When Crowds Fail: Predicting Failures in Collective Wisdom through Discourse Cues

In today's increasingly interconnected world, the ability to anticipate future developments has become more critical than ever. While individual predictive abilities are often limited by cognitive biases, the wisdom of crowds—the aggregation of human judgments—has shown remarkable success in improving prediction accuracy. Group discussion can have mixed effects on this success. On one hand, it can enhance collective intelligence through information sharing; On the other hand, it may also undermine crowd wisdom by introducing social pressures that reduce independence and diversity of thought. Using data from a community forecasting site, we developed an interpretable, predictive model to investigate how structural and linguistic-psychological characteristics of discourse affect crowd accuracy. A simple model with 14 independent variables explained 28.6% of the variance in group prediction accuracy on a held-out test set. Notable predictors included a high comments-to-predictions ratio and informal language use (e.g., profanity, religious references, speech disfluencies), which correlated with increased crowd error. Emotional language also played a role, with the use of exclamation marks associated with increased prediction error, while anxiety-related language and detachment showed a modest association with decreased error rates. These findings demonstrate how discourse markers can serve as early indicators of potential crowd prediction failures, offering a practical tool for calibrating collective intelligence and enhancing forecasting accuracy.