researchers concluded that an abrasive environment suppresses the voices of the silent majority while enabling a less sensitive minority to dominate the conversation. They noted that the interaction between the objective level of toxicity and a user’s personal sensitivity to hostility helps determine behaviour.
Interventions to improve conversation quality yielded mixed results. Financial incentives did encourage more people to speak up, narrowing the gap between lurkers and power users. Direct rewards successfully motivated hesitant participants to join the discussion.
However, the moderation condition failed to change participation levels or improve discourse quality. Reminding users to stay civil did not reduce the participation gap, nor did it lower the overall toxicity of comments, suggesting that simply urging politeness may not alter underlying dynamics.
The study also examined how social feedback influenced behaviour. Positive votes on comments were associated with an increase in posting the following day, while negative feedback via downvotes reduced subsequent participation.
Over the four weeks, there were no statistically significant differences in political attitude changes between treatment and control groups. Measures of political trust and polarization remained relatively stable regardless of group rules or incentives.
Descriptively across all groups, participants reported lower political interest and issue knowledge by the study’s end, a finding the researchers attributed to an intimidation effect. The comments during the study were longer and more detailed than typical Reddit posts, which may have discouraged some participants from adding further input.
The experiment took place in a tightly controlled environment that differs from organic social media use. The participants were predominantly liberal, and the overall toxicity of comments was relatively low from the outset, leaving limited scope for the moderation treatment to demonstrate an impact.
The researchers suggest future work could explore whether stricter rules, such as banning hostile users, yield different outcomes, and to investigate how these dynamics play out on other platforms with different community structures and algorithms.
The study, “Disentangling participation in online political discussions with a collective field experiment,” was authored by Lisa Oswald, William Small Schulz and Philipp Lorenz‑Spreen.
