Post No.: 0347
Identifying upside-down faces is particularly difficult compared to trying to identify other objects that are presented to us upside-down, suggesting that we use a different brain mechanism when identifying faces. This may possibly be because we need to identify different individual (right-way-up) faces all of the time as social animals hence our brains have developed time-and-effort-saving shortcuts when identifying faces. These work fine most of the time but when these shortcuts fail because the context these shortcuts mostly evolved for or during has changed (in this case by unusually presenting faces upside-down), they can fail badly.
It is theorised that we use two mechanisms for identifying objects – one is by dissecting the object into its component parts or features (the windows, doors, bricks and tiles if it’s a house), and the other is through a holistic view of the object (seeing the entire building’s form). But when identifying faces, we recognise them almost entirely holistically and only a tiny part by dissection. Unless they’re highly distinctive features, we find it difficult to identify a person from only seeing their nose and mouth, for instance. This might be why traditional photofit or identikit techniques tend to be very unreliable. (A ‘photofit’ is an image of a face composited from parts of photographs of other people’s faces. An ‘identikit’ is closely similar but is an image of a face constructed from strips showing facial features which have been selected to match a witness’s description.)
Facial features from one person to the next tend to be only incredibly subtly different (we’re talking about differences of range from the biggest to smallest or tallest to widest feature, for most people, of maybe only a few percent). Some outliers may possess particularly odd or distinctive features but most people don’t. And we may be able to narrow an identity down to, perhaps, ‘a white, middle-aged female with auburn hair’, but many people within a population would likely fit such a description. The way we tend to struggle to describe or draw from memory a face – even one we’re supposedly familiar with – suggests that we mainly only recognise faces holistically i.e. we don’t look at the eyes, nose then mouth to identify someone. We look at an entire face all at once. In fact, even when the target face is in view, we tend to get poor likenesses when using traditional photofits(!) Most people find it difficult to draw accurate faces from life models who are right in front of them.
‘EvoFITs’ or ‘EFIT-Vs’, therefore, ask eyewitnesses to judge whole faces for likenesses rather than individual features. They deliberately blur out the ‘external’ features (the hair, ears and neck) so that people will focus on the ‘internal’ features (such as the eyes, eyebrows, nose and mouth) since hair is easily changeable and ears and necks can be obscured thus these features can be distracting to what’s important. They also allow holistic adjustments on a sliding analogue/finely-discrete scale (such as the proportions, age, health, ‘personality’ and overall appearance of the face).
Yet even this technique only works to allow others to accurately identify the resultant face up to 25% of the time! Though this is much better than the 5% with traditional photofits.
Verbally trying to describe a person’s face to a sketch artist also creates poor results, even though these sketches are done holistically because of the limitations of verbal descriptions when concerning very subtle details. People also tend to struggle to differentiate between different individual members of an ethnicity that they’re not personally familiar enough with because they don’t see them on a frequent or regular basis. Not all black people look the same, not all oriental people look the same, not all white people look the same, and so forth. Indeed, not all dogs of the same breed look the same, not all cats of the same breed look the same, not all chimpanzees look the same, and so forth. When people perceive that members of their own ethnicity look diverse but members of other ethnicities look more alike, this speaks about their own unrefined internal models (the cross-race effect). People who live with, say, two Rottweilers of the same age, will be able to differentiate between their faces, whereas those who don’t might struggle if they look very similar. Woof!
The administrator of a photographic or video line-up or identity parade (live line-ups are rare nowadays) may already know who the suspect is thus may influence and goad a witness with instructions, cues and biasing behaviours that, either intentionally or unintentionally, try to push the witness into making a particular choice – hence the need for double-blinding. This means that neither the administrator nor the witness will know beforehand which picture is of the real suspect.
But biasing behaviours from the administrator and witness may still be present even if the administrator doesn’t know who the suspect is, due to assumptions and prejudices. For example, witnesses may assume that the suspect ‘must be’ in the line-up and so will feel obliged to choose who most closely resembles the criminal in their memories, which could be someone completely innocent. This response bias can be easily solved by stating that the suspect may or may not be in the line-up – but this may reduce the number of true positives as well as false positives; albeit it’s on balance considered more important for furry innocents to go free than for fuzzy criminals to get caught.
