Research Proposal Assignment | Essay Help Services

Please see attached documents for prompt and sources. Use only the sources attached for this assignment, and use the below research question.

Research Question- Can educating people about eyewitness identifications reduce the tendency for a jury to wrongfully convict on the basis of an eyewitness ID

Eyewitness Testimony in Occupational Accident
Investigations: Towards a Research Agenda
E. Kevin Kelloway,1,2 Veronica Stinson,1 and Carla MacLean1
Accident investigation is frequently cited as the cornerstone of an effective occupational
health and safety program. We suggest that the literature on accident investigation is
based on a model of witnesses as neutral and accurate recording devices. The literature
on eyewitness testimony and criminal investigation offers strikingly different conclusions.
We review these findings and point to their implication for research on accident
investigation in occupational health and safety contexts.
KEYWORDS: eyewitness testimony; accident investigation.
Two employees were moving stock in a furniture warehouse. Both were experienced
employees who have rearranged stock hundreds of times. Large appliances were
stacked four high on pallets. One of the clothes dryers on the top tier began to fall.
Attempting to catch the appliance before it smashed to the floor, one employee suffered
severe muscle damage to his back and shoulder. As a result of the injury, the
employee is unable to return to his regular employment. The state-sponsored Workers’
Compensation fund has awarded him a long-term pension and is now paying
for his retraining into a more suitable occupation. The employer is paying increased
Workers’ Compensation premiums for the foreseeable future. The employee is paying
with the loss of his career and his health. All parties have lost as a result of the
incident.
This brief description illustrates many of the central features of occupational
accidents. For purposes of exposition, we highlight three of these features. First,
although the outcome of the incident is clear in terms of both physical and economic
consequences, exactly what happened is not clear. Were the appliances improperly
stacked in the first place? Did the employees bump the tiers dislodging the dryer?
Were other parties involved? Second, in a related vein, it is difficult to determine
why the accident happened.Were the workers careless?Were they untrained?What
differentiated this day from the hundreds of similar events in the past that did not
1Department of Management, Saint Mary’s University, Halifax, Nova Scotia, Canada.
2To whom correspondence should be addressed at Department of Management, Saint Mary’s University,
Halifax, Nova Scotia, Canada B3H 3C3; e-mail: kevin.kelloway@stmarys.ca.
115
0147-7307/04/0200-0115/1 C ° 2004 American Psychology-Law Society/Division 41 of the American Psychology Association
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
116 Kelloway, Stinson, and MacLean
result in an accident? Finally, most importantly for this review, in most cases, attempts
to answer these questions are going to rely on interviews of eyewitnesses.
Given this emphasis on eyewitness testimony as a means to investigate and understand
occupational accidents, it is perhaps surprising to note that our literature
searches revealed two striking gaps. First, the literature on accident investigations is
largely practitioner-based and does not reference the empirical research literature
on eyewitness testimony. Rather, the overwhelming impression is that accident investigators
treat eyewitnesses as neutral and accurate recording devices. Second, the
empirical research literature on eyewitness testimony has focused almost exclusively
on memory for criminal events and has not considered generalizing to other relevant
contexts such as occupational accidents.
Our purpose in this paper is to begin to rectify both omissions. First, we establish
some context by reviewing the role of accident investigations in health and safety
programming. Second, we briefly summarize the literature on eyewitness testimony
with a specific focus on identifying similarities and differences between accident and
criminal investigations. Finally, we attempt to bridge these literatures by framing
specific questions for future research. It is our belief that both field of enquiry benefit
from this endeavor. Accident investigations could be designed and conducted on a
strong empirical base while theories or hypotheses about eyewitness testimony could
be tested in a situation analogous to criminal investigations thereby providing a basis
for generalization.
THE ROLE OF ACCIDENT INVESTIGATIONS IN HEALTH
AND SAFETY PROGRAMS
In 2001 (the last year in which data are available), the U.S. Bureau of Labor
Statistics reported 5,900 workplace fatalities and 5.2 million occupational illnesses
and injuries. Moreover, the occupational fatality rate exceeds the annual death rate
attributable to breast cancer, prostate cancer, colorectal cancer, firearms, and AIDS
(Leigh, Markowitz, Fah, Shin, & Landrigan, 1997; Sauter, Hurrell, Fox, Tetrick, &
Barling, 1999). These figures become more striking when one recognizes that data on
occupational injuries tend to be underreported (Conway&Svenson, 1998; Eisenberg
& MacDonald, 1988).
In addition to the human costs of workplace fatalities and injuries, accidents
exert serious negative financial effects on organizations. For example, Dupr´e (2000)
estimated that in approximately half the accidents that occurred in the European
Union in 1996, the resulting absence from work was between 2 weeks and 3 months.
In United States, 80 million days of lost productivity were associated with workplace
accidents in 1998 (United States Census Bureau, 2000). Leigh et al. (1997) estimated
the 1992 costs of injuries conservatively at $145 billion. In Canada the cost of each
workplace injury is estimated to be $6,000, with the cost of each workplace fatality
estimated to be $492,000 (Marshall, 1996).
Not surprisingly, human resource and safety professionals have increasingly
turned toward systematic safety programming in an attempt to reduce injuries,
fatalities, and costs. The conduct of an investigation after each workplace accident
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 117
is frequently cited as a critical component of a firm’s occupational health and safety
program (Montgomery & Kelloway, 2002). Such investigations are aimed at recreating
the events leading up to the accident and identifying the causes of the event.
The results of such investigations are used to form preventative safety policies and
practices that will result in a reduced likelihood of reoccurrence.
The principal role of the accident investigation is to prevent future occurrences
of a similar nature. Such investigations serve this purpose through a variety of routes
including, but not limited to, the determination of direct and contributing causes,
the prevention of similar accidents, the creation of a permanent record to be used in
further analysis, and programming and raising safety awareness (Ferry, 1988; Laing,
1992).
Although the importance of effective investigation procedures to a health and
safety program cannot be overstated, the reality of such investigation stands in
stark contrast to the list of goals described above. Ferry (1988, p. 3) describes the
modal pattern when he says “Most mishaps are investigated by persons without
any investigative background who have no particular approach to the task. They
usually have minimal resources to meet minimum company or government
regulations.”
The literature on accident investigations typically offers only rudimentary suggestions
on interviewing techniques (e.g., Ferry, 1988; Montgomery & Kelloway,
2002) or the provision of some broad outlines for investigation as guides to the investigators.
For example, Montgomery andKelloway (2002) follow standard practice
in suggesting that investigators need to focus on three broad areas in their investigation;
human factors (e.g., What was the worker doing at the time of the accident?
Was he or she performing a regular task, a different task, doing maintenance work,
or helping a coworker?Were the tasks or procedures new?What was the posture and
location of the employee?); situational factors (e.g.,What was the site/location of the
accident? What tools and equipment or objects were involved in the accident? Was
the correct equipment available and being used to do the job?What personal protective
equipment (gloves, goggles, etc.) was being worn? Were guards in place? What
time of day did the accident occur?); and environmental factors (e.g., noise, heat,
light).
Indeed much of this literature is devoted to developing “models” of accident
causation (e.g., Reason, 1990) that provide organizing schemas for the results of the
accident investigation. Comparatively little attention is paid to the actual process of
data collection. For example, in their report on accident investigation techniques,
Livingston, Jackson, and Priestley’s sole reference to data collection (as opposed to
data organization and analysis) is to observe that:
The first stage of the incident investigation involves obtaining a full description of the
sequence of events which led to the failure. This will require interviews with key personnel
and examination of the physical evidence in order to piece together the circumstances of
the incident (Livingston et al., 2001, p. 4).
Implicit in this approach is the assumption that witnesses to an accident act
as neutral and accurate recording devices. The role of the investigator is merely to
elicit those recordings by asking a comprehensive series of questions covering the
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
118 Kelloway, Stinson, and MacLean
human, situational, and environmental elements of the situation.We suggest that the
burgeoning literature on criminal investigations (e.g., Fisher, 1995), and research on
eyewitness testimony (e.g., see reviews by Haber&Haber, 2000, andWells&Olson,
2003), casts this assumption into considerable doubt.We now turn to a review of this
literature with reference to accident investigation.
