“Eyewitnesses who are extremely confident are probably accurate.”

• Meta-analytic and laboratory evidence shows that when an identification is made under 'pristine' conditions — a fair (non-suggestive) lineup, a double-blind administrator, proper instructions warning the culprit may not be present, and confidence recorded immediately at the FIRST test — high confidence is associated with very high accuracy, often 95%+ suspect-ID accuracy in the top confidence bin (Wixted & Wells, 2017).
• Wixted, Mickes, Clark, Gronlund & Roediger (2015, American Psychologist) reanalyzed data including field studies (e.g., Houston Police Department lineups) and found initial eyewitness confidence was a strong, reliable predictor of accuracy, contradicting the older blanket claim that confidence is uninformative.
• Calibration analyses (rather than point-biserial correlations, which are attenuated by restricted range) generally show a monotonic confidence–accuracy relationship: witnesses expressing near-certainty are meaningfully more likely to be correct than low-confidence witnesses.
• Because of this research, the U.S. National Academy of Sciences (2014) and many courts (e.g., the New Jersey Supreme Court in State v. Henderson) recommend recording a witness's confidence statement in the witness's own words at the time of the first identification — an implicit acknowledgment that immediate confidence carries diagnostic value.
• Signal-detection framing implies that high-confidence IDs arise from strong memory matches, so at the moment of first test, extreme confidence is a genuine (if imperfect) signal of memory strength.

• The claim omits the conditions that make it true. Under suggestive lineups, non-blind administrators, poor viewing conditions, cross-race identification, long delays, or repeated testing, the confidence–accuracy relationship weakens sharply or collapses.
• Confidence is highly malleable. Wells & Bradfield (1998) demonstrated the post-identification feedback effect: telling witnesses 'Good, you identified the suspect' dramatically inflates their retrospective certainty — including among witnesses who picked the wrong person — and also distorts their reported view quality and attention.
• Courtroom confidence is not initial confidence. By trial, witnesses have typically been exposed to feedback, repeated viewings, mugshots, and preparation, so the extreme confidence jurors actually see is often contaminated and far less diagnostic than confidence at the first test.
• Repeated identification procedures corrupt the evidence: once a witness has seen a face in a lineup, subsequent identifications reflect memory of the lineup rather than of the crime (Wixted, Wells, Loftus & Garrett, 2021).
• Real-world error data: the Innocence Project reports that eyewitness misidentification was a contributing factor in roughly 60–75% of DNA-based exonerations in the U.S., and many of those witnesses testified with total certainty (e.g., the Jennifer Thompson / Ronald Cotton case).
• Even in favorable conditions, high confidence is not near-certainty of accuracy: some studies show high-confidence error rates well above zero, and a confident identification of an innocent 'filler-like' suspect in a biased lineup can be systematically wrong.
• Jurors overweight confidence. Research on juror decision-making finds confidence is the single strongest driver of perceived credibility, which makes an unqualified version of this claim practically dangerous in legal settings.
• The word 'probably' is doing a lot of hidden work: base rates matter. If a lineup contains an innocent suspect, no level of witness confidence can make the identification correct.

84/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Wixted & Wells, Psychological Science in the Public Interest (Association for Psychological Science)journal · 95 · National Research Council / National Academies of Sciences, Engineering, and Medicinegov · 95 · Wixted, Wells, Loftus & Garrett, Psychological Science in the Public Interestjournal · 93 · Wells & Bradfield, Journal of Applied Psychology (APA)journal · 92 · Wixted, Mickes, Clark, Gronlund & Roediger, American Psychologist (APA)journal · 92 · National Institute of Justice, U.S. Department of Justice — 'Eyewitness Evidence: A Guide for Law Enforcement'gov · 88 · American Psychological Associationedu · 85

• Scientific reviews find a strong confidence-accuracy relationship when confidence is recorded immediately after the witness's first identification and the procedure is fair, double-blind, and non-suggestive; under those 'pristine' conditions, very high-confidence identifications are often highly accurate (Wixted & Wells, Psychological Science in the Public Interest).
• The National Academies concluded that confidence can have evidentiary value if it is captured at the time of the initial identification, before police feedback or other contamination; this supports a limited version of the claim.
• NIJ guidance for law enforcement recommends obtaining and documenting a witness's confidence statement in the witness's own words at the time of identification, reflecting that contemporaneous confidence is considered relevant evidence.

• The claim is misleading if applied to courtroom confidence or later confidence: confidence is malleable and can be inflated by confirming feedback, repeated questioning, co-witness discussion, media exposure, or suggestive lineup procedures.
• Wells & Bradfield's Journal of Applied Psychology study found that simple confirming feedback such as 'Good, you identified the suspect' distorted witnesses' later reports, including making them more confident, even when their identifications were wrong.
• The National Academies emphasized that eyewitness confidence is not a fixed measure of memory strength and that jurors may overvalue confident testimony, especially when confidence has been contaminated after the initial identification.
• High-stress witnessing conditions can reduce eyewitness accuracy, according to a Law and Human Behavior meta-analysis; high subjective confidence does not necessarily correct for poor encoding conditions.
• Wrongful-conviction data show that eyewitness misidentification has contributed to many DNA exonerations, demonstrating that eyewitness certainty can coexist with serious error, particularly when procedures are flawed.

