Thinking more or thinking differently? Using drift-diffusion modeling to illuminate why accuracy prompts decrease misinformation sharing

Contributions

• Mechanistic decomposition: Drift-diffusion modeling separates accuracy prompts' effects into two competing mechanisms: increased deliberation (larger decision boundary) vs. shifted attention to accuracy (increased weight on accuracy). The study tests these competing hypotheses with unprecedented precision.
• Rejection of deliberation account: Accuracy prompts do NOT increase the amount of deliberation (no significant effect on boundary parameter), contradicting a common interpretation of dual-process theories.
• Support for attention account: Accuracy prompts significantly increase the weight participants place on perceived headline accuracy during deliberation (effect b = 0.19, supporting Hypothesis 2), showing that prompts work by reweighting information, not by increasing cognitive effort.
• Robustness across formats: Effects hold across binary and Likert response scales, partisan and COVID-19 headlines, and multiple recruitment platforms (Amazon Mechanical Turk and Lucid), suggesting the mechanism is fundamental.

Method

Participants and design: Six experiments (N = 5,633 after exclusions, from 6,961 total recruited via Amazon Mechanical Turk and Lucid) with random assignment to control (news-sharing task only) or treatment (accuracy prompt before sharing task) conditions.

Task: Participants evaluated true and false news headlines on sharing likelihood using 6-point Likert scales or binary responses. False headlines came from fact-checking sites (e.g., Snopes); true headlines came from mainstream news. Headlines covered politics and COVID-19.

Accuracy prompt: In the treatment condition, participants first rated the accuracy of a neutral headline, drawing attention to accuracy before the main sharing task.

Computational model: Multi-attribute drift-diffusion model (DDM) with decision boundaries for "share" (upper) and "do not share" (lower). The model decomposes each sharing decision into five parameters:

• Decision boundary: Distance between boundaries (larger = more deliberation)
• Weight on accuracy (β₁): How much perceived headline accuracy influences the drift rate
• Drift rate (β₀): Default decision value
• Starting-point bias: Initial inclination before deliberation
• Non-decision time: Stimulus encoding and motor response time

The key equation: v_t = β₀ + β₁ accuracy_t + ε_t, where changes in the drift rate over time capture deliberative adjustment.

Analysis: Bayesian mixed-effects models with varying intercepts and slopes by study and platform; posterior density intervals (95% and 89% HPD); Bayes factors for hypothesis testing.

Results

Sharing outcomes (replicating prior work): Accuracy prompts increased sharing discernment (proportion true shared minus proportion false shared) by approximately 5% (b = 0.05, 95% CI [0.004, 0.08], d = 0.20, BF = 3.90). No meaningful difference in overall response times (b = 0.23, 95% CI [−0.12, 0.60], BF = 0.47).

Model results—boundary parameter (deliberation account): Accuracy prompts did NOT significantly increase the decision boundary (b = −0.06, 95% CI [−0.24, 0.14], d = −0.02, BF = 0.13). This null effect on boundary contradicts Hypothesis 1 (the deliberation account) and suggests accuracy prompts do not cause people to deliberate more.

Model results—weight-on-accuracy parameter (attention account): Accuracy prompts significantly increased the weight placed on headline accuracy during deliberation (b = 0.19, 95% CI [0.02, 0.37], d = 0.19, BF = 3.35). The Bayesian posterior density was concentrated at positive values, supporting Hypothesis 2. This effect was robust across response formats (binary and Likert) and a preregistered robustness check using independent accuracy ratings.

Effect heterogeneity: Across studies, the weight-on-accuracy effect was positive in all but one study, though effect sizes varied (ranging from small to moderate across platforms). Platform and condition-platform interactions on both boundary and weight parameters were null.

Notes

Strengths: The drift-diffusion approach is methodologically novel in the misinformation literature and enables a direct test between two plausible but mutually exclusive theoretical mechanisms. The sample is large and preregistered, reducing p-hacking risk. The models are interpretable (parameters map cleanly onto psychological constructs), and results are robust across response scales and headline types. The null effect on deliberation (boundary parameter) is as informative as the positive effect on weight-on-accuracy, because it rules out a leading alternative explanation.

Limitations: (1) Response times in a lab setting may not reflect real social-media contexts, where distraction and limited attention are more pronounced. (2) The effect on weight-on-accuracy is small to moderate (b = 0.19), and the variation across studies suggests the mechanism may be context-sensitive. (3) All studies are U.S. online-recruited samples; cross-cultural and field-settings generalizability remains untested. (4) The paper does not test whether other misinformation interventions (media literacy, fact-checks, source-credibility cues) operate via similar or different mechanisms.

Broader implications: This paper's core insight—that "thinking better" does not require "thinking more"—has implications beyond accuracy prompts. It suggests that many misinformation phenomena may stem not from inability or unwillingness to deliberate, but from the saliency and weight of available information. This reframes interventions away from trying to force deeper processing and toward redesigning what gets attended to in information-rich environments.