Using confidence judgments provided during versus after memory retrieval to predict eyewitness memory accuracy - Emily Spearing, Kimberley Wade
1. Using confidence judgements provided
during versus after memory retrieval to
predict eyewitness memory accuracy
Emily Spearing & Kimberley Wade
2. What we know
There is a growing-body of research on the confidence-accuracy relationship
(Wells et al., 1998; Brewer, 2006)
Recent research suggests that confidence judgements could be useful for
police investigations (Wixted, Mickes & Fisher, 2018)
However, little is known about how confidence judgements should be
collected during police interviews (see: Brewer et al., 2018; Saraiva et al.,
2020; Sauer & Hope, 2016)
3. What we know (cont.)
Studies have examined confidence given AFTER each response (immediate
confidence) or confidence given at the END of the memory test (delayed
confidence)
Only one study has compared immediate and delayed confidence
judgements (Robinson & Johnson, 1996)
Mock Crime
Memory Test
4AFC or cued recall
questions
Confidence
1 – 9 scale
Immediate
Delayed
Filler
4. What we know (cont.)
Studies have examined confidence given AFTER each response (immediate
confidence) or confidence given at the END of the memory test (delayed
confidence)
Only one study has compared immediate and delayed confidence
judgements (Robinson & Johnson, 1996)
The confidence-accuracy correlation was similar in the immediate and
delayed confidence conditions
5. What we know (cont.)
There are several reasons to investigate the timing of confidence
judgements further:
Previous research calculated the point-biserial correlation which can be
misleading about the strength of the confidence-accuracy relationship
Identification research shows that immediate confidence judgements
show less underconfidence than delayed (5 min) confidence judgements
(Brewer, Weber & Semmler, 2007)
Different results may be obtained for different events and under different
encoding conditions (Lindsay et al., 1998)
6. Experiment 1:
How does the timing of confidence
judgements influence the CA relationship for
eyewitness memory?
7. Experiment 1
The source monitoring account
suggests that metacognitive cues
decay over time (Johnson,
Hashtroudi & Lindsay, 1993).
Thus, delayed-confidence subjects
should be more likely to rely on
heuristics, such as perceived
familiarity, to make confidence
judgements, than immediate-
confidence subjects.
25
40
55
70
85
100
25 - 40 45 - 60 65 - 80 85 - 100
%Correct
Confidence Level
Delayed
Immediate
8. Filler (3 mins)
Event
18 MCQs
“What was the
colour of the
stolen car?”
Black
Blue
Silver
White
Memory Test
Immediate
Delayed
“How confident
are you in your
response?”
25 – 100 scale
Confidence
Experiment 1: Method
N = 341
13. Experiment 2:
How does misleading questioning and the
timing of confidence judgements influence
the CA relationship for eyewitness
memory?
14. Experiment 2: Hypotheses
1. If metacognitive cues decay over time, then delayed confidence
judgements will show a weaker CA relationship than immediate
confidence judgements.
2. If people are misled and fail to accurately source monitor, then misled
items will show more overconfidence than non-misled and control
items.
15. Experiment 2: Items
1. Misled: We misled subjects about these items during the
misleading question phase.
2. Non-misled: We asked subjects about these items during the
misleading question phase but did NOT mislead them.
3. Control: We did NOT ask subjects about these items during the
misleading question phase.
All 3 item types were included in the final memory test.
16. Experiment 2: Method
Event Misleading Qs
8 MCQs (4
Misleading, 4
Non-Misleading)
“Was the stolen
car parked in
Parking Zone 4
or Parking Zone
D?”
Correct answer:
Parking Zone 3
Confidence
“How confident
are you in your
response?”
0 – 100 scale
12 Cued Recall
Qs (4 Misled, 4
Non-Misled, 4
Control)
“Which parking
zone was the
stolen car
parked in?”
Memory Test
2.5 min
DC
IC = Immediate Confidence, DC = Delayed Confidence N =
IC
3 min
23. Experiment 3:
How does the timing of confidence
judgements influence the CA relationship
and completeness of eyewitness reports?
24. Experiment 3: Hypotheses
1. If metacognitive cues decay over time, then delayed confidence
judgements will show a weaker CA relationship than immediate
confidence judgements.
2. If immediate confidence judgements disrupt the recall process, then
immediate confidence subjects will produce less complete reports than
delayed confidence subjects.
25. Filler (3 mins)
Event
1. Free Recall
2. 10 Cued
Recall Qs
“Which parking
zone was the
stolen car
parked in?”
Memory Test
Immediate
Delayed
“How confident
are you in your
response?”
0 – 100 scale
Confidence
Experiment 3: Method
N = 212
Repeat for Alternate Event
26. Summary
The timing of confidence judgements did not influence the CA relationship
when the delay was short.
Subjects adjusted their confidence to compensate for lower accuracy when
they saw an event under poor visibility.
Repeated and misleading questioning were detrimental to the CA
relationship.
27. Questions
1. Under what conditions might you expect a difference between
immediate and delayed confidence judgements?
2. What would be compelling evidence that a delay does not affect the CA
relationship?
3. Given the challenges of studying the CA relationship for recall (few low
confidence responses etc.), how else could we investigate the CA
relationship for recall?
>>> Visibility conditions for range of confidence ratings
>>> Between-subjects
>>> Varying difficulty of questions
We know that misinformation can reduce the accuracy of eyewitness reports
Few studies have investigated how misinformation influences the CA relationship
YOU MIGHT HAVE TO EXPLAIN AT THIS POINT WHAT NON-MISLED AND CONTROL ITEMS ARE (OR MAKE IT CLEAR ON THE NEXT PAGE)…. PEOPLE OFTEN USE THESE TERMS INTERCHANGEABLY SO SOME AUDIENCE MEMBERS WON’T UNDERSTAND WHY THEY ARE DIFFERNET TYPES OF QUESTIONS HERE.
ALSO EXPLAIN THE MECHANISM LEADING TO OVERCONFIDENCE
THESE DATA ARE BEAUTIFUL. IT’S WORTH SPENDING A BIT OF TIME ON THIS SO THE AUDIENCE CAN ABSORB JUST HOW BAD THE AC IS WHEN QUESTIONS GO AWRY
OUR ULTIMATE AIM WAS TO EXPLORE THE CA IN A FREE RECALL TASK BECAUSE IT’S MORE ECOLOGICALLT VALID BUT THERE ARE CHALLENGES IN RECALL DATA WHICH I’LL RETURN TO AT THE END
Real eyewitnesses are often subjected to lengthy interviews
Free recall responses are usually associated with high accuracy and high confidence
We used cued recall questions to produce sufficient variation to plot calibration cruves
>>> Visibility conditions for range of confidence ratings
>>> Between-subjects
>>> Varying difficulty of questions
CODING ONGOING
1. We probably found no effect because the delay was quite short
2. Subjects appropriately adjusted their confidence ratings under the low-accuracy/poor visibility conditions fits with some lineup research showing that people can calibrate quite well in certain situations (e.g., older adults show good calibration when making lineup decisions. See Melissa’s line-up paper on unfair lineups). So it looks like people can use their metacognitive beliefs (what they know about memory) to adjust their confidence ratings appropriately in some situations. But can’t adjust their confidence when they are unknowingly confronted by misleading questions.