Weber, Nathan &
Brewer, Neil. (2003). The Effect of Judgment Type and Confidence
Scale on Confidence-Accuracy Calibration in Face Recognition. Journal of Applied Psychology, 88,
490-499.
Introduction
The issue concerning whether or not witness confidence is
indicative of their decision accuracy has been a debatable one. Studies that used calibration as a way of
measuring the confidence-accuracy (CA) have shown advantages over the standard
CA correlation. Calibration is the
amount in which the participant’s confidence level matches the probability that
they made the correct decision.
Although there is supporting evidence for CA calibration in
identification, little is known about the factors affecting it. Therefore Weber and Brewer take a look at
the confidence scales that witnesses are asked to use, the type of judgment
strategy used on CA calibration, and the decision latency accuracy relationship.
Method
Each participant was tested in four blocks of trials within one
session. Participants first engaged in
practice trials to get comfortable with the instructions for that particular
condition. After that participants were
shown 25 photos of faces at 500 ms each and were asked to remember the faces as
best they could. A delay of 10 minutes
followed study before participants were given the recognition test. In the single-face trials, participants were
shown one face at a time and were asked to judge whether the face was shown at
study. In the pair-face trials,
participants were shown two faces simultaneously and were asked to judge which
face had been shown at study. In each
condition, 25 target and 25 distractor faces were presented (50 trials for the
single-face and 25 for the pair-face).
After each recognition judgment, participants rated how confident they
were in their decision. In the
full-scale condition participants rated their confidence on a scale from
0%-100%, and in the half-scale condition participants used a scale from
50%-100%.
Results
The test revealed no differences in accuracy in the either the
single-face condition (between full-scale and half-scale) or the pair-face
condition. Also, half-scale actual
confidence scores were significantly greater than full-scale scores. Yet the confidence in the rescaled condition
appeared much larger in the full-scale conditions, meaning that responses on
the full-scale were closer to the high end vs. when the half-scale was
used. Not only that but pair-face
judgments were made with more accuracy and confidence than single-face
judgments. These results indicate that
relative judgments were more accurate than absolute judgments. It also took longer for participants to make
relative judgments than absolute judgments, causing a negative relationship
between accuracy and decision latency.
Figure 1 shows CA calibration was better for half-scale
conditions rather than full-scale conditions.
It also suggests that pair condition responses were more underconfident
while the single condition responses exhibited a closer relationship between
accuracy and confidence. This means
that absolute judgments are favored according when CA calibrations is produced. Another CA calibration compared the positive
recognition decisions (old) with the negative recognition decisions (new). Positive decisions produced well-calibrated
confidence judgments, while negative decisions demonstrated the opposite
effect. Overall, the half-scale,
single-face, positive recognition decisions seem to have a stronger CA
calibration.
Discussion
The results from this study have increased our understanding of
CA calibration in face recognition and will hopefully be an aid in eyewitness
identification. The favored absolute
strategy vs. the relative strategy opens doors for continued debate between
sequential and simultaneous lineups.
Several of the findings in the face recognition experiment can be
applied to eyewitness identification; however, future research should attempt
to bridge some of the gaps that still exist between the laboratory and the
actual courtroom, such as the lineup presentation.