Fig. 4: Trivial to small effects of hostility on the persuasiveness of misinformation and rebuttal.

Internal meta-analyses of neutral misinformation versus hostile misinformation (a, b) and neutral rebuttal versus hostile rebuttal (c, d) on intention and attitude of the audience using random effects models. Negative effect sizes indicate that the audience’s attitude towards a behaviour dismissed by science deniers (b, d) or the intention to perform this behaviour (a, c) was lower when the respective speaker used hostile compared to neutral language. Thus, for the advocate (c, d), negative effect sizes indicate a decrease in the persuasive power of rebuttal when hostile language was used, while for the denier (a, b), positive values indicate a decrease in the persuasive power of misinformation when hostile language was used. The y-axes represent experiments, and the x-axes represent Cohen’s ds. Mean differences are adjusted for baseline values and standardisers are pooled standard deviations for all of the cells in the respective design⁵⁶. Diamonds show averaged effects. Red error bars indicate 95% confidence intervals. Blue error bars indicate 90% confidence intervals. The dotted central line is the zero line. If the red confidence interval does not include the zero line, then the effect is significantly different from zero. The left (−0.2) and right (0.2) dotted lines mark the equivalence bounds. If the effect lies between these lines and the blue confidence interval does not include any of the equivalence bounds, then the result is significantly equivalent, i.e., trivial. Q tests and I² indicate heterogeneity of results.