ACCIDENTS AS NEGATIVE EMOTIONAL EVENTS
We begin our analysis with the recognition that a workplace accident3 or incident
prompting an investigation shares some similarity with the “crime scenes” that
have provided the focus for much of the eyewitness literature. The experience of
witnessing a crime may be akin to that of witnessing a workplace accident. For eyewitnesses
to crimes and workplace accidents, the events are usually unanticipated,
can result in death or injury, and are frequently traumatic to both the witnesses and
the participants. Christianson (1992, p. 285) describes negative emotional events as
“distinct events or scenes that have unpleasant visual features (e.g., blood, injuries)
and have the potential to invoke strong unpleasant feelings (emotional stress) in the
viewer.” Pratt and Barling (1988) referred to such events as acute or catastrophic
stressors, having a specific time of onset, a limited duration and evoking intensely
negative reactions among the participants/witnesses. Indeed, witnesses to and victims
of both violent crimes and workplace accidents may suffer from some form of posttraumatic
stress symptoms (e.g., Barling, Bluen, & Fain, 1987; LeBlanc & Kelloway,
2002; Rogers & Kelloway, 1997; Schat & Kelloway, 2000, 2003; Schooler & Baum,
1999).
In addition to some topographical similarity between criminal activity and occupational
accidents, there is also some similarity in the interviewing process, from
the perspective of both the interviewer and interviewee (i.e., the eyewitness). Accident
investigations typically include interviews with eyewitnesses or coworkers
whose accounts of the events may help investigators identify the factors that caused
or contributed to an accident or the severity of the consequences. Criminal investigations
share some of the same elements of accident investigations. Law enforcement
officers usually gather evidence to find leads and often to determine
the identity of the perpetrator; sometimes that evidence includes eyewitness
testimony.
As Wells (1995) points out, eyewitness testimony can be compared to other
forms of physical evidence because an eyewitness has a memory trace in the brain.
Like physical evidence, this memory trace may be altered or destroyed if it is not
“handled” properly. Considerable research has now identified factors that affect the
integrity of the memory trace in criminal investigations and we suggest that these
results have direct implications for occupational accident investigation.We begin our
review by discussing the research on the factors that impact eyewitness testimony,
identifying the similarities and differences between witnessing aspects of accident
3The classic definition of an accident is an unplanned and unpredictable event resulting in the loss of
property or personal injury. We use this definition recognizing that health and safety professionals now
frequently object to the term accident and the inference that such events are unpredictable.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 119
and criminal investigations, and pinpointing the gaps in our knowledge of eyewitness
memory that would prevent generalizations to workplace accidents.
FACTORS AFFECTING THE MEMORY TRACE
To illustrate the importance of eyewitness memory on accident investigations,
we point to a report of an airplane crash that killed all nine people aboard. Dozens of
bystanders witnessed the crash. One eyewitness insisted at a formal hearing that the
airplane nose-dived straight into the ground. By coincidence, several photographs
had been taken moments before the crash that proved the airplane coasted down and
skidded for nearly 1,000 ft (Flying Magazine, 1977, cited by Loftus & Doyle, 1997).
In this case, the evidence clearly refuted the convinced eyewitness’s testimony illustrating
our contention that memory for accidents is fallible. Unfortunately, accident
investigators rarely have alternative forensic evidence that definitively identifies the
precise sequence of events or the cause(s) of the mishap. Moreover, a lack of research
attention has resulted in the fallibility of memory being largely overlooked in the
health and safety literature.
What we know about eyewitness memory comes from hundreds of studies, most
of which attempt to recreate some of these elements by exposing research participants
to a staged crime and measuring their memory for the event and perpetrator.
Overall, this body of research tells us that eyewitness testimony is not like a videotape
recorder; memory is fragile, malleable, and susceptible to forgetting, even in
optimal conditions (for reviews see Cutler & Penrod, 1995; Loftus & Doyle, 1997;
Ross, Read, & Toglia, 1994). Threats to the accuracy of eyewitness testimony may
be present during the encoding or acquisition of the memory (when the eyewitness
perceives the event), storage (the time lapse between the event and the subsequent
attempt to recall the event), and retrieval (when the eyewitness accesses the memory
of the event).
Threats at Encoding
Several factors can affect eyewitness memory during the encoding of information.
Haber and Haber (2000), for example, suggest that encoding of information
will be impaired if the witness cannot perceive the event, or is attending elsewhere.
Diverse factors may affect the perception of the event. For example, insufficient light
obviously impairs the encoding of information, but changes in lighting (from good
illumination to poor and vice versa) can also cause difficulties in seeing (Loftus &
Doyle, 1997). Moreover, although researchers have long established that eyewitness
accuracy is positively correlated with the duration of the event, people are notoriously
poor at estimating the duration of time. Most people overestimate the duration
of events. In one study, people estimated the duration of a 30-s mock crime to be
nearly 2 min long (on average), and some people even estimated the crime to have
lasted over 15 min (Loftus, Schooler,&Boone, 1987). People are also poor estimators
of distance and speed (Loftus & Doyle, 1997).
These observations point to the likelihood that witnesses to an occupational accident
may not able to provide accurate estimates of time, sequencing, or distances—
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
120 Kelloway, Stinson, and MacLean
data frequently collected in the course of the investigation (Laing, 1992). Equally
troublesome is the notion that witnesses may not be attending to the sequence of
events leading up to the accident (Haber & Haber, 2000). Models of accident causation
often invoke the “domino sequence” in which an accident is viewed as the
outcome of a sequence of events (Montgomery & Kelloway, 2002) and the goal of
many accident investigations is to recreate this sequence through interviews with the
witnesses or victims. This attempt is based on the assumption that witnesses were
in fact attending to the relevant events in their environment. Again, the data would
suggest that this may not be a reasonable assumption.
In the context of criminal investigations, the presence of a weapon during the
commission of a crime tends to direct attention to the weapon rather than on the perpetrator
or other factors. Numerous studies and a meta-analytic review confirm that
weapon focus clearly impairs memory (Maas & Kohnken, 1989; Shaw & Skolnick,
1994; Steblay, 1992). Extrapolating from this data, Haber and Haber (2000) suggest
that the presence of a weapon at a crime scene narrows attention and that a
similar narrowing of attention will emerge whenever the event is dramatic, violent,
or distasteful to the witness. In the case of workplace accidents, it is possible that
sudden changes in the environment, or the introduction of a novel stimulus might
similarly narrow attention and impair encoding of events in the workplace. For example,
consider a hypothetical scenario in which Kay, a control room operator at a
manufacturing plant, notices a warning light and siren indicating the overpressurization
of a tank. Kay’s focus on that warning could impede her ability to notice other
signals on the control panel that might indicate another more serious problem is developing.
This situation would be exacerbated if control panel malfunctions activate
contradictory warning signals to flash (this was the situation that precipitated the
crash of Aero Peru Flight # 603 on October 2, 1996).
Negative emotions and stress reactions often elicited during a crime are also
typically present for witnesses to workplace accidents. The emotions elicited by
witnessing a workplace accident compared with a crime may be different in some
situations, however. Consider the following situation: Joe, who is mopping floors,
observes a coworker slip on a wet floor, trip over a bucket of water, fall, and suffer
a serious head injury. Joe (our eyewitness) may experience dismay at seeing his
coworker’s condition, anxiety in finding someone to call for help, and frustration
over the ambulance not arriving quickly enough. But Joe may also experience guilt
over creating the conditions that contributed to the accident (i.e., failing to alert
his coworker that the area was wet or failing to ensure that the floor was dry). He
may also experience fear that his carelessness and negligence caused his coworker’s
injury and that he may lose his job. He may worry that such an accident might happen
to him and that he might be similarly debilitating injury. In some situations,
eyewitnesses to crimes might experience a similar array of emotions (e.g., consider
motor vehicle accidents in which several drivers’ errors contributed to a serious
accident).
We believe that herein lies one gap in the eyewitness literature; there is a paucity
of research that explores the psychological factors (e.g., emotions of guilt, defensive
motivations) that are present in real-world events. Indeed, the vast majority of studies
in eyewitness memory test eyewitness memory for a stranger’s characteristics (i.e.,
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 121
facial features) or actions. How does knowing the “players” involved in a crime or
accident affect eyewitness memory?
Attitudes, Scripts, and Stereotypes
Attitudes affect memory in many ways. Making an attitudinal judgment about
someone (e.g., determining that Joe is a safe worker) increases the likelihood that
attitude-relevant information will be recalled (e.g., he completed the safety checks)
and that attitude-consistent inferences will be made (Fiske&Taylor, 1991). Attitudes
affect memory by inducing people to recall the attitude first rather than retrieving and
processing all the relevant bits of information about the person (Boon&Davies, 1988;
Loken, 1984). Organizations expend considerable resources in attempts to develop
safety consciousness (e.g., safety knowledge and safety behavior) and safety climate
(Barling, Loughlin, & Kelloway, 2002) in the organization and it is conceivable that
these attempts affect the encoding of memories about an accident.That is, individuals
may be more likely to encode information consistent with the safety climate of the
organization or to make inferences consistent with the company’s safety culture
(Zohar, 1980).