88/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Psychological Science in the Public Interestjournal · 96 · National Academies Pressother · 95 · National Institute of Justicegov · 92 · Journal of Applied Psychologyjournal · 94

• Recent psychological research has demonstrated a 'new synthesis' regarding eyewitness memory: when initial identification is made under pristine, double-blind lineup conditions, an eyewitness's confidence is highly predictive of their accuracy.
• If an eyewitness expresses extreme confidence at the very first, uncontaminated identification attempt, statistical models show they are highly likely to be accurate.

• Eyewitness memory is highly malleable. Confidence can be artificially inflated by confirming feedback (even unintentional) from police, lawyers, or co-witnesses.
• Confidence expressed later in the legal process, such as on a witness stand in a courtroom, has a weak correlation with accuracy because the witness's memory has often been altered or reinforced over time.
• Data from DNA exonerations reveals that eyewitness misidentification is a leading cause of wrongful convictions, and many of these cases involved witnesses who testified with absolute, yet incorrect, confidence.

95/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Psychological Science in the Public Interestjournal · 95 · National Academies of Sciences, Engineering, and Medicineedu · 95

• Research distinguishes confidence expressed immediately during the initial identification from confidence expressed later in court. Studies published after the National Academies report found that confidence recorded at the initial identification can be strongly correlated with accuracy under carefully controlled procedures.[https://www.pnas.org/doi/10.1073/pnas.1706891114]
• The National Research Council reported that the confidence–accuracy relationship is likely strongest at the time of the initial identification, supporting limited circumstances in which high confidence is informative.[https://www.nationalacademies.org/news/report-urges-caution-in-handling-and-relying-upon-eyewitness-identifications-in-criminal-cases-recommends-best-practices-for-law-enforcement-and-courts]

• The statement is too broad because courtroom confidence is not a reliable predictor of eyewitness accuracy; confidence can be affected by feedback, repeated questioning, and the passage of time.[https://www.nationalacademies.org/news/report-urges-caution-in-handling-and-relying-upon-eyewitness-identifications-in-criminal-cases-recommends-best-practices-for-law-enforcement-and-courts]
• A peer-reviewed PNAS review states that courtroom statements of confidence are very poor predictors of accuracy, while emphasizing that confidence at the initial identification is more informative.[https://www.pnas.org/doi/10.1073/pnas.1706891114]
• Real-world DNA-exoneration data show that eyewitness misidentification is a major contributor to wrongful convictions: the Innocence Project reports that 69% of U.S. DNA exonerations in its cited dataset involved eyewitness misidentification.[https://innocenceproject.org/news/how-eyewitness-misidentification-can-send-innocent-people-to-prison/]
• The National Academies recommends caution in relying on eyewitness identification and notes that confidence at trial is not a reliable accuracy indicator, contradicting the unqualified claim that extremely confident eyewitnesses are probably accurate.[https://www.nationalacademies.org/read/18891/chapter/1]

96/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Proceedings of the National Academy of Sciencesjournal · 97 · National Academies of Sciences, Engineering, and Medicinegov · 98 · National Academies of Sciences, Engineering, and Medicineedu · 98 · Innocence Projectother · 88

• Under fair/pristine lineup procedures with confidence recorded immediately after the ID, very high-confidence identifications are often highly accurate in lab and some field data (commonly cited rates well above chance, sometimes ~90%+).
• There is a positive confidence-accuracy calibration: people who express extremely high confidence (e.g., 90–100%) are, on average, more likely to be correct than those who express moderate or low confidence.
• Some later syntheses argue the older ‘confidence is useless’ view overstated the problem specifically for initial high-confidence IDs made under good conditions.

• Many meta-analyses find only a modest overall confidence–accuracy correlation for eyewitness identification, so a large share of extremely confident witnesses are still wrong.
• Confidence is highly malleable: confirmatory feedback, co-witness talk, repeated questioning, and delay inflate confidence without raising accuracy.
• DNA exonerations show mistaken eyewitness identifications—often delivered with high confidence—are the single most common factor in known wrongful convictions.
• The relationship weakens or breaks for event details (not just person ID), poor viewing conditions, long delays, biased lineups, and confidence assessed later rather than immediately.
• Legal decision-makers systematically overweight witness confidence, which is a documented source of error.

82/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Psychological Science in the Public Interest / Association for Psychological Sciencejournal · 93 · National Academies of Sciences, Engineering, and Medicineother · 96 · Innocence Projectother · 78 · National Institute of Justice (U.S. Department of Justice)gov · 90