Cognitive Ability, Overconfidence, and Confirmation Bias in Lie Detection

This study examined individuals' ability to detect lies and truths in short videos using coin-toss reports across three pre-registered experiments. Results showed detection rates only slightly above chance, with participants better at identifying truths than lies, and revealed that cognitive ability does not improve detection accuracy while feedback increases overconfidence without enhancing performance.

Three pre-registered experiments where participants judged whether statements in 20 short videos describing coin-toss outcomes were true or false, with analysis of cognitive ability, overconfidence measures, facial expression cues, and feedback effects on detection and sharing behavior.

Background & Context

The digital age has created unprecedented challenges for evaluating the veracity of video-based information, particularly with the proliferation of short-form video content on social media platforms. This study examines individuals' ability to distinguish truthful from deceptive communication in short videos, addressing a critical gap in understanding how cognitive biases affect lie detection in modern media contexts.

While traditional deception detection research has consistently shown human accuracy rates averaging 54% correct lie-truth discrimination, the specific dynamics of video-based lie detection involving cognitive ability, overconfidence, and confirmation bias remained underexplored. This research builds on established findings that people feel overconfident in their evaluation despite low detection accuracy, extending this to the domain of short video content.

The study's focus on coin-toss reports provides a controlled paradigm for examining deception in a context relevant to misinformation spread online, where individuals must rapidly judge video content authenticity with limited information.

Research Design & Methodology

The researchers conducted three pre-registered experiments using coin-toss reports, ensuring methodological rigor and transparency. Participants were tasked with identifying whether statements made in 20 short videos describing the outcome of a coin toss were true or false, creating a realistic yet controlled deception detection scenario.

The experimental design incorporated multiple dimensions of analysis:

• Assessment of detection accuracy for both truthful and deceptive videos
• Measurement of cognitive ability and its relationship to detection performance
• Evaluation of overconfidence through self-reported detection ability versus actual accuracy
• Emotional and facial expression analysis to identify subtle emotional cues in videos
• Feedback conditions that heightened cognitive confidence to assess effects on overconfidence and recommendations to others

The pre-registration of all three experiments represents a methodological strength, reducing publication bias and ensuring analytic transparency in examining the interplay between cognitive factors and deception detection.

Results & Key Findings

Detection rates were only slightly above chance, with higher accuracy for truthful than deceptive videos. More specifically, individuals performed slightly better than chance when detecting truthful videos, but their accuracy in detecting lies was worse than random, revealing an asymmetric pattern in detection capability.

The study revealed significant gender differences and the role of overconfidence:

• Gender effects: Females outperformed males, while males displayed greater overconfidence
• Cognitive ability: Cognitive ability does not improve detection, but feedback that heightens cognitive confidence increases overconfidence and recommendations to others
• Facial cues: Favorable first impressions and facial cues influence judgments, yet reliance on such signals proves ineffective
• Emotional analysis: Subtle emotional cues in videos provided some signals of truthfulness, but participants failed to effectively utilize these cues

These findings highlight systematic biases that undermine accurate evaluation of video-based information, suggesting that even educated observers struggle with video-based lie detection and that confidence calibration is particularly problematic.

Discussion & Significance

This research makes a critical contribution to understanding how cognitive biases shape video-based deception detection in the digital information age. The finding that cognitive ability fails to improve detection accuracy challenges assumptions that more intelligent individuals are better lie detectors, while the demonstration that feedback increases overconfidence without improving accuracy has important implications for training programs.

The gender differences align with broader patterns in deception detection literature, where female participants generally outperformed male participants in accuracy, while males exhibited higher levels of overconfidence in their ability to detect lies. The asymmetry in detecting truths versus lies—performing above chance for truths but below chance for lies—suggests a truth bias that may be exploited by deceivers in video content.

The study's demonstration that people rely on facial cues and first impressions despite these being ineffective indicators points to the persistence of folk psychology beliefs about deception. This has profound implications for the spread of misinformation through video platforms, where confident but inaccurate judgments may lead to viral sharing of deceptive content.

Limitations & Considerations

The use of coin-toss reports, while providing experimental control, represents a relatively low-stakes deception scenario that may not generalize to high-stakes lies involving significant emotional investment. The facial cues and behavioral patterns in such controlled laboratory deceptions may differ substantially from real-world video misinformation where deceivers have time to prepare and rehearse their presentations.

The study's focus on short videos of coin-toss outcomes limits ecological validity when considering diverse types of video content encountered online, from political statements to product testimonials. Additionally, the specific characteristics of the video stimuli—length, production quality, and sender demographics—may influence generalizability to the broader landscape of digital video content.

Practical Applications

These findings carry urgent implications for combating video-based misinformation on social media platforms. The demonstration that overconfidence increases with feedback suggests that simply exposing people to more video content may paradoxically worsen their calibration, leading them to share misinformation more confidently. Platform designers and media literacy educators should focus on reducing overconfidence and encouraging epistemic humility rather than promoting faith in intuitive judgments.

For investigators and fact-checkers, the results underscore that relying on facial cues and first impressions when evaluating video authenticity is counterproductive. Training programs should emphasize statistical approaches, content verification, and source checking rather than enhancing confidence in behavioral cue interpretation. The gender differences also suggest that diverse teams may bring complementary strengths to video content verification tasks.