Moreover, there are now data to suggest that experiencing a workplace accident
may result in changes in attitudes. Specifically, Barling, Kelloway, and Iverson (2003)
found that individuals who had experienced an occupational injury reported a sense
of lost control and diminished trust in management. In turn, these attitudes lead
to a greater sense of job dissatisfaction and intent to turnover. Similarly, Cree and
Kelloway (1997) found that individuals who had experienced a workplace accident
subsequently reported a higher perception of risk in the workplace. How quickly
these changes in attitudes occur and how they might affect the memory trace remains
a question for future research.
A schema is a “cognitive structure that represents knowledge about a concept
or type of stimulus, including its attributes and the relations among those attributes”
(Fiske & Taylor, 1991, p. 98). Knowledge about the sequence of events is called
a script; people have scripts for numerous activities such as going to the movies.
Triggered when people encounter a familiar event, scripts serve to guide expectations,
makeinferences, process information, and fill in gaps. Reliance on scripts during recall
rather than on the original memory trace is more likely when script items are central
or highly related to the script and when the retention interval is longer (Greenberg,
Westcott, & Bailey, 1998).
Stereotypes are special schemas about roles that organize people’s expectations
of people who fit certain groupings (Fiske & Taylor, 1991). Stereotype-induced expectations
influence memory (Allport & Postman, 1947; Lenton, Blair, & Hastie,
2001; Sherman & Bessenoff, 1999, but see Treadway & McCloskey, 1989 for opposing
view). People activate stereotypes at encoding, particularly when their cognitive
resources are strained (Sherman & Bessenoff, 1999). Stereotypes may also affect
memory during the retrieval stage (discussed below). In summary, attitudes, scripts,
and stereotypes may all contribute to memory errors.
The role of scripts is likely to be especially salient in organizational contexts.
Many jobs consist of repetitive actions or familiar sequences of events. These scripts
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
122 Kelloway, Stinson, and MacLean
of safe work performance may be invoked to fill in gaps in actual memory. This
suggestion is consistent with empirical data suggesting that individuals are not able
to recall accurately a specific instance of a repeated event (Friedman, 1990; Linton,
1986). Scripts for the “typical” event sequence may be used to fill in the details when
recall fails.
With regard to stereotypes, we know that employees are able to make reliable
judgments about supervisory, coworker, and managerial commitments to safety
and that these perceptions shape employees’ own perceptions of risk in the workplace
(Cree & Kelloway, 1997). This suggests the possible influence of stereotyping
individuals as being safety conscious or not and gives rise to the possibility that
memories may be distorted by the invocation of these same stereotypes. Moreover,
a worker’s perceptions of risk, coupled with a defensive motivation to feel safe at
work may also play a role in a worker’s encoding, storage, and retrieval of information
related to a workplace accident.
We suggest that the role of attitudes, scripts, and stereotypes on memory for
crimes committed by a person unknown to the eyewitness may be different for workplace
accident situations. The attitudes of worker-witnesses toward coworkers who
were involved in a workplace accident should be better developed, stronger, and
more complex than the attitudes of an eyewitness who sees a stranger commit a
crime.Workers’ greater knowledge of their job should also make them more vulnerable
to relying on scripts than eyewitnesses in a criminal setting.
Threats During Storage
Passage of Time and Postevent Information
Ordinarily, police officers try to interview eyewitnesses to a crime as soon as
possible, but occasionally the interview does not take place for quite some time.
Delaying the interview increases the likelihood that postevent information or suggestion
will distort eyewitness reports (Hoffman, Loftus, Greenmun, & Dashiell,
1992). The deleterious effect of delay is exacerbated by the use of leading or suggestive
questioning because eyewitnesses are more likely to incorporate the false
information into their subsequent reports, particularly when the source of the misinformation
is credible (Toland, Hoffman, & Loftus, 1991). People who witness violent
events are particularly susceptible to the deleterious effect of postevent information,
presumably because misinformation is less likely to be at odds with the
weaker memory trace (Loftus & Doyle, 1997). Additionally, it is important to note
that cognitive processes in eyewitnesses themselves may contribute to the postevent
information problem. For instance, because people tend to remember themselves
in a favorable light, eyewitnesses’ own thoughts and desires may taint their reports
(Loftus & Doyle, 1997). As mentioned earlier, people’s sense of responsibility,
guilt, and fear may have a similar effect on the memory trace and the subsequent
report.
In the context of workplace accidents there is consistent evidence that witnesses
to traumatic events experience repetitive, uncontrollable, and intrusive thoughts or
memories of the event (see for example, Schooler & Baum, 1999). Moreover, the
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 123
witness to an accident may be required to repeat his/her version of the events several
times to different investigators (e.g., supervisor, health and safety specialist, insurance
investigator). There is a substantial body of literature suggesting that such reviews of
memory lead to systematic changes in the memory trace and that these changes may
be related to the nature or purpose of the review. In particular, Haber and Haber
(2000) note that individuals may actively restructure memories on repetition to make
sense of seemingly incoherent events.
A related issue involves generating possible causes for an accident. The vast
majority of studies looking at the postevent information effect on memory
have tested the impact of external sources of postevent information on memory performance
(e.g., Loftus & Palmer, 1974). However, accident investigators, coworkers,
and others may ask an eyewitness to a workplace accident to speculate on possible
causes. How does the generation of plausible reasons for the event affect the memory
trace?4 We have found no research on this question.
Factors During Retrieval
When eyewitnesses attempt to access their memories of the event, several factors
may contribute to errors in their testimony or may change their confidence in the
accuracy of their testimony.
Stereotypes
As mentioned above, stereotypes activated during the retrieval stage may contribute
to memory errors. When an event is difficult to retrieve, people rely on
stereotypes as a backup to “fill in the gaps” (Sherman & Bessenoff, 1999). Consequently,
stereotypes may cause people to attribute stereotype-consistent behaviors
to a person and fail to recall stereotype-inconsistent behavior (Sherman&Bessenoff,
1999).
Interviewing Strategy
The key to inducing eyewitnesses to recall as much accurate information as possible
about the target event lies in the interviewing strategy used by interviewers. The
standard police interview tends to consist of a series of closed-ended questions (e.g.,
“Did he have a weapon?”) that are often leading or suggestive (e.g., “The getaway
vehicle was a white van, right?”). Numerous studies have shown that the standard
police interviewing procedure is far less effective than a psychologically based approach
known as the cognitive interview (CI; Fisher, 1995; Fisher, Geiselman, &
Amador, 1989; Fisher, McCauley, & Geiselman, 1994). The CI allows the eyewitness
ample opportunities to recall the event without interruptions, and it also encourages
the use of imagery and context-reinstatement. More importantly, the CI elicits more
information from the eyewitness without increasing the reporting of incorrect information
(Fisher et al., 1994). Thus, the fact that eyewitness reports vary as a function
of interviewing technique highlights the importance of optimizing the interviewing
conditions.
4We thank the Editor for this suggestion.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
124 Kelloway, Stinson, and MacLean
In contrast to typical police procedures, accident investigators are frequently
advised to use open-ended questions (e.g., Montgomery&Kelloway, 2002) although
there are no data on how often this advice is followed. However, the popularity of
accident causation models and taxonomies of potential causes may lead investigators
to focus on some factors to the virtual exclusion of others.The direction of questioning
may, therefore, affect the nature of the retrieved memory trace.
Repeated Questioning
Eyewitnesses to both workplace accidents and crimes are often interviewed
repeatedly by several people. Interestingly, repeatedly interviewing eyewitnesses
whowere wrong about their testimony tends to inflate their confidence in the accuracy
of their testimony (Shaw, 1996; Shaw & McClure, 1996).
Memory Accuracy and Eyewitness Confidence
Typically, police officers ask eyewitnesses to indicate their level of confidence in
the accuracy of their testimony. Most of the literature examines the relation between
eyewitness confidence and lineup identification accuracy. These studies tend to show
that the confidence-accuracy correlation is small (Sporer, Penrod, Read, & Cutler,
1995) because there are numerous factors that affect eyewitness confidence (e.g.,
feedback such as “good, you identified the suspect”) that do not affect identification
accuracy and vice versa (Wells & Bradfield, 1998). Feedback confirming eyewitness
identifications moderates the relationship by inducing eyewitnesses to report having
stronger memories, optimal viewing conditions, and so forth. Note that confirmatory
feedback affects people’s evaluations of the encoding conditions, precisely the kinds
of questions that would be of interest to someone who was evaluating the information
provided by eyewitnesses (e.g., lawyers). Thus, when their suspicions are confirmed,
investigators may share their excitement with eyewitnesses, thus hopelessly (if inadvertently)
tainting the eyewitness’ subsequent reports.
