The relationship between eyewitness memory confidence and accuracy deteriorates over time. While early recall opportunities can protect memory from decay, less is known about their potential to preserve confidence and the confidence-accuracy alignment against the effects of time. This in-person study investigates whether an initial recall test mitigates the impact of time on confidence, confidence-accuracy calibration, and memory clarity, as well as whether this potential impact depends on the type of initial recall test.
Sixty participants viewed a mock-crime video and 24 hours later recalled the event in response to a Free Recall invitation or a Forensic Interview. A third (control) group did not recall the event at this stage. A week later, all participants answered memory questions and provided (i) confidence ratings for each response, (ii) a global measure of confidence, and (iii) judgments on the clarity of their memory for the details recalled.
Results and implications are discussed.

When eyewitnesses make lineup identifications, police in many countries are instructed to collect confidence statements. While previous research has focused on numeric confidence (e.g., “80% sure”), recent interest has shifted to verbal statements (e.g., “I’m very certain”). In previous work, we have shown that verbal and numeric confidence provide non-overlapping diagnostic information (Seale-Carlisle et al., 2022). But, the cognitive mechanisms behind this remain unclear. In this presentation, we examine three potential accounts: a single-process account (granularity), a dual-process account (recollection sensitivity), and a metacognitive noise account (multiple ratings). We conclude that a single-process account is insufficient, as the size of the numeric scale has little effect on the diagnostic value of verbal confidence (Seale-Carlisle et al., 2024). We propose additional studies to test the other accounts and finish by discussing whether AI-driven verbal classifiers could help human evaluators (e.g., police, jurors) to better understand witnesses’ verbal confidence.

Eyewitness confidence is more reflective of identification accuracy than previously believed; however, challenges still exist. We will present a replication of our prior work demonstrating that what people believe constitutes low, medium, or high confidence varies considerably regardless of whether eyewitnesses provide confidence numerically or verbally. In the replication, participants’ best numeric estimates of what constitutes low (M = 24.82%, SD = 15.25), medium (M = 48.77%, SD = 9.27), and high (M = 84.50%, SD = 9.78) confidence were highly variable. Likewise, interpretations of phrases commonly believed to indicate low, medium, and high confidence were highly variable. For example: “not quite sure” (M = 24.9%, SD = 15.29), “pretty sure” (M = 60.1%, SD = 14.63), and “certain he/she did it” (M = 90.3%, SD = 10.05). There is an urgent need for tools that can create common ground between eyewitnesses and those who must interpret their confidence judgments.

The pristine conditions hypothesis postulates that highly confident witnesses are highly accurate, even when witnessing conditions are poor. Since then, research in adults has shown that, in some poor witnessing conditions, the high confidence-accuracy relationship breaks down. Researchers have concluded that this relationship applies to children in good witnessing conditions. We tested how the witnessing condition of exposure duration impacted this relationship in children (5-8-yrs = 485, 9-13-yrs = 357) and adults (N = 213). Participants viewed a 360-degree live-action mock-crime video in a VR headset. Surprisingly, memory strength was weak under all witnessing conditions. There were so few high confidence identifications in adults that the confidence-accuracy relationship could not be plotted. Importantly, we found that the pristine conditions hypothesis does not hold in children regardless of the state of the witnessing condition. This suggests that there are important boundary conditions to the pristine conditions hypothesis.