ABSTRACT
In Experiment 1, simulated social pressure was manipulated through two factors: whether participants believed they were interacting with others or not via a webcam and whether they believed they were being recorded or not. Participants who believed they were being recorded, were significantly less accurate at recognising faces than those who did not believe they were being recorded. For Experiment 2, we found that the recognition of own-ethnicity faces was negatively affected by observation but not the recognition of other-ethnicity faces, and then only when observed during learning. Experiment 3 demonstrated that observation affected the recognition of upright faces more so than that of objects and inverted faces. Experiment 4 showed that observation does not affect the amount of holistic processing engaged in, but does affect how people view faces. Such results indicate that expert face recognition is susceptible to increased error if participants are being observed whilst encoding faces.
Acknowledgements
The authors would like to thank David White and an anonymous reviewer for helpful comments on a previous version of this manuscript. Further, the authors would like to thank Gizem Arabacı for help in constructing the figures.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Peter J. Hills http://orcid.org/0000-0002-0097-9170
Notes
1 Note, that because of the design, this could only be done in a by-item analysis.
2 Mauchley’s Test of sphericity was significant, therefore the Huynh-Feldt correction was applied.
3 In the partial design, the same top of a face is always paired with a different bottom leading to an incongruence in the response, whereas when the different top is presented, there is a congruence in the response. This suggests that there might a congruence effect in the partial design. Due to the statistical calculation of the composite face effect in the complete design, Rossion (Citation2013) argues that this is measuring response bias rather than holistic processing. Since we are not able to resolve this debate, we have employed a task where both measures can be calculated.
4 Differences in degrees of freedom across this Experiment represent data loss due to programming error or failure to calibrate to the eye-tracker.
5 Incidentally, using the simpler WW-WP measure of holistic processing, there was a significant result, t(100) = 4.43, p < .001, Cohen’s d = 0.89. Similar to the reported measure, this was not significantly affected by observation condition, t(99) = 0.97, p = 336, Cohen’s d = 0.19.
6 Mauchley’s test of sphericity was significant, therefore we employed the Greenhouse-Geisser correction (Field, Citation2010).