Studies examining the relationship between eyewitness confidence and recall accuracy
have focused on memory for personal identifying attributes (e.g., age, height,
weight; Ebbesen & Rienick, 1998; Tollestrup, Turtle, & Yuille, 1994; Yarmey, Jacob,
& Porter, 2002; Yuille & Cutshall, 1986). In general, these studies reveal a small
to moderate correlation between eyewitness confidence and accuracy of memory
for criminal events (Yarmey, 1993; Yarmey, Jacob, & Porter, 2002). We have found
no research examining the eyewitness confidence-accuracy correlation in workplace
accidents.We propose that eyewitness confidence may affect workplace accident investigations
in at least three ways. First, accident investigators may weigh the reports
of more confident eyewitnesses more heavily than those of less confident eyewitnesses.
Of course, there is no problem if the memories of highly confident eyewitnesses
are correct; difficulties could arise when highly confident eyewitnesses are
incorrect in their testimony.5 Most investigations are conducted by individuals with
little training in proper investigative or interviewing techniques (Ferry, 1998). It is
highly unlikely that they have any special training on the psychology of memory
5We focus our attention here on memory issues, not on situations in which eyewitnesses to workplace
accidents intentionally distort their reports)
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 125
and decision-making; it is far more likely that investigators share many of the misconceptions
about eyewitness memory. Most people believe in a positive eyewitness
confidence—accuracy correlation, and most believe that our memories for events
are stable over time (i.e., do not decline; Haber & Haber, 2000). Second, giving
undue weight to the testimony of highly confident eyewitnesses may lead accident
investigators to change the nature of their search for information. They may disregard
once-plausible explanations for the accident or change the direction of their
investigation. Finally, other eyewitnesses to the event may be more vulnerable to suggestibility
effects when exposed to the reports of highly confident eyewitnesses. This
effect may be especially true in a workplace setting where employees are typically
well-known to each other.
Investigative Procedures and Context
Clearly, eyewitness memory is not perfect; it is susceptible to the influence of
numerous factors that usually decrease rather than increase the accuracy of the
testimony. Accident investigators rely on employee eyewitness accounts, so accident
eyewitness reports should be susceptible to the same problems. How do accident
investigators actually handle memory evidence in their analysis of the event? We
know of no research exploring this question, but criminal investigations may shed
some light into this issue.
Unlike most employees who conduct accident investigators, police officers are
highly trained in matters dealing with the handling of physical evidence. Officers are
instructed to follow strict science-based procedures when handling physical evidence.
Law enforcement training and protocol relating to eyewitness testimony is far more
rudimentary; most police officers have little understanding of memory processes and
the social-cognitive factors that affect it (Wells & Loftus, in press). Given the lack of
specific training in investigative techniques (Laing, 1992), accident investigators are
probably similarly ignorant of these psychological processes.
Interestingly but not surprisingly, health and safety professionals have made
interviewing recommendations that either contradict or are unsupported by psychological
research. For example, accident investigators are advised to interview first
people who were not directly involved in the incident (e.g., senior management) so
that the investigator has a good idea of what happened (Ammerman, 1998). This
strategy has at least three drawbacks. First, it provides the investigator with schemas
that will influence the both the investigator’s understanding of the incident and his
or her subsequent interviews with the key players involved.6 Second, delaying the
interviews with those who were directly involved provides more opportunities for
postevent information and contamination from other witnesses, supervisors, etc. to
impair their memories; the effect of postevent information is positively correlated
with the duration of the retention interval (Belli, Windschitl, McCarthy, & Winfrey,
1992). Finally, knowing that the investigator consulted with senior management first
may increase the stress and fear of those who were directly involved in the incident.
Their motivation to defend their actions (or inactions), defend the actions or inac-
6Of course, there is no problem if the schemas are accurate. The difficulty arises when the schemas are
wrong because it is difficult to change schemas (see Fiske & Taylor, 1991, for a review).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
126 Kelloway, Stinson, and MacLean
tions of respected coworkers, and provide a clear and complete description of the
events preceding and following the incident will be affected.
Another interviewing strategy that seems at odds with psychological research is
the use of “ok,” “uh huh,” and “okay” during interviews. Ammerman (1998) advocates
the use of such expressions to convey the “impression of responsiveness and
attentiveness” (p. 55). Psychological research would caution against the use of these
expressions because they may convey approval of the eyewitness’ testimony and
encourage the eyewitness to provide testimony that elicits the same approval.
The criminal investigations involving eyewitnessesmayalso be different from occupational/
workplace accident investigations in several respects. Most eyewitnesses
sincerely try to help police investigators and do not deliberately distort their reports
(Wells, 1993). The same may not be true for witnesses to workplace accidents.
Coworkers who witnessed an accident may intentionally cover up for a victim. This is
particularly likely in organizational contexts that penalize unsafe work performance.
Just as government sanctions might lead organizations to underreport accidents and
injuries, organizational sanctions may lead individuals to suppress or distort their
reports of accidents.
There are some data to support this suggestion. Reason, Parker, and Lawton
(1998) note that management may well react to injuries, especially serious injuries,
by tightening their use of procedures and rules, thereby exerting greater control.
In their development of an accident reporting system for the offshore oil industry,
Gordon, Mearns, Flin, O’Connor, and Whitaker (2000, p. iii) reported that “The
main problem in gathering human factors causal data was respondents’ reluctance
to given open and candid responses to the forms.”
Unlike in many real life situations, the research models used in the eyewitness
memory literature reflect the assumption that there is a single “truth” (e.g., the
perpetrator committed the robbery or the vehicle failed to yield to the oncoming car).
Indeed, most experiments on eyewitness memory involve the use of one perpetrator
(for exceptions, see Clifford & Hollin, 1981; Geiselman, MacArthur, & Meerovitch,
1993). In contrast, workplace accidents are usually much more complex and focus on
“why” events occurred rather than just “what” events occurred in a specific incident.
For example, the theory of normal accidents (Perrow, 1984, 1994) particularly in
high reliability organizations (e.g., chemical plants, nuclear plants; Weick, Sutcliffe,
& Obstfield, 1999) suggests that accidents result from the interactive complexities in
the technological system. That is, there is no single event that causes an accident and
the search for a single discrete cause, analogous to a single perpetrator, might well
be fruitless in such an environment. The futility of the endeavor may be difficult to
recognize given the common tendency to make sense out of organizational events.
As Weick (1995, p. 28) notes “people who know the outcome of a complex prior
history of tangled, indeterminate events, remember that history as being much more
determinant, leading ‘inevitably’ to the outcome they already knew.”
In many workplace accidents the actions or inactions of the accident victim and
other coworkers may contribute to the event. The eyewitnesses are usually coworkers
who know the victim, so schemas, expectations, and affect may affect memory
(Fiske & Taylor, 1991). Information that is consistent with a schema is more likely to
be recalled than information that is deemed to be irrelevant (Fiske & Taylor, 1991),
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 127
so office politics may shape the witnesses’ subsequent memories of the incident and
attributions of responsibility. Organizational safety culture (Zohar, 1980) may also
play a role in shaping the witnesses’ memories of the incident and their attributions
of responsibility. Moreover, adverse consequences for employees who were involved
in a workplace mishap may shape both their reports and those of their coworkers.
Research tells us that people tend to remember more information about in-group
members than out-group members, and they tend to recall more negative information
about out-group members (Fiske & Taylor, 1991). Other factors, such as union
membership (Kelloway, in press), may contribute to the eyewitnesses and victim
adopting a particular perspective and recalling only certain aspects of the event.
Given considerable ambiguity inherent in assigning causality for a workplace
accident, attributional errors are likely. Defensive attribution refers to the notion
that people will assign greater responsibility to people when their actions lead to
severe rather than minor consequences (Shaver, 1970). However, people tend to
attribute less responsibility to an accident perpetrator whose actions lead to serious
consequences when they are similar to the perpetrator. When people are dissimilar
to an accident perpetrator whose actions lead to severe consequences, they tend to
attribute more responsibility to the perpetrator. Defensive attributions also affect
eyewitness recall memory and related judgments in a way that reduces the perceived
threat of seeing someone similar to the eyewitness be a crime victim (Marsh, 1997).
Although we have found no empirical research supporting this notion, defensive
attributions might explain why employees tend to place blame for their injuries on
management. The fundamental attribution error refers to the tendency for people to
attribute other people’s behavior to internal stable factors (e.g., personality) and to
underestimate the role of external situational factors (Fiske & Taylor, 1991). It leads
to the prediction that employees will assign blame for their injuries to managers. In
both cases, the role of personal and situational factors would be minimized (Hofmann
& Stetzer, 1998). Clarke (1999) for example, found that train drivers erroneously
estimated that their supervisors and managers had less knowledge, and cared less
about occupational safety than they actually did.
