A criterion-placement theory of face matching

In the paper

“Similar dissociations are seen in training effects for unfamiliar face matching, where instructions to attend to some features (e.g., the eyebrow regions) yields improved performance on match trials (Megreya & Bindemann, 2018), while increased attention to other features (e.g., facial marks such as moles and blemishes) improves performance on mismatch trials (Towler et al., 2021).”

What the sources say

• Supported

  The Experiment 1 Discussion says that eyebrow instructions enhanced accuracy and that the benefit was carried primarily by identity-match trials. The General discussion says the benefit was obtained only for identity matches. This supports the part of the sentence attributed to Megreya and Bindemann; the facial-marks finding is attributed to a different source. (LLM, conf=0.92)

  cites Megreya & Bindemann, 2018 source
• ?Not checked — no full text of the sourcecites Towler et al., 2021 source

In the paper

“For example, in a recognition memory test for names, a naturally occurring bias to classify names as 'unfamiliar' reduces the mirror effect.”

What the sources say

• Supported

  The Conventional analysis reports a strong conservative bias for unfamiliar names: participants said “yes” to them less often, and the bias made the mirror effect appear to fail on the H/FA Index. Experiment 3 describes this as a bias that concealed the mirror effect; the paper does not claim the underlying effect disappeared. (LLM, conf=0.91)

  cites Hilford et al., 2015 source
• Partly supported

  The Bias Effects section says a general bias toward “new” responses can weaken or eliminate the mirror effect. It does not identify a naturally occurring bias to classify names as unfamiliar. In the New/New Slope Correlation section, it reports that recognition experiments using name familiarity produced a mirror pattern, without linking those results to such a bias. Page numbers are unavailable in our copy. (LLM, conf=0.85)

  cites Glanzer et al., 2009 sourceExperimental — accuracy not yet measured

In the paper

“criterion is influenced by task difficulty and the relative frequency with which matches and mismatches occur (e.g., Baker & Bindemann, 2025;Stabile et al., 2024)”

What the sources say

• Partly supported

  The Experiment 3 Discussion (page unavailable) reports that unequal match–mismatch base rates produced a more liberal decision-making criterion. The passages do not establish that task difficulty influences criterion; they discuss face matching as challenging but do not link difficulty to criterion shifts. (LLM, conf=0.85)

  cites Baker & Bindemann, 2025 sourceExperimental — accuracy not yet measured
• ?Not checked — no full text of the sourcecites Stabile et al., 2024 source

In the paper

“The unrotated PCA detects that ability carries the most variance in scores, as is reflected by component 1 which has positive component loadings on both match and mismatch scores.The second component captures variations in bias.This is reflected by positive component loadings for match scores and negative loadings for mismatch scores.”

In the paper

“In the unrotated PCA, the first component, which captures the larger source of variance, corresponds to bias.This is reflected by the positive component loadings for match scores and negative component loadings for mismatch scores.”

In the paper

“The second component corresponds to ability.This is reflected by the positive component loadings for both match and mismatch scores.”

In the paper

“This simulation replicates the findings with human participants in Experiments 1 and 2, by demonstrating that match and mismatch performance can be correlated negatively when bias values vary freely, but correlated positively when bias is controlled.”

In the paper

“We provide evidence that the placement of these decisionmaking thresholds is stable at the individual level and transfers across tests, by demonstrating split-half reliability and correlations of criterion values.”

In the paper

“The first of these simulations reproduced the dissociation between match and mismatch performance when criterion is free to vary, and the positive relationship between match and mismatch accuracy when criterion is controlled.”

In the paper

“the maximum amount that any individual match or mismatch score deviated from the reported value above was less than ± 0.10.”

In the paper

“When bias varies only at the participant level (C and D), there is no association between hits and correct rejections at the participant level (r = -0.15,p = 0.13) but a strong positive correlation at the item level (r = 0.92, p < 0.001).”

In the paper

“Finally, when variability in bias is present at both the participant and the item levels (G and H), performance is not correlated either for items (r = -0.08,p = 0.71) or participants (r = 0.21, p = 0.15).”

In the paper

“When bias only varies at the item level (E and F), while there is a strong positive relationship between hits and correct rejections at the subject level (r = 0.87, p < 0.001), but the correlation is non-existent at the item level (r = -0.11,p = 0.60).”

In the paper

“Four additional participants were excluded for recognising identities used in one of the other tests (n = 1) and inattention (i.e., repeatedly pressing the”

In the paper

“Individuals with lower face-matching ability adopt more extreme criteria when there are more match than mismatch trials, or the cost of mismatch errors is high, and as such show higher levels of liberal and conservative tendencies, respectively (Stabile et al., 2024).”

What the sources say

• ?Not checked — no full text of the sourcecites Stabile et al., 2024 source

In the paper

“When ability dominates, scores are positively correlated; when bias dominates, the correlation is negative.”

