Towards an evidence-based approach for interpreting eyewitness expressions of confidence

Research Institution / Organisation

Queen Margaret University

Principal Researcher

Pia Pennekamp

Level of Research


Project Start Date

September 2019

Research Context

Eyewitnesses are often asked to make an identification from a lineup containing a suspect among fillers (i.e. people that are not suspected of having committed a crime). Crucially, this piece of evidence can be a decisive factor for investigations and in applied settings, even more so when the eyewitness identifies a suspect. Ultimately, when an eyewitness makes an identification, they provide evidence that the suspect is guilty (by identifying him/her) or innocent (by not identifying him/her). Eyewitness accounts, especially when given with high confidence, are often more persuasive than any other type of evidence (Brewer & Burke, 2002). Memory, however, is reconstructive, easily influenced by external sources and, above all, prone to error (Loftus, 1981). In legal settings, this fallibility of memory has detrimental consequences. Mistaken eyewitness identifications were involved in 70% of 362 later exonerated convictions based on DNA in the United States (Innocence project, 2019). 

The reliability of eyewitness evidence is commonly determined using eyewitness confidence. Research shows that eyewitness confidence can be predictive of accuracy when it is immediately obtained after a lineup decision has been made (Wixted & Wells, 2017). Yet, the body of literature on how to interpret confidence judgments in relation to eyewitness evidence is limited. In practice, confidence is typically obtained verbally (in the witness' own words; NAS, 2014). Research suggests that individuals prefer to give confidence verbally rather than numerically. However, verbal probability estimates are more likely to be misinterpreted than numeric estimates.  

To accurately interpret verbal confidence statements, we need a reliable way to obtain and interpret eyewitness identification confidence. Fields like climate science and intelligence analysis use lexicons (i.e., dictionaries), which translate verbal confidence (e.g., "likely") into a range of numeric values (e.g., 40%-75%).

The proposed research involves creating and testing a lexicon for confidence in lineup decisions. To minimize miscommunication, this research will provide preliminary data on the practical performance of an evidence-based lexicon for eyewitness identification confidence. The outlined project tests the utility of an evidence-based approach in the eyewitness context and extends the existing literature on the use of lexicons.

Research Methodology

How should verbal confidence statements be interpreted to minimize the miscommunication of eyewitness confidence?

My PhD project uses quantitative approaches and will span across three studies.  The studies are registered on the Open Science Framework. Samples sizes and analyses are pre-registered.

  1. Does the order (verbal, then numeric vs. numeric, then verbal) of confidence statements matter? Open Science Framework (OSF) Pre-registration 

  2. Creation of lexicon.  Open Science Framework (OSF) Pre-registration

  3. Validation of lexicon

The experimental design will depend on results of study 2 (creation of lexicon). Sample sizes and analyses will be pre-registered on the Open Science Framework.

Date due for completion

September 2022
Return to Research Map