Eyewitnesses are not the only ones who are affected by schemas and expectations.
Like law enforcement officers who sometimes have a suspect in mind when they
interview eyewitnesses, accident investigators often expect the accident victim to be
at fault for the workplace accident. This bias probably induces investigators to look
for information that confirms their hypothesis regarding the cause of the accident
and ignore other potentially relevant information. This is particularly troublesome in
light of the often-cited, but rarely substantiated, claim that “Human error is the cause
of most accidents” (Dekker, 2002, p. 372). As Dekker (2002, p. 372) notes “An investigators’
emphasis on proximal causes ensures that the mishap remains the result of
a few uncharacteristically ill-performing individuals who are not representative: : :of
the larger population.” These expectations probably guide their search for evidence
as well as their analysis of the information gleaned. Thus, most investigators (both
police and accident) don’t realize that their expectations guide their thinking about
their case. The tunnel vision problem is serious, so much so that investigation reports
of wrongful convictions have advised police departments to provide annual training
to police officers (www.gov.mb.ca/justice/sophonow).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
128 Kelloway, Stinson, and MacLean
There is one aspect of accident investigations that differs from that of police
investigations. In accident investigations, managers typically have the responsibility
of investigating accidents. Because managers are ultimately responsible for anything
that happens on their watch, they have a personal role in the investigation. AsVincoli
(1993) points out, “the fact that an accident occurs is a strong indication that a manager
somewhere within the organization has made a bad decision” (p. 11; e.g., a floor
manager may have erred in the job assignment, there was some miscommunication
regarding the proper use of equipment, etc.). Thus, accident investigators have a
vested interest in the outcome of their investigation. They will be motivated to find
that the accident was, for example, due to employee carelessness and not to some
factor under the control and responsibility of the manager. Police investigators are
certainly motivated to solve crimes, but their investigations do not typically involve
inquiries into their own shortcomings or errors.
IMPLICATIONS FOR RESEARCH
Perhaps the most striking implication of our review is the obvious need for more
research on workplace accident investigations. Although there are many manuals and
instructional guides telling investigators “what” to look for in conducting an accident
investigation, the almost universal assumption is that witnesses and victimswhomight
be interviewed are both (a) capable of recalling the exact sequence of events, and
(b) willing and capable of disclosing all relevant information. Indeed in preparing
this paper we consulted a colleague who specializes in health and safety and were
promptly informed that we were off track because “everyone knows what happened”
in any given accident. We know of no data that would justify this position and have
reviewed a number of studies conducted in the context of criminal investigations that
would suggest otherwise.
We suggest that a more fruitful starting point is to recognize that the process of
investigation is an attempt to retrieve a memory trace (Wells, 1995) and that memory
is fragile, malleable, and susceptible to forgetting, even in optimal conditions. Starting
from this position points to the need to identify the factors that may distort or change
individual memories of the event. In the long run, such research would also point to
more effective investigative techniques. We believe that some of the data collected
in the study of criminal investigations should readily transfer to the health and safety
setting allowing researchers to build on a considerable body of empirical data that
has already accumulated.
We suggest that a starting point for a research agenda on accident investigation is
to explore the accuracy of eyewitness testimony in this context. Noting that accidents
often unfold over a considerable period of time and are rarely preceded by unusual
events that would act as warning signs (Dekker, 2002) the conditions for eyewitness
accuracy would seem to be suboptimal. However, given the absence of empirical
evidence on accident investigations, we believe there is some value to research simply
demonstrating the generalizability of eyewitness findings to this new domain.
In identifying workplace accident investigations as a focal point for future research,
we believe that there is an opportunity to test the generalizability of findings
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 129
from the forensic setting. Much of the research we reviewed has been based on simulated
or actual crime settings. Although this is an appropriate focus given the intent
of the research, reliance on a small set of situations gives rise to the possibility of constructing
and disseminating a truncated body of knowledge.We have proposed that
workplace accident investigation may provide an analogue to criminal settings and
that both health and safety and forensic researchers would benefit from expanding
the research focus in this way.
In our analysis of the eyewitness literature and the “practice” of accident investigations
we identified several variables that should be studied further and several
questions that researchers could address. First, in accident investigations the eyewitnesses
may (a) know the victims and perpetrators (if any), (b) have well-established
scripts for how these individuals perform tasks in the workplace on a day-to-day basis,
and (c) experience emotional states such as guilt or fear. In brief, eyewitnesses to
accidents are intimately connected to the individuals and events involved. How does
this level of connection (which may also be present in some criminal investigations)
affect eyewitness testimony?
Second, there are good grounds to suggest that (a) individuals actively sort
through and reorganize information to make sense of complex information (e.g.,
Dekker, 2002;Weick, 1995); (b) accidents are most likely the result of multiple causes
(e.g., Perrow, 1984) some of which may not be readily observable; and (c) there is
a tendency for accident witnesses and investigators to blame individual rather than
system causes (e.g., Dekker, 2002). There is also a tendency for eyewitnesses to cognitively
restructure the event witnessed to place some distance from the victim and
reduce threat (Janoff-Bulman, 1982). Is there a bias toward identifying a perpetrator
in accident investigations? If so, how does this impact on eyewitness testimony?
Third, what is the impact of having investigations conducted by managers who
might bear some responsibility for the incident under investigation? How does management
style, or management reaction to an accident, affect eyewitness accounts?
Does a punitive management style result in distorted accounts?
Finally, does the extensive research database on eyewitness testimony in criminal
investigations provide grounds for improving eyewitness testimony in accident
investigations? Do techniques such as the Cognitive Interview or other interviewing
techniques result in more detailed, more accurate, or less biased eyewitness accounts?
Accident investigations have as their primary purpose to find out why accidents
occurred and to prevent similar accidents in the future. We suggest that this goal is
unlikely to be achieved if investigations are predicated on unrealistic assumptions
about the quality of the data. Assessment of the accident investigation process and
development of more effective techniques are a suitable focus for future research. It
is our hope that this review might stimulate research in this area.
ACKNOWLEDGMENTS
Preparation of this paper was supported by grants from the Social Sciences
and Humanities Research Council of Canada and the Nova Scotia Health Research
foundation to the first author.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
130 Kelloway, Stinson, and MacLean
REFERENCES
Allport, G.W., & Postman, L. J. (1947). The psychology of rumor. Oxford, England: Holt.
Ammerman, M. (1998). The Root Cause Analysis Handbook. NY: Productivity Inc.
Barling, J., Bluen, S. D., & Fain, R. (1987). Psychological functioning following an acute disaster. Journal
of Applied Psychology, 72, 683–690.
Barling, J., Kelloway, E. K., & Iverson, R. (2003). Accidental outcomes: Attitudinal consequences of
workplace injuries. Journal of Occupational Health Psychology, 8, 74–85.
Barling, J., Kelloway, E. K., & Zacharatos, A. (2002). Occuptional safety. In P. B.Warr (Ed.), Psychology
at work (5th ed). London: Penguin.
Barling, J., Loughlin, C.,&Kelloway, E. K. (2002). Development and test of a model linking safety-specific
transformational leadership and occupational safety. Journal of Applied Psychology, 87, 488–496.
Belli, R. F., Windschitl, P. D., McCarthy, T. T., & Winfrey, S. E. (1992). Detecting memory impairments
with a modified test procedure: Manipulating retention interval with centrally presented event items.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 356–367.
Boon, J. C.W., & Davies, G. (1988). Attitudinal influences on witness memory: Fact and fiction. In M. M.
Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Practical aspects of memory: Current research and
issues (Vol. 1, pp. 53–58). New York:Wiley.
Christianson, S. A. (1992). Emotional stress and eyewitness memory: A critical review. Psychological
Bulletin, 112, 284–309.
Clarke, S. (1999). Perceptions of organizational safety: Implications for the development of safety culture.
Journal of Organizational Behavior, 20, 185–198.
Clifford, B. R., & Hollin, C. R. (1981). Effects of the type of incident and number of perpetrators on
eyewitness memory. Journal of Applied Psychology, 66, 364–370.
Conway, H.,&Svenson, J. (1998). Occupational injury and illness rates 1992–1996:Why they fell. Monthly
Labor Review, 121, 36–58.
Cree, T., & Kelloway, E. K. (1997). Responses to occupational hazards: Exit and participation. Journal of
Occupational Health Psychology, 2, 304–311.
Cutler, B. L.,&Penrod, S.D. (1995). Mistaken identification: The eyewitness, psychology, and the law. New
York: Cambridge University Press.
