High eyewitness confidence is always compelling: that’s a problem

ABSTRACT

Recent research shows a strong positive relationship between eyewitness confidence and identification accuracy, assuming the confidence judgment results from a first, fair test of memory. The current study examines whether jurors understand this relationship, and the boundary conditions under which this understanding holds. Mock jurors read a trial transcript in which we manipulated the eyewitness’ level of confidence (high vs. low), the timing of the confidence judgment (initial, courtroom), and its consistency (if the eyewitness expressed initial and courtroom confidence, did the two judgments match). Mock jurors voted guilty more when confidence was high, regardless of when the confidence judgment was made, or whether there were inconsistencies in the confidence levels. Jurors need a more nuanced appreciation of the role of eyewitness confidence, and we discuss ideas for potential interventions that may aid jurors’ decision making.

KEYWORDS:

• Eyewitness confidence
• juror decision-making
• testimonial inconsistency
• immediate confidence
• credibility

Acknowledgments

We used R (Version 3.6.1; R Core Team, 2019) and the R–packages dplyr (Version 0.8.3; Wickham et al., 2019), extrafont (Version 0.17; Winston Chang, 2014), gdata (Version 2.18.0; Warnes et al., 2017), ggplot2 (Version 3.2.0; Wickham, 2016), knitr (Version 1.23; Xie, 2015), MOTE (Version 1.0.2; Buchananet al., 2019), papaja (Version 0.1.0.9842; Aust & Barth, 2018), psych (Version 1.8.12; Revelle, 2018), and readxl (Version 1.3.1; Wickham & Bryan, 2019) to help with these analyses.

Open Scholarship

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The eyewitness’ level of confidence at the time of the identification procedure is unknown, as this information was not recorded and/or analyzed. For a more cautious take on these findings, see Berkowitz et al. (2020).

2 An additional N = 698 people began the study but were excluded by Qualtrics for failing to meet the quota, failing to complete the survey, or for spending less than 4 s on each page of the trial transcripts. Qualtrics uses ⅓ the median response time as a standard quality check. Usually this is done using total completion time, but we used this cutoff per page of the trial transcript to ensure that participants actually read the transcript. In our data, 5 s was ⅓ the median time, so to be more conservative, we requested Qualtrics remove any data with response times lower than 4 s on any page of the transcript.

3 Withdrawing did not significantly differ across experimental conditions, χ2 (df = 7) = 8.00, p = .33.

4 The authors are grateful to Dr. Amy Bradfield-Douglass for providing these transcripts.

5 We included all participants in our analyses regardless of whether they could correctly recall trial details including the eyewitness confidence level(s) because those participants who incorrectly recalled might be less affected by our manipulations. Consequently, excluding these participants might bias the results towards significance. Also, as mentioned above, excluding these participants decreases external validity.

6 None of the demographic variables emerged as significant predictors of verdict; therefore, demographics were removed from remaining analyses to increase power. 

7 We thank an anonymous reviewer for this suggestion.

8 There were also three variables with statistically significantly higher ratings in the high-match compared to the inflation condition: Confidence in Witness, Probability of Guilt, and Witness Correct.

Additional information

Funding

This work was supported by UAHuntsville Individual Investigator Distinguished Research (IFDR) Program.