Presenting photographs simultaneously may encourage relative rather than absolute judgements too. This is a problem because we’re looking for the one true suspect, not ‘who most looks like the suspect’; and we’re trying to compare the face in the photograph with the face in the eyewitness’s memory, not one face with another in front of them. Therefore presenting the photographs sequentially (without the eyewitness knowing in advance how many photographs will be in the line-up) and stopping on the first positive identification helps to encourage more absolute judgements. Another technique is to first present a line-up that does not include the suspect at all – if the eyewitness picks one of these faces then we may call into question their judgement with the real line-up.
An eyewitness may have mostly seen a face from the side rather than straight on, and while moving rather than static. Thus there are systems being developed that allow eyewitnesses to rotate images of faces to whichever orientations are more informative for them.
Choosing the right ‘foils’ or ‘fillers’ (the innocent members in a line-up who aren’t the suspect, who are there to act as distractors) is important. Foils must wear the same clothes as and must bear some resemblance to the suspect, but an arguably better practice is to make sure the foils match the verbal description of the criminal that the witness has already provided. This is because people who look similar may still have certain characteristics that the eyewitness has already firmly discounted in their verbal description (e.g. a blue eye colour), thus a portion of the line-up would be effectively useless and be automatically ruled out, which reduces the number of effective foils in the line-up and making it more likely that the suspect is picked via deduction rather than memory. By only using foils that all fit the verbal description of the criminal, we can choose foils whom may vary considerably on the characteristics that were not a part of the provided description too. One can also be too successful in finding similar-looking foils – the eyewitness’s memory of the suspect’s face may be quite good yet he/she still might not be able to tell the people apart in such a line-up.
Like with any attribute that falls under a normal distribution when looking at the population, there are some people who are far above average at recognising faces after only a brief look at them. They’ve been coined ‘super recognisers’. But even they’re not perfectly accurate all of the time.
A very high confidence in one’s memory may very weakly predict the accuracy of that memory – but level of confidence is still best disregarded as an indicator of eyewitness accuracy. The longer the duration an eyewitness had viewed the perpetrator then the logically more that his/her confidence should correlate with his/her accuracy though. The correlation between confidence and accuracy also depends on what the outcome was – when eyewitnesses don’t identify the suspect as being the criminal then the correlation is very low, but if they do identify the suspect as being the criminal then the correlation is medium-low. A child’s level of confidence is relatively even more unreliable compared to an adult’s, on average. If an innocent person looks incredibly similar to the real perpetrator then the correlation obviously decreases too.
Another major factor is how confidence can be manipulated with incredible ease, such as by casting or removing doubt by giving discouraging or encouraging feedback to the witness, especially in a trial setting, even inadvertently (e.g. directly by saying, “That’s not what the other witnesses said” or indirectly by saying, “You must’ve had a pretty clear view of the criminal from where you were.”) Feedback can even influence whether an eyewitness is willing to testify at all.
Some suggest that, instead of witnesses indicating whether a photograph is that of the suspect or not via a ‘yes or no’ response, they should indicate their level of confidence on a ‘0-100%’ scale. And then compare the suspect’s rating with that of the foils.
Fast identifications are also more likely to be accurate, and vice-versa – hence giving no more than 3 seconds to look at each picture (presented sequentially) in a line-up, and asking for a confidence rating, may help prevent conscious biases from influencing the judgements (e.g. rationalising why a particular picture ‘must be’ of that of the suspect rather than basing one’s decision purely from memory).
With all of these techniques though, there are limitations, and confidence still shouldn’t be used as a measure of whether to believe in somebody’s eyewitness testimony in real-life forensic situations. No matter these small improvements to the procedure of identifying a suspect – confidence is still overall a poor predictor of eyewitness accuracy. These techniques may help during the police investigation to narrow down suspects, but not when it reaches the courtroom.
It’s also worth adding that witnesses need to be protected against ‘stitches for snitches’. Sometimes witnesses are intimidated by violent persons or gangs they’re giving statements against. It’s not right for innocent people, who are trying to do the right thing, to be put in the crosshairs of violent gangs for trying to testify against their members. Such intimidation can result in the witnesses changing or retracting their statements, or getting harmed one day – hence the importance of adequate witness protection systems.
Woof. This post essentially continues on from Post No.: 0339, which took a look at police interviews and interrogations. I really enjoy the study of criminal justice, particularly the psychological aspects of the field, and I hope you do too!