Dekker, S. W. A. (2002). Reconstructing human contribution to accidents: The new view on error and
performance. Journal of Safety Research, 33, 371–385.
Dupr´ e, D. (2000). Accidents at work in the E.U. in 1996. Statistics in focus: Population and Social Conditions,
3(4), 1–7.
Ebbesen, E.B.,&Rienick,C.B. (1998). Retention interval and eyewitness memory for events and personal
identifying attributes. Journal of Applied Psychology, 83, 745–762.
Eisenberg, W. M., & MacDonald, H. (1988). Evaluating workplace injury and illness records: Testing a
procedure. Monthly Labor Review, 111, 58–60.
Ferry, T. D. (1988). Modern accident investigation and analysis (2nd ed). New York:Wiley.
Fisher, R. P. (1995). Interviewing victims and witnesses of crime. Psychology, Public Policy, and Law, 1,
732–764.
Fisher, R. P., Geiselman, R. E., & Amador, M. (1989). Field test of the cognitive interview: Enhancing the
recollection of actual victims and witnesses of crime. Journal of Applied Psychology, 74, 722–727.
Fisher, R. P., McCauley, M. R., & Geiselman, R. E. (1994). Improving eyewitness testimony with the
Cognitive Interview. In D. F. Ross, J. D. Read, & M. P. Toglia (Eds.), Adult eyewitness testimony:
Current trends and developments (pp. 245–272). New York: Cambridge University Press.
Fiske, S. T., & Taylor, S. E. (1991). Social Cognition (2nd ed.). New York: McGraw-Hill.
Friedman,W. J. (1990). About time. Cambridge, MA: MIT Press.
Geiselman, R. E., MacArthur, A.,&Meerovitch, A. (1993).Transference of perpetrator roles in eyewitness
identifications from photoarrays. American Journal of Forensic Psychology, 11(4), 5–15.
Gordon, R., Mearns, K., Flin, R., O’Connor, P., & Whitaker, S. (2000). Factoring the Human into Safety:
Translating Research into Practice. Vol 2: The Development and evaluation of a human factors accident
and near miss reporting form for the offshore oil industry. Aberdeen, UK: University of Aberdeen.
Greenberg, M. S., Westcott, D. R., & Bailey, S. E. (1998). When believing is seeing: The effect of scripts
on eyewitness memory. Law and Human Behavior, 22, 685–694.
Haber, R. N., & Haber, L. (2000). Experiencing, remembering and reporting events. Psychology, Public
Policy, and Law, 6, 1057–1097.
Hoffman, H. G., Loftus, E. F., Greenmun, N., & Dashiell, R. L. (1992). The generation of misinformation.
In F. Losel,D. Bender,&T. Bliesener (Eds.), Psychology and law: International perspectives (pp. 292–
301). Berlin: Walter de Gruyter.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Accident Investigation 131
Hofmann, D. A., & Stetzer, A. (1998). The role of safety climate and communication in accident interpretation:
Implications for learning from negative events. Academy of Management Journal, 41,
644–657.
Janoff-Bulman, R. (1982). Esteem and control bases of blame: “Adaptive” strategies for victims versus
observers. Journal of Personality, 50, 180–192.
Kelloway, E. K. (2003). Labor unions and occupational safety: Conflict and cooperation (pp. 249–264). In
J. Barling & M. Frone (Eds.), Psychology of occupational safety.Washington, DC: APA Books.
Laing, P. (1992). Accident prevention manual for business and industry: Administration and programs (10th
ed.).Washington, DC: National Safety Council.
LeBlanc, M., & Kelloway, E. K. (2002). Predictors and outcomes of workplace violence and aggression.
Journal of Applied Psychology, 87, 444–453.
Leigh, J. C., Markowitz, S. B., Fah, M., Shin, C., & Landrigan, P. J. (1997). Occupational injury and illness
in theU.S.: Estimates of cost, morbidity, and mortality. Archives of Internal Medicine, 157, 1557–1568.
Lenton, A. P., Blair, I. V., & Hastie, R. (2001). Illusions of gender: Stereotypes evoke false memories.
Journal of Experimental Social Psychology, 37, 3–14.
Linton, M. (1986).Ways of searching and the contents of memory. InD. C. Rubin (Ed.), Autobiographical
memory (pp. 50–67). New York: Cambridge University Press.
Livingston, A. D., Jackson, G., & Priestley, K. (2001). Root causes analysis: Literature review. Contract
Research Report 325/2001. London, UK: Health and Safety Executive.
Loftus, E. F., & Doyle, J. M. (1997). Eyewitness testimony: Civil and criminal (3 rd ed). Charlottesville,
VA: Lexis Law.
Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction
between language and memory. Journal of Verbal Learning and Verbal Behavior, 13, 585–
589.
Loftus, E. F., Schooler, J. W., & Boone, S. M. (1987). Time went by so slowly: Overestimation of event
duration by males and females. Applied Cognitive Psychology, 1, 3–13.
Loken, B. (1984). Attitude processing strategies. Journal of Experimental Social Psychology, 20, 272–296.
Maas, A., & Kohnken, G. (1989). Eyewitness identification: Simulation the “weapon effect.” Law and
Human Behavior, 13, 397–408.
Marsh, D. P. (1997). Defensive attribution, witness’ similarity to the victim of a criminal incident, and
implications for the witness’ recall of the incident. Unpublished doctoral dissertation, University of
Pittsburgh, CA.
Marshall, K. (1996, Summer). A job to die for. Perspectives on Labour and Income. 26–31.
Montgomery, J., & Kelloway, E. K. (2002). Management of occupational health and safety, (2nd ed.).
Toronto: Nelson.
Perrow, C. (1984). Normal accidents: Living with high-risk technologies. New York: Basic Books.
Perrow, C. (1994). Accidents in high risk systems. Technological Studies, 1, 1–20.
Pratt, L. I., & Barling, J. (1987). Differentiating between daily events, acute and chronic stressors: A
framework and its implications. In J. R. Hurrell, L. R. Murphy, S. L. Sauter, & C. L. Cooper (Eds.)
Occupational stress: Issues and developments in research (pp. 41–53). London: Taylor & Francis.
Reason, J. (1990). Human error. Cambridge, UK: Cambridge University Press.
Reason, J., Parker, D., & Lawton, R. (1998). Organizational controls and safety: The varieties of rulerelated
behaviour. Journal of Occupational and Organizational Psychology, 71, 289–304.
Roberts, K. (1989). Some characteristics of high reliability organizations. Organization Science, 2, 160–
176.
Rogers, K., & Kelloway, E. K. (1997). Violence at work: Personal and organizational outcomes. Journal
of Occupational Health Psychology, 2, 63–71.
Ross,D. F., Read, J.D.,&Toglia, M. P. (1994). Adult eyewitness testimony: Current trends and developments.
New York: Cambridge University Press.
Ryan, T. (1991). Accident investigations: Group investigations. In F. Briggs (Ed.), Guide to health and
safety management. Don Mills, ON: Southam Business.
Sauter, S. L., Hurrell, J. J., Fox, H., Tetrick, L. E., & Barling, J. (1999). Occupational health psychology:
An emerging discipline. Industrial Health, 37, 199–211.
Schat, A.,&Kelloway, E. K. (2000). Effects of perceived control on the outcomes of workplace aggression
and violence. Journal of Occupational Health Psychology, 5, 386–402.
Schat, A., & Kelloway, E. K. (2003). Reducing the adverse consequences of workplace aggression and
violence: The buffering effects of organizational support. Journal of Occupational Health Psychology,
8, 110–122.
Schooler, T. Y., & Baum, A. (1999). Memories of a petrochemical explosion: A cognitivephenomenological
study of intrusive thoughts. In L. Williams and V. L. Banyard (Eds.), Trauma
& memory (pp. 189–201). Thousand Oaks, CA: Sage.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
132 Kelloway, Stinson, and MacLean
Shaver, K.G. (1970). Defensive attribution: Effects of severity and relevance on the responsibility assigned
for an accident. Journal of Personality and Social Psychology, 14, 101–113.
Shaw, J. S., III (1996). Increases in eyewitness confidence resulting from postevent questioning. Journal
of Experimental Psychology: Applied, 2, 126–146.
Shaw, J. S.,&McClure, K. (1996). Repeated postevent questioning can lead to elevated levels of eyewitness
confidence. Law and Human Behavior, 20, 629–653.
Shaw, J. I., & Skolnick, P. (1994). Sex differences, weapon focus, and eyewitness reliability. Journal of
Social Psychology, 134, 413–420.
