Fake news, fast and slow: Deliberation reduces belief in false (but not true) news headlines

Contributions

• First causal evidence directly manipulating deliberation to measure its impact on fake news belief, isolating deliberation from mere cognitive ability.
• Demonstrates that deliberation reduces belief in false headlines under time pressure and working memory load, consistent with System 1 (intuitive) errors being corrected by System 2 (deliberative) processing.
• Shows political polarization is equivalent across intuitive and deliberative responses, contradicting the Motivated System 2 Reasoning (MS2R) hypothesis that deliberation magnifies partisan bias.
• Provides empirical support for the classical reasoning account over the MS2R account in the context of media truth discernment.

Method

Participants (N=1,635 from MTurk; 1,012 in main experiment, 503 females, Mage = 36.9 years) rated 16 actual news headlines for accuracy. Headlines varied on two dimensions: veracity (true vs. false) and political alignment (Republican-consistent, Democrat-consistent, or neutral). The study employed three experimental conditions:

1. Two-response experiment (main condition): Participants made an initial response under time pressure (7 seconds) and concurrent working memory load (memorizing a dot pattern). They then re-rated the same headlines with no time pressure or load, enabling deliberation.
2. One-response baseline: Control condition where participants made a single, unconstrained judgment.
3. Politically neutral pre-test: Separate participants (N=623) rated neutral headlines to establish that deliberation improves accuracy when political motivation cannot explain the pattern.

Analyses used mixed-effect logistic regression models with headlines and subjects as random intercepts. The preregistered sample was 1,000 participants; final N=1,012 after accounting for exclusion criteria (missed initial response deadline, incomplete political ideology data).

Results

Politically neutral headlines: Deliberation significantly increased accuracy for false headlines. The interaction between headline veracity and response number was significant (b = 0.47, 95% CI = [0.29, 0.65], p < 0.0001); perceived accuracy of false headlines increased from ~37% in initial (intuitive) responses to ~31% in final (deliberative) responses, and to ~29% in the one-response baseline. This pattern is opposite to what MS2R predicts and consistent with a classical reasoning correction mechanism.

Political headlines (main analysis): Within-subject comparison of initial versus final responses revealed a significant interaction between headline veracity and response number (b = 0.36, 95% CI = [0.2, 0.52], p < 0.0001). Final responses rated false headlines as less accurate than initial responses. Critically, there was no interaction between political concordance and response number (b = 0.004, 95% CI = [-0.16, 0.17], p = 0.96), contradicting MS2R's prediction that deliberation would increase partisan bias. Political concordance affected accuracy ratings (people rated concordant headlines as more accurate), but this effect was equivalent across intuitive and deliberative responses.

Between-subject analysis: Comparing the two-response experiment to the one-response baseline further supported the classical account. Deliberation significantly improved truth discernment for both discordant and concordant headlines, with no evidence of increased partisan bias under deliberation.

Dispositional reasoning: Participants scoring higher on the Cognitive Reflection Test (CRT) showed stronger relationships between deliberation and truth discernment, though both high- and low-CRT participants benefited from deliberation.

Notes

This is among the strongest empirical pieces in the literature on the role of thinking style in fake news belief. The two-response paradigm is a clever methodological choice—it isolates the causal effect of deliberation itself while controlling for individual differences in cognitive ability. The key finding—that deliberation improves accuracy without amplifying partisan bias—directly contradicts a prominent theory (MS2R) and provides causal support for the classical reasoning account. The authors appropriately note that their sample is non-nationally representative (MTurk participants are not ideologically balanced), but the use of political concordance/discordance within participants sidesteps the need for a balanced sample; comparing across concordance conditions collapses across Democrat/Republican, making the test of MS2R robust. The main limitation is reliance on online participants with news headlines, which may not generalize to real-world, ecologically valid misinformation exposure. Future work testing the impact of deliberation on belief persistence (the continued influence effect) and real-world sharing behavior would be valuable.