Effect of Feedback on the Accuracy of Future News Assessments
This graph measures the intent-to-treat effect of negative feedback on future news veracity assessments. Coefficients correspond to P(True | True News; Treatment) - P(True | True News; Control) and P(Fake | Fake News; Treatment) - P(Fake | Fake News; Control).
The first row shows that treated subjects are more likely to correctly detect whether news is true or fake. The second and third rows show that this improvement is entirely driven by treated subjects becoming less likely to believe that Fake News is truthful.
Error bars indicate 95% confidence intervals.