Sherman, J. W., & Bessenoff, G. R. (1999). Stereotypes as source

Confidence and Eyewitness Identifications: The Cross-Race Effect, Decision
Time and Accuracy
CHAD S. DODSON1* and DAVID G. DOBOLYI2
1Department of Psychology, University of Virginia, Charlottesville, USA
2McIntire School of Commerce, University of Virginia, Charlottesville, USA
Summary: Participants encountered same-race and cross-race faces at encoding, completed a series of line-up identification tests
and provided confidence ratings by using one of nine different confidence scales. Confidence was less well calibrated with identification
accuracy when participants selected a cross-race than a same-race face because of overconfidence. By contrast, there
was no cross-race effect on confidence–accuracy calibration when participants responded ‘not present’. Whereas confidence
was a very strong predictor of accuracy for fast identifications of a line-up face, this was much less the case for slower decisions.
Highly confident identifications showed a dramatic drop in accuracy from faster decisions to slower decisions, whereas there was
little change in accuracy between faster and slower decisions for moderately confident or weakly confident identifications. Finally,
we observed little influence of the format of the nine different confidence scales: numerical and verbal scales produced comparable
calibration scores, as did scales with few or many points. Copyright © 2015 John Wiley & Sons, Ltd.
Individuals show less accurate memory for cross-race than
same-race faces. This cross-race effect (CRE) is a welldocumented
phenomenon that occurs across a variety of
different kinds of memory tests (Meissner & Brigham,
2001; Sporer, 2001; see Young, Hugenberg, Bernstein &
Sacco, 2012, for review). Individuals are worse at recognizing
previously encountered cross-race than same-race faces
(e.g., Marcon, Susa & Meissner, 2009; Meissner, Brigham
& Butz, 2005; Rhodes, Sitzman & Rowland, 2013). They
are less able to remember source information about the
context in which they encountered cross-race versus samerace
faces (e.g., Horry & Wright, 2008; Horry, Wright &
Tredoux, 2010). And, of importance for the present paper,
individuals are generally less accurate at identifying a
cross-race face than a same-race face from a line-up of faces
(e.g., Evans, Marcon & Meissner, 2009; Jackiw, Arbuthnott,
Pfeifer, Marcon & Meissner, 2008; Smith, Stinson &
Prosser, 2004; Wright, Boyd & Tredoux, 2001).
One goal of the present paper is to examine the cross-race
effect on (a) confidence ratings that are given to line-up identifications
and (b) on the relationship between decision time,
confidence and identification accuracy. As a foundation for
the study, the following sections will briefly review research
on (1) the confidence–accuracy relationship, (2) the interplay
between decision time, confidence and accuracy, and (3)
ways of measuring confidence.
Confidence–accuracy relationship
The relationship between confidence and accuracy is frequently
measured in two different ways, broadly referred to
as relative and absolute measures of monitoring (e.g., Koriat
& Goldsmith, 1996). In the eyewitness literature, relative
measures of the confidence–accuracy relationship have been
measured with gamma scores, which refer to how well confidence
predicts overall accuracy, such as when individuals
give higher confidence ratings to accurate than to inaccurate
responses. One limitation, however, with relative measures
is that the particular levels of the confidence rating for correct
and incorrect responses can have no bearing on the strength of
the association between confidence and accuracy. For example,
the identical gamma score is obtained when individuals
use the lowest confidence rating (e.g., 0) for all incorrect responses
and they use either the second lowest confidence rating
(e.g., 10) or the highest confidence rating (e.g., 100) for all
correct responses. For gamma, the absolute confidence rating
does not matter; all that matters is the degree to which lower
confidence ratings are assigned to inaccurate than to accurate
responses. It is for this reason that these measures of the
confidence–accuracy relationship are referred to as relative
measures of monitoring because it is the relative difference
in confidence for correct and incorrect responses that matters
(e.g., Koriat & Goldsmith, 1996). But, as Juslin, Olsson &
Winman (1996) argued, because the particular value of the
confidence rating is not necessarily important, these relative
scores of the confidence–accuracy relationship can have little
practical value. Because police do not know a priori whether
an identification is correct or incorrect, they must focus on the
absolute level of confidence of the eyewitness.
Absolute measures of the confidence–accuracy relationship,
such as calibration and over–underconfidence scores,
focus on the particular level of the confidence rating that is
assigned to correct and incorrect responses (e.g., Juslin
et al., 1996; Koriat & Goldsmith, 1996). Perfect calibration
is shown when linear increases in confidence are paralleled
by linear increases in accuracy, such as when individuals
are poorly, moderately and highly accurate for responses that
are assigned low, moderate and high levels of confidence,
respectively. Knowing that a condition is associated with
excellent calibration is useful knowledge for investigators
because it indicates that high confidence is associated with
high accuracy and likewise for low confidence and low accuracy.
By contrast, it is impossible to make this inference
about high confidence and high accuracy from even an excellent
gamma score because it is possible that individuals
assigned a high confidence rating to both correct and
* Correspondence to: Chad Dodson, Department of Psychology, University
of Virginia, Charlottesville, VA 22904-4400, USA.
E-mail: cdodson@virginia.edu
Copyright © 2015 John Wiley & Sons, Ltd.
Applied Cognitive Psychology, Appl. Cognit. Psychol. 30: 113–125 (2016)
Published online 12 October 2015 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/acp.3178
incorrect responses [e.g., a confidence rating of 100 to all
correct and a rating of 90 (or even 99) to all incorrect responses
would produce a perfect gamma score].
There are factors in line-up identification studies that
reliably affect the confidence–accuracy relationship. Much
research shows that when participants choose a face from a
line-up—so-called choosers—they generally show a reasonable
relative relationship between confidence and accuracy
—e.g., Sporer et al. (1995) observed an rpb = .37 and
Sauerland and Sporer (2009) observed a similar correlation
score for choosers. Likewise, absolute measures of the
confidence–accuracy relationship, such as calibration scores,
also show a reasonable relationship for choosers (e.g., Weber
& Brewer, 2003; Weber, Brewer, Wells, Semmler and Keast,
2004; Brewer & Wells, 2006; Sauerland & Sporer, 2009).
For example, Sauerland and Sporer (2009) observed that
accuracy increased from nearly 20% to nearly 80% with
increases in confidence from the least confident (i.e., those
expressing 0–20% confidence) to the most confident
(i.e., those expressing 80–100% confidence). By contrast,
the confidence–accuracy relationship for nonchoosers
(i.e., those who respond ‘not present’) is much worse than
that for choosers. There frequently is no relationship at all
between the confidence in and accuracy of a nonchooser
response (i.e., confidence does not predict accuracy;
e.g., Sauerland & Sporer, 2009; see Sporer et al., 1995,
for a meta-analysis and similar findings). Similarly,
nonchoosers tend to show very poor calibration scores
because they exhibit similar levels of accuracy across all
levels of confidence (e.g., Brewer & Wells, 2006). Overall,
many studies have observed that confidence is both reasonably
correlated and calibrated with chooser accuracy, but
this is not the case for nonchooser accuracy.
Is there a cross-race effect on the relationship between
confidence and accuracy? Many studies show that individuals
give higher confidence ratings for correct recognition responses
to same-race than cross-race faces and that they give
lower confidence ratings to incorrect recognition responses
to same-race than cross-race faces (e.g., Horry & Wright,
2008; Rhodes et al., 2013). This pattern for confidence to
correct and incorrect responses means that confidence–
accuracy resolution (i.e., a relative measure) is better for
same-race than for cross-race faces, but it is impossible to
infer anything from these data about confidence–accuracy
calibration (i.e., an absolute measure). Further support for a
CRE on relative measures of the confidence–accuracy relationship
comes from both recognition and line-up identification
studies that have observed stronger gamma scores for
same-race than for cross-race faces (e.g., Corenblum &
Meissner, 2006; Wright, Boyd & Tredoux, 2001, 2003).
These data show that confidence better predicts the accuracy
of responses to same-race than cross-race faces. Notably,
Wright et al. (2001, 2003) are the only researchers to use a
line-up identification paradigm and observed that the relative
relationship between confidence and accuracy (e.g., gamma
scores) is no different from zero for cross-race identifications.
Is there a CRE on either confidence–accuracy calibration
or other absolute measures of monitoring (e.g., over–
underconfidence)? Are individuals prone to excessive overconfidence
when making cross-race line-up identifications?
The answers to these questions are unknown as no one has
examined them. But, given the value of this knowledge to investigators,
the answers to these questions are of great
importance.
