Face masks are less effective than sunglasses in masking face identity

Abstract

The effect of covering faces on face identification is recently garnering interest amid the COVID-19 pandemic. Here, we investigated how face identification performance was affected by two types of face disguise: sunglasses and face masks. Observers studied a series of faces; then judged whether a series of test faces, comprising studied and novel faces, had been studied before or not. Face stimuli were presented either without coverings (full faces), wearing sunglasses covering the upper region (eyes, eyebrows), or wearing surgical masks covering the lower region (nose, mouth, chin). We found that sunglasses led to larger reductions in sensitivity (d') to face identity than face masks did, while both disguises increased the tendency to report faces as studied before, a bias that was absent for full faces. In addition, faces disguised during either study or test only (i.e. study disguised faces, test with full faces; and vice versa) led to further reductions in sensitivity from both studying and testing with disguised faces, suggesting that congruence between study and test is crucial for memory retrieval. These findings implied that the upper region of the face, including the eye-region features, is more diagnostic for holistic face-identity processing than the lower face region.

Figure 1

Example faces of the same identity in the 3 disguise types. All faces were displayed against a black screen, with clearly distinguishable head outlines.

Figure 2

Experiment 1 results: (a) Hit rates and false alarm rates, and (b) sensitivities d’ and criteria c (indicating response biases) across the three conditions. All error bars represent ± 1 SEM. The asterisks (*) indicate significant pairwise differences between conditions. The hash signs (#) in (b) represent d’ or c significantly different from zero in the indicated conditions. The unmarked bars and comparisons were not statistically significant. The significance level was set at 0.05.

Figure 3

Experiment 2 results: (a) Hit rates and false alarm rates, and (b) sensitivities d’ and criteria c (indicating response biases) across the three conditions. All error bars represent ± 1 SEM. The asterisks (*) indicate significant pairwise differences between conditions. The hash signs (#) in (b) represent d’ or c significantly different from zero in the indicated conditions. The unmarked bars and comparisons were not statistically significant. The significance level was set at 0.05.

Figure 4

Experiment 3 results: (a) Hit rates and false alarm rates, and (b) sensitivities d’ and criteria c (indicating response biases) across the three conditions. All error bars represent ± 1 SEM. The asterisks (*) indicate significant pairwise differences between conditions. The hash signs (#) in (b) represent d’ or c significantly different from zero in the indicated conditions. The unmarked bars and comparisons were not statistically significant. The significance level was set at 0.05.

Figure 5

Sensitivities d’ and criteria c (indicating response biases) across the three experiments. Sensitivities: Incongruent disguises (Experiments 2 & 3) led to significantly lower d’ than congruent disguises (Experiment 1), and sunglasses led to significantly lower d’ than masks, as indicated by the asterisks (*). Broken line (in orange): d’ for presenting only full faces, averaged across all three experiments. Criteria: No significant effects were found between sunglasses and masks, and among congruent and incongruent conditions. Presenting only full faces resulted in no significant bias (mean c = – 0.01, p = 0.81; not drawn). All error bars represent ± 1 SEM. The hash signs (#) represent significant differences from zero in the indicated conditions. The unmarked bars and comparisons were not statistically significant. The significance level was set at 0.05.