Does being good with faces make you good at recognising people?
People who scored well on a standard face-matching test were mostly not better at matching identities from bodies or from movement, even though group statistics suggested they were.
Source
Face recognition ability does not predict person identification performance: using individual data in the interpretation of group results
Study at a glance
- Design
- Cross-sectional — Single-session correlational test battery: GFMT screening score used to form low/medium/high groups, then related to accuracy on a second face test, a face-hidden body-matching test and a point-light biological-motion test
- N
- N=90 · 90 undergraduates, split by GFMT score into three groups of 30; 14 scored at or above 95% and were treated as 'top performers'
- Population
- Undergraduate students at the University of Texas at Dallas, mostly female, aged 18-34
- Outcome
- Percent correct on same/different identity matching for faces (EFCT), bodies with faces masked, and point-light walking videos
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
Ninety students took a short standard face-matching test (the GFMT) plus three identity-matching tasks: a harder face test, a test with photos of people's bodies with the face covered, and a test with point-light videos showing only a person's movement. The researchers first plotted each individual's performance across tasks, then ran correlations, and finally ran the kind of group ANOVA that is common in this literature.
What they found
GFMT scores correlated moderately with the other face test (r = .54) but only weakly with body matching (r = .25) and biological-motion matching (r = .23), explaining around 5-6% of the variance for the non-face tasks. Only 20% of the low-GFMT group were low on all three tasks, and only 20% of the high group were high on all three; just 4 of 14 top performers were in the top band on every task. Yet the group ANOVA showed a significant effect of face ability with no interaction, which on its own would suggest the face test predicts all tasks.
The limits
What it doesn't show
The sample was 90 undergraduates tested once on a single day, so the study cannot say whether performance is stable over time or whether results apply to professional examiners or verified super-recognisers, who were not specifically recruited. The biological-motion test had only 20 trials and the body test was custom-made, so their low reliability could partly explain weak correlations. As a correlational study it cannot separate true ability from general motivation, which the authors note can inflate correlations across tasks.
Key terms
- Glasgow Face Matching Test (GFMT)
- A standardised test in which people decide whether two face photos show the same person; widely used to screen face-matching ability.
- Biological motion
- Movement of a living body, here shown as point-light displays where only dots on the joints are visible.
- Variance explained (r squared)
- The square of a correlation, indicating what proportion of variability in one measure is shared with another.
- Correspondence analysis
- An exploratory multivariate method that maps patterns among categorical data, used here to plot each person's performance profile across tasks.
- Super-recogniser
- A person with exceptionally good face recognition, often defined by very high scores on face tests.
Flashcards
0 of 11 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
Which task did GFMT scores predict best?
Common questions
If the ANOVA was significant, why do the authors say the face test is not a good predictor?
A significant group effect only means the relationship beats chance; individual data showed so much overlap that using the test to pick good body or motion matchers would produce many errors.
Why does it matter for real-world jobs?
Police and security agencies screen staff using face tests, but real identification often relies on bodies and gait, and this study suggests face skill does not transfer well to those cues.
Were body and motion matching related to each other?
No; accuracy on the body task and the biological-motion task was essentially uncorrelated (r = .01).
More on Face processing