Bad news for passport control: face-matching is harder than we realised

Experiments suggest that telling if two unfamiliar faces are the same or different is no easy task. Such research has sometimes presented participants with full body shots, has more commonly used cropped shots of people's heads, but almost never placed the faces in a formal context, such as on a photographic ID card. But these are the situations in which face-to-photo matching is most relevant, when a shop assistant squints at a driver's license before selling alcohol to a twitchy youth, or an emigration official scrutinises passports before their holders pass ports. Moreover, it's plausible that the task is harder when juggling extra information, something already found in the realm of fingerprint matching, where biographical information can lead to more erroneous matches because it triggers observer prejudices. A new article in Applied Cognitive Psychology confirms these fears, suggesting that our real-world capacity to spot fakes in their natural setting is even worse than imagined.

Jennifer McCaffery and Mike Burton at the University of York asked 80 adult participants to make judgments on pairs of face photographs, deciding when the two shots were of the same person or two different individuals. In some instances, one of the two faces was embedded in a passport image (see below), and in these cases the participant was also asked to keep an eye out for data errors in the passport, which were sprinkled through a minority of instances. McCaffery and Burton found that when matching against a passport-embedded photo, participants showed a bias towards thinking the two faces were of the same person, a bias which led them to miss more instances where the photos didn't match; in a real situation, this translates to letting someone through control on another person's passport.

In another experiment, the researchers investigated how the task of face-matching affects the way we evaluate passport data and vice versa. As well as making face judgments, participants had to look out for errors in the passport, such as: the gender of the name not matching the gender of the supposed passport owner; the wrong year of birth, which could be off by 20 years (this should be an easy task, as people can typically judge age from a photograph to within five years); and an error in the place of birth, which was indicated by spelling mistakes in well-known British place names, such as "Luuton". The participants' error-spotting performance was compared against control participants recruited in earlier research who were challenged to look for errors without also having to perform face-matching.

Combining face-matching and data-matching into one situation harmed performance in both areas. Once again, faces in a passport context were more likely to be falsely waved through as a match than when the comparison was between headshots alone. And participants were poorer at spotting the data errors, compared to control participants, only catching one in five of the place of birth and year of birth errors, versus well over half caught by control participants. What's more, participants were worse at spotting data errors when the faces were a true match, suggesting the confidence from this task leaked over to the other and made them less vigilant.

It's well-known that tasks with multiple demands are harder, loading on things for our mind to deal with and allowing the possibility of interference between the two tasks. So the findings are understandable, but still troubling. Perhaps the procedure behind identity checks needs tuning by breaking it into discrete steps – perhaps even undertaken by different individuals. Moreover, this work is an important reminder that even well-established findings can be wide of the mark of how things truly operate, when based on laboratory research that doesn't take account of real-world conditions.

Further reading

—Passport Checks: Interactions Between Matching Faces and Biographical Details