In the paper

“As reported in Baker et al. (2023), an additional four participants were tested, but were excluded for failing attention checks.”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Baker et al., 2023 source

In the paper

“This indicates the presence of systematic response biases, whereby observers are biased towards match or mismatch decisions in their attempts to solve these tasks.”

In the paper

“We also suggest that item difficulty and the number of trials might be moderating factors, whereby increases in the number of items stabilizes the relative influence of difficulty values for each item, and therefore strengthens the correlation between match and mismatch performance when bias is controlled.”

In the paper

“Increasing the number of items reduces the average effect of this variation, which increases the correlation further.”

In the paper

“Five participants recognized one identity, so these trials were removed from data analysis.”

In the paper

“These values also exhibited good split-half reliability for criterion (both rs > 0.62, ps < 0.001) and adequate split half reliability for d' (both rs > 0.29, ps < 0.001).”

In the paper

“Notably, the mirror effect is also attenuated by response biases, such as a tendency to categorise items as 'new' under conditions of uncertainty.”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Hilford et al., 2015 source
• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Glanzer et al., 2009 source

In the paper

“Participants first applied Testable's screen calibration tool to ensure that stimulus size was held constant across the experiment.”

In the paper

“Again, this interpretation is strengthened by a correlation of criterion values across the two tests, which suggests that these biases are stable at the level of the individual and show transfer across different face-matching tests (Baker et al., 2023).”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Baker et al., 2023 source

In the paper

“These values were previously reported to have good test-retest reliability (i.e., criterion: both rs > 0.52, ps < 0.001; d': both rs > 0.53, ps < 0.001; Baker et al., 2023).”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Baker et al., 2023 source

In the paper

“Collectively, this suggests that such item-level biases do exist in unfamiliar face identification and might influence the dissociation found in old/new tasks.”

In the paper

“This is an important simulation as it shows that just as individual differences in observers' criterion can drive the dissociation of match and mismatch performance, it also can drive the dissociation of old/new performance.”

In the paper

“Forensic facial examiners, on the other hand, are practitioners with a trained ability for face identification and tend to be more balanced in the placement of the decision-making criterion between matches and mismatches.These experts are also more cautious in expressing confidence when errors are made (Towler et al., 2023).”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Towler et al., 2023 source

In the paper

“Whereas recognition (match trials, old trials) and discrimination (mismatch trials, new trials) are associated strongly when criterion is controlled, variability in criterion disconnects these measures.”

In the paper

“Criterion shifts have been observed under a range of conditions, for example, when face matching is performed in an airport context (Bindemann et al., 2022), in response to costs, payoffs and unequal base rates (Baker & Bindemann, 2025;Stabile et al., 2024”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Bindemann et al., 2022 source
• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Baker & Bindemann, 2025 source
• ?Not checked — no full text of the sourcecites Stabile et al., 2024 source

In the paper

“When sensitivity is poor, as is represented by partially overlapping signal and noise distributions, criterion placement should matter (e.g., a placement that yields more hits will also yield more false alarms) and this might lead to a negative relationship between match and mismatch performance.”

In the paper

“The unrotated PCA results behave exactly as they should, identifying the two underlying independent sources of variance in the data and allocating the first component to the source with greater variance.”

In the paper

“The current findings extend recent calls to integrate decision-making measures into models of face identification (Baker et al., 2023;Bindemann & Burton, 2021).”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Baker et al., 2023 source
• ?Not checked — no full text of the sourcecites Bindemann & Burton, 2021 source

In the paper

“Overall, these results extend the findings from our experiments and simulations, by demonstrating how individual differences in sensitivity and criterionor ability and biasinteract and suggest that both are important for understanding face matching.”

In the paper

“Notably, the correlations at the item level and the participant level are each dependent on the amount of variation in the bias at the corresponding level, and as such it is possible to get a correlation at the item-but not the participant-level, or vice versa.”

In the paper

“It is possible that enhanced accuracy with one of these stimulus types reflects differences in observer's decision-making thresholds.”

In the paper

“These observers exhibit a tendency to classify face pairs as matches and are biased to make highly confident errors on mismatch trials (Towler et al., 2023).”

What the sources say

• Not checked yet

  Checked before — its earlier verdict is held back until a new check runs.

  cites Towler et al., 2023 source

In the paper

“This phenomenon was observed with large sample sizes, totalling nearly 550 participants across two experiments that were conducted in two different labs, and three face-matching tests (AIFMT, GFMT, KFMT).”

In the paper

“When sensitivity is high, as is represented by a large separation of signal and noise distributions, criterion placement should not matter and there should be a positive relationship between match and mismatch performance.”

In the paper

“While accuracy and sensitivity can determine the number of errors one makes, the placement of criterion determines the type of error.”