When criterion control in face matching induces correlation: Commentary on Baker et al. (2026)

Area of Science:

• Cognitive psychology
• Psychometrics
• Human-computer interaction

Background:

• Face matching accuracy shows weak correlations between match and mismatch trials.
• Previous research suggested individual differences in response criterion explained this weak correlation.
• Baker et al. (2026) proposed controlling for criterion revealed a shared ability.

Purpose of the Study:

• To examine a methodological issue in residualization procedures used in signal detection analyses.
• To determine if statistical artifacts can induce correlations between match and mismatch accuracies.
• To clarify the interpretation of residual correlations in face matching research.

Main Methods:

• Monte Carlo simulations were used to model performance under independent and dependent conditions.
• Criterion was computed using hit and false-alarm rates from the same datasets.
• Reanalysis of original datasets was performed to validate simulation findings.

Main Results:

• The residualization procedure can induce large positive correlations even with independent performance.
• The magnitude of the induced correlation depends on the coupling between outcomes and covariates.
• Correlations were maximal when criterion was derived from the same task and reduced when estimated independently.

Conclusions:

• Residual correlations in signal detection analyses are not definitive proof of a shared underlying ability.
• Statistical conditioning can introduce artifacts that inflate correlations, requiring cautious interpretation.
• Distinguishing genuine shared variance from statistical artifacts is crucial for accurate signal detection analysis.