Monitoring the Opinion of the Crowd: Psychological Mechanisms Underlying Public Opinion Perceptions on Social Media

Abstract

Social media technologies offer several features that allow users to monitor other people’s opinions on public issues. Initial research showed that user-generated content can shape recipients’ perceptions of the majority opinion on societal problems. Still, it remains largely unexplored under which circumstances people gauge other users’ opinions through social media and whether perceived opinion climates affect people’s opinions and communication behavior in these environments. Results of a two-session experiment revealed that people’s fear of isolation sharpens their attention toward user-generated comments on Facebook which, in turn, affect recipients’ public opinion perceptions. The latter influenced subjects’ opinions and their willingness to participate in social media discussions. These findings are discussed in light of the spiral of silence theory and the social projection hypothesis.

Notes

1. While it was intended to invite only those who completed the first questionnaire to the second session, due to a technical error all 8,800 panelists who were invited in the first round (including those who did not complete the first questionnaire) were invited again.

2. Moreover, two participants were excluded since they gave unrealistic and uniform answers when asked to recall the number of positive and negative comments presented in the stimulus (e.g, “77777”); another participant was excluded due to the extreme value of the time spent viewing the main stimulus (4.75 h).

3. Descriptive examinations revealed that participants paid the greatest attention to peer comments, M = 4.57, SD = 1.84, followed by the main picture, M = 4.10, SD = 1.80, and the number of likes, M = 2.69, SD = 1.82. An ANOVA with repeated measures yielded a significant effect of the within-subject factor message type, Greenhouse–Geisser F(2, 1310.89) = 254.58, p < .001, η²_p = .28, revealing that self-reported attention differed between all message types at p < .001 (based on a post hoc comparison with Bonferroni correction).