Affects on Face Recognition: When are people confident about what they see?
Studying the relationship between a person’s ability to recognize a face accurately and their subsequent confidence level regarding the judgment is important in the realm of eyewitness testimony. Research has suggested that the apparent confidence of a witness’ face identification affects how jurors’ evaluate testimony. That is, if the witness seems very sure that the person identified is the criminal, the jury will be very convinced by their testimony; however, if the witness is unsure, the jury will also be unsure. This fact is scary considering the results of several studies that found a very weak relationship between the accuracy of face identification and subsequent confidence (Bothwell, Deffenbacher, & Brigham, 1987; Wells & Murray, 1983).
Several studies have been conducted in order to determine what makes people confident about face recognition. One theory that predicts the relationship between the accuracy of an identification and a person’s confidence is the Optimality Hypothesis proposed by Deffenbacher (1980, cited in Wells & Murray, 1984). According to the theory, the relationship between accuracy and confidence is a function of the condition in which the face was observed. When the condition is good, the relationship between accuracy and confidence should be good; when the condition is bad, the relationship between accuracy and confidence should be bad (McKelvie, 1991, 1993).
While the optimality hypothesis explains the relationship between accuracy and confidence, the Elaboration Hypothesis explains factors involved in recognition accuracy. According to the Elaboration Hypothesis, the level of processing involved in face recognition is a better predictor of accuracy than confidence. One study showed that accuracy of face recognition is improved if participants are required to increase the level of processing each face by answering questions about specific traits during the study phase (Winograd, 1981).
The following study uses a standard levels of processing paradigm to manipulate the condition in which participants engage in face processing. Specifically, during the test phase of the experiment, participants will be asked either about a certain characteristic or about the distinctiveness of a physical feature. Following the test phase, participants will see a series of faces and will be asked to determine whether each face appeared during the study phase or not. Further, after making judgments about whether they had previously seen the face, they will be asked to indicate their confidence in the judgment.
According to the Elaboration Hypothesis and the Optimality Hypothesis, participants should be most accurate and confident identifying faces for which they have required to answer a question about the distinctiveness of the face. They should be less accurate and also less confident when they have only had to answer a question about a single physical feature of the face.
Participants: A simple random sample of 50 people participated in the experiment (36 females and 14 males) as part of a class assignment. Participants ranged from 18 – 33, with the majority between the ages of 18 and 24.
Materials: The face stimuli consisted of slides of photographs depicting the head and shoulders of Caucasian females who were approximately 18 to 20 years old. None of the pictured females had easily identifiable features like glasses, jewelry or a distinctive hairstyle. In the study phase 36 slides were shown to participants. In the test phase, 72 slides were shown – 36 of which had been shown during the study phase; 36 of which were novel.
Participants were given experimental booklets which included a cover sheet, study questions, response sheets and scoring protocols. The cover sheet indicated the title of the experiment and the version. The cover sheet also contained demographic questions like age and gender.
The questions that were asked of the faces in the study phase were included in the booklet and were randomized between physical feature and distinctive feature. Examples of physical feature questions include:
Does this face have a big nose?
Does this face have close-set eyes?
Does this face have thin lips?
While examples of distinctive feature questions include:
Is the nose the most distinctive feature?
Are the eyes the most distinctive feature?
Is the mouth the most distinctive feature?
During the test phase of the experiment, participants were instructed to write a yes or no next to each number (1 through 72) in their answer book, depending on whether they had seen the face in the study phase or not. Next to where participants indicated yes/no was a space that asked for their confidence rating. Confidence ratings were between 1 (not at all confident) and 5 (totally confident) on a Likert scale.
The scoring protocols replicate the response sheet, and are coded so that each item numbered is identified with the appropriate condition (old or new face, and physical or distinctive feature condition).
Two combinations of the study and test lists were compiled (Versions A and B), such that each face occurred once in each question type across participants. This controlled for variation that may result from a particular face being more or less easily recognized.
Design: The independent variable is the type of question asked during encoding: either physical feature or distinctive feature. The dependent variable is recognition performance: the yes/no responses and confidence ratings. This is a repeated dependent measures design.
Procedure: The slides were presented at a constant rate within each phase: participants saw each face for 5 seconds during the study phase and 10 seconds during the test phase. The participants were told, going in to the experiment, that they would be participating in a memory experiment, and that they are required to answer one question for each face as it is presented. The recognition test requires participants to respond YES if they think the face appeared during the study phase and NO if they think they are seeing the face for the first time. They were also asked to indicate their confidence level for each judgment.
Complete results are shown in Table 1 below. Participants were fairly accurate across the board with higher than chance proportions of correct responses for both conditions and a low proportion of false responses (.139). The proportion of hits in the distinctive feature condition (.721) exceeded the proportion of hits in the physical feature condition (.609).
Confidence ratings were highest in when participants had correctly identified a face in the physical feature condition (4.28) and next highest when participants correctly identified a face in the distinctive feature condition (4.08). Participants were less confident on trials when they identified a face incorrectly: (3.27) in the physical feature condition and (3.42) in the distinctive feature condition.
Mean Confidence Ratings
Correct Physical Feature
Incorrect Physical Feature
Correct Distinctive Feature
Incorrect Distinctive Feature
The results above replicate some of the findings from previous research, particularly that the depth of processing affects the accuracy of recognition. The fact that confidence ratings are higher when participants correctly identify faces is also consistent with previous theory; however, we would expect the confidence rating for correct responses in the distinctive feature condition to be higher than the confidence rating for correct responses in the physical feature condition. The difference is not large and therefore may not be statistically different; however, if it is, it could be attributable to participants’ not realizing how much they retained from the distinctive feature questions. It is possible that the learning they did was implicit and therefore, they are more accurate, but fail to realize that they are more accurate in that condition.
These results support the optimality hypothesis in that the different conditions resulted in differing levels of confidence in the face recognition judgment. While the experiment has been able to demonstrate the optimal conditions for preparing people for face recognition, the experiment is not readily applicable to the courtroom problem with eyewitness testimony. That is, one cannot expect an eyewitness to engage in a depth of processing task when they first see the criminal in order to improve face recognition later on. Instead, we could use this information to instruct the jury on how to evaluate the witness’s testimony and on the questions that the lawyers might ask the witness. Rather than asking “did you get a good look at the defendant?” one might ask “did you notice anything distinctive about their face?” If the witness noticed something distinctive, then maybe their report will be more reliable.
Bothwell, R. K., Deffenbacher, K. A., & Brigham, J. C. (1987). Correlations of eyewitness accuracy and confidence optimality hypothesis revisited. Journal of Applied Psychology, 72, 691-695.
Deffenbacher, K. A., Carr, T. H., & Leu, J. R. (1981). Memory for words, pictures and faces: Retroactive interference, forgetting, and reminiscence. Journal of Experimental Psychology: Human Learning and Memory, 7, 299-305.
McKelvie, S. J. (1991). Effects of processing strategy and transformation on recognition memory for photographs of faces. Bulletin of the Psychonomic Society, 29, 98-100.
McKelvie, S. J. (1993). Effects of spectacles on recognition memory for faces: Evidence from a distractor-free test. Bulletin of the Psychonomic Society, 31, 475-477.
Wells, G. L., & Murray, D. M. (1983). What can psychology say about Neil v. Biggers criteria for judging eyewitness accuracy. Journal of Applied Psychology, 68, 347-362.
Winograd, Terry (198), “What does it mean to understand language?,” Perspectives on Cognitive Science, Ablex and Erlbaum Associates, 1981, 231-264.