Mixed News about the Bad News Game

Abstract

Basol et al. (2020) tested the "the Bad News Game" (BNG), an app designed to improve ability to spot false claims on social media. Participants rated simulated Tweets, then played either the BNG or an unrelated game, then re-rated the Tweets. Playing the BNG lowered rated belief in false Tweets. Here, four teams of undergraduate psychology students each attempted an extended replication of Basol et al., using updated versions of the original Bad News game. The most important extension was that the replications included a larger number of true Tweets than the original study and planned analyses of responses to true Tweets. The four replications were loosely coordinated, with each team independently working out how to implement the agreed plan. Despite many departures from the Basol et al. method, all four teams replicated their key finding: Playing the BNG reduced belief in false Tweets. But playing the BNG also reduced belief in true Tweets to the same or almost the same extent. Exploratory signal detection theory analyses indicated that the BNG increased response bias but did not improve discrimination. This converges with findings reported by Modirrousta-Galian and Higham (2023).

Keywords: Bad News Game; Fake news; gameified inoculation; social media.

Figure 1

Mean Truth Ratings for True and False Tweets Before and After Playing the Bad News Game (Inoculation) or Tetris (Control) in Each of the Four Replication Attempts.

Figure 2

Mean Truth Ratings for True and False Tweets Before and After Playing the Bad News Game (Inoculation) or Tetris (Control) Across the Four Replications and for False Tweets in the Basol et al. (2020) Data.

Figure 3

ROC Curves by Condition and Pre/Post. Note: Dashed line represents chance sensitivity, with curves bowing further from the dashed line representing higher sensitivity.

Figure 4

Response Bias (B”D) by Condition and Pre/Post. Note: Error bars = 95% CIs (between-subjects). Dashed line at 0 B”D indicates “neutral” response bias.

Figure 5

Change in Truth Ratings as a Function of Tweet Ground Truth and Initial Truth Ratings. Note: Red points = model-predicted means, red error bars = model-predicted 95% CIs, histograms = relative frequency of each post-inoculation effect in each pre-inoculation rating category.