Detecting Children’s Lies: Credibility Assessments and Nonverbal Facial Expressions

• In Event: Cultural Transmission of Information: Evidence of Epistemic Vigilance Across Children, Adults, and Non-human Informants

Abstract

Legal decision-makers (e.g., jurors, judges, police, social workers) are frequently tasked with evaluating the veracity of children’s reports to determine if an event or crime has occurred. However, research has shown that adults are very poor detectors of children’s lies (Gongola et al., 2017). Not being able to distinguish honest reports from dishonest reports can have severe implications, such as a child being kept in an unsafe environment or someone being wrongfully accused of a crime. It is therefore critical that researchers identify factors related to the effective identification of children’s lies. Here we present evidence from two studies examining this topic among human and automated computer-based lie detectors: Study 1 examined assessments of children’s credibility and adult’s lie detection accuracy, while Study 2 examined the use of nonverbal facial expressions in the automatic classification of children’s lies.

Adults’ decisions regarding the veracity of children’s reports (lie-judgements) are influenced by whether they perceive the child to be a credible source of information, based on their perceived trait honesty (e.g., trustworthiness) and cognitive ability (e.g., intelligence; Ross et al., 2003). However, it is unclear whether such credibility assessments are related to adult’s accuracy and bias in their lie-judgements (O’Connor et al., 2019). Study 1 examined this research question. Adults’ (N = 176; Mage=21.45) watched 8 videos of children (7-9 years-old) playing a game with an experimenter, half of whom lied about having cheated in the game. Participants first indicated how credible they believe each child to be (based on ratings of perceived trait honesty and cognitive ability) before judging whether they believe the child was telling the truth or a lie. Adults’ lie-detection accuracy and response bias (i.e., whether adults tend to label all children as liars/truth-tellers) was assessed using signal detection theory (MacMillan & Creelman, 1991). Credibility assessments were not significantly related to adults’ ability to accurately detect children’s lies. However, adults who believed children to be higher in trait honesty were biased towards thinking children were being truthful, while adults who believed children to be higher in cognitive ability were biased towards thinking children were lying (p<.05).

Adults in Study 1 achieved a lie-detection accuracy rate of 64% which, although higher than chance level (p<.05), is still quite low. The goal of Study 2 was to examine whether automated computer-based lie detection could achieve a higher accuracy rate than human judges by analyzing children’s nonverbal behaviors. Using computer vision technology, videos of children (4-9 years-old; N = 158) telling the truth or a lie regarding a transgression were automatically coded based on the Facial Analysis Coding System (Ekman & Friesen, 1978). Machine learning was then used to discriminate between liars and truth-tellers at an accuracy rate of 73% – significantly above both chance levels and human accuracy (p<.05). Two emotions, surprise and fear, were more strongly expressed by liars than truth-tellers. These findings provide evidence to support the use of automatically coded facial expressions to detect children’s lies. Theoretical and practical implications of both studies will be discussed.

Authors

• Sarah Zanette, University of Regina
  Presenting Author

• Kaila Bruer, University of Regina
  Non-Presenting Author

• Makenzie Furlong, University of Regina
  Non-Presenting Author

• Xiao Pan Ding, National University of Singapore
  Non-Presenting Author

• Thomas D Lyon, University of Southern California
  Non-Presenting Author

• Kang Lee, University of Toronto
  Non-Presenting Author