Decision time, confidence and accuracy
Much research shows that there is a reliable relationship
between decision time and the accuracy of positive
identifications from lineups (i.e., choosers; e.g., Brewer &
Wells, 2006; Dunning & Perretta, 2002; Sauerland & Sporer,
2007, 2009; Smith, Lindsay, Pryke & Dysart, 2001; Sporer,
1993, 1994; Weber et al., 2004). Faster decisions are generally
more accurate than slower decisions. For example,
Sauerland and Sporer (2009) observed that faster line-up
decisions (i.e., 6 seconds or less) showed 72% accuracy,
but accuracy dropped in half (36% accuracy) for slower
decisions (i.e., >6 seconds). One drawback for real world
application, however, is that the optimal decision time for
an identification varies from study to study and is related to
retention interval and other factors (e.g., Brewer, Caon, Todd
& Weber, 2006; Sauerland & Sporer, 2007; Weber et al.,
2004). So, there does not appear to exist a single time interval
that one can point to as always indicating a fast or slow
decision. In addition, this relationship between decision time
and accuracy appears only to apply to choosers, as many
studies have observed that decision time is not meaningfully
associated with the accuracy of ‘not present’ responses to
lineups (e.g., Sauerland & Sporer, 2009).
Combining confidence and decision time has proven
particularly powerful for predicting accuracy of positive
identifications from lineups (e.g., Brewer & Wells, 2006;
Sauerland & Sporer, 2007, 2009; Weber et al., 2004). For
example, Sauerland and Sporer (2009) observed that fast
(6 seconds or less) and confident (90–100%) individuals
showed an impressive 97% accuracy rate when they selected
someone from a line-up. But, nonchoosers show a very different
pattern from choosers: the combination of confidence and
decision time provided little value at distinguishing between
correct and incorrect nonchooser responses (e.g., Brewer &
Wells, 2006).
What of the cross-race effect and confidence and decision
time for line-up identifications? Smith et al. (2001) were the
only researchers to examine this question, and their study is
problematic because they did not observe a CRE on line-up
identification performance; they found no significant differences
in accuracy between own-race choosers and cross-race
choosers. Thus, it is still an open question about the relationship
between confidence and decision time and the accuracy
of cross-race decisions.
Measuring confidence
Does the format of the confidence scale influence the relationship
between confidence and accuracy? When measuring
the confidence–accuracy relationship, does it matter if individuals
use numbers or words or if the scale contains few
or many points? Existing research by Weber, Brewer and
Margitich (2008) within the context of an eyewitness identification
task and by Wallsten and Budescu (1983) within the
context of a probability judgment task shows nearly identical
114 C. S. Dodson and D. G. Dobolyi
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
patterns of calibration of confidence to accuracy when participants
use either numeric or verbal confidence scales (see
also Wallsten, Budescu & Zwick, 1993; see Wallsten &
Budescu, 1995, for review). For example, Wallsten et al.
(1993) required participants to judge the likely accuracy of
general knowledge statements with either numeric or verbal
confidence scales and observed nearly identical calibration
scores for both confidence formats. Despite these findings,
many eyewitness researchers assume that calibration scores
can be computed properly only when (a) a numerical scale
is used and (b) in the context of identifications from a typical
line-up that the confidence scale must range from 0% to
100% (e.g., Wells & Penrod, 2011). We examine these latter
two assumptions in our study.
We used nine different confidence scales to examine
whether the format of the confidence scale matters when
measuring the relationship between confidence and accuracy.
First, we used a variety of different verbal and numeric
scales that allow us to replicate the findings of Weber et al.
(2008) and Wallsten and colleagues (e.g., Wallsten et al.,
1993) about a similar confidence–accuracy relationship with
verbal and numeric scales. Second, we varied the number of
points on the scale (6 points vs. 11 points vs. 101 points) to
examine the hypothesis that more points will increase the
sensitivity of the scale. Third, all scales included identical
verbal anchors at the endpoints (i.e., not at all confident
and completely confident). But, we included numeric scales
that ranged either from 0% to 100% or from 50% to 100%.
Typically, calibration studies use numeric scales that correspond
to chance performance, such as using a 50% to
100% confidence scale for a two alternative forced choice
task because chance accuracy is 50%. We deliberately used
different numeric ranges in order to examine the assumption
that assessing calibration with a typical line-up requires that
the confidence scale ranges from 0% to 100%. Put another
way, we examine how flexible individuals are at mapping a
sense of confidence onto a confidence scale that substantially
varies in range (i.e., 0–100% vs. 50–100%), as well as other
dimensions, such as verbal versus numeric format. Finally,
for the verbal confidence scales, we manipulated whether every
point on the verbal scale was labeled or only the endpoints
on the scale were labeled. In other words, one would
expect that when given a 6-point or an 11-point scale—ranging
from ‘not at all confident’ to ‘completely confident’—
that individuals should be better calibrated at mapping their
subjective sense of confidence on to a verbal scale that has
a label at every point, as opposed to a scale that consists of
unlabeled points except for the anchor labels at the endpoints.
Overall, these nine different confidence scales allow
us to examine the degree to which confidence–accuracy relationships
(e.g., calibration) are tied to the particular format of
the confidence scale (see the Procedure section for a list and
example of each of the scales).
Overview
This study answers a number of questions about the crossrace
effect on line-up identification accuracy and the interplay
between the CRE, accuracy, confidence and decision time.
We used a line-up recognition paradigm (e.g., Meissner,
Tredoux, Parker & MacLin, 2005; Dobolyi & Dodson,
2013) in which participants first encoded a series of samerace
and cross-race faces and then encountered a series of
lineups consisting of a mixture of lineups that either
contained or did not contain a different photo of a previously
seen person (i.e., target-present lineups and target-absent
lineups, respectively). Participants provided a confidence
rating for all responses, which allowed us to examine the
relationship between confidence and accuracy.
METHOD
Participants
Participants were 1656 individuals who completed the task
over the Internet via Amazon’s Mechanical Turk (www.
mturk.com). Individuals were restricted to US users, and a
check of IP addresses showed that 96.92% of participants
were located in the USA, with all states—except Wyoming
—represented by the sample. There were 1482 Caucasian-
Americans (mean age = 23.74 years, SD = 3.13, range = 18–
30, 49.83% female)1 and 174 African-Americans (mean
age = 23.54 years, SD = 3.29, range = 18–30, 64.37% female).
Design
We used a 9 (confidence-scale format) × 2 (line-up race:
black, white) × 2 (participant race: black, white) × 2 (targetpresent
vs. target-absent lineups) mixed factorial design,
with confidence-scale format and participant race as
between-subjects factors and line-up race and target
presence/absence as within-subjects factors (see later for
how we implement this design by using 12 lineups). A priori
power analyses using G*POWER (Faul, Erdfelder, Lang &
Buchner, 2007) showed that our sample size and an alpha
level of .05 provide us with over 90% power to detect
small-sized effects—using Cohen’s (1988) criteria—that
are between factors (e.g., confidence-scale format) in the
ANOVA and we will have over 99% power to detect
small-sized effects that are within factors.
Materials
We used the materials, procedure and browser-based framework
that were used in Dobolyi and Dodson (2013). There
were 12 target faces (six black and six white), with a casual
version (i.e., smiling facial expression and wearing street
clothes) and a formal version (i.e., neutral facial expression
and wearing a maroon t-shirt) of each face. The photos of
the faces were from a noncommercial database (the Meissner
Face Database). The casual photo of the person was shown at
encoding, whereas the line-up contained the formal version
of the person. Each target face was associated with two
six-person lineups: (a) a target-present line-up consisted of
the target face along with five foil faces and (b) a targetabsent
line-up consisted of six foils (i.e., aforementioned five
foils and an additional foil that replaced the target face). A
mock witness paradigm verified that all 12 lineups were fair:
that is, using the scoring method of Tredoux (1999), we
1 Technical error prevented the recording of the age of one white participant.
Confidence and eyewitness identifications 115
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
obtained average E-scores across all lineups of 4.66 (95%
CI = 4.26, 5.07) and an average target selection rate of 18%
(95% CI = 10%, 26%). See Dobolyi and Dodson (2013) for
details about the materials.
Procedure
Participants were instructed to pay careful attention to a series
of faces as their memory for them would be tested later.
They further were informed that the faces would appear one
at a time and that some of them may repeat.
During the encoding phase, a casual version of each face
was shown for 3 seconds with a 1-second interstimulus
interval. Each of the 12 target faces was presented twice in
a random order, with the following three constraints: (a) that
all 12 target faces appeared once before any repeated; (b) no
identical target face appeared consecutively; and (c) that no
more than three faces of a particular race appeared back-toback.
To counteract primacy and recency effects, the first
two faces and last two faces of the encoding phase were filler
faces that never appeared during the test phase. Overall, then,
the encoding phase contained 28 trials: the 12 target faces
seen twice, plus the four filler faces. When the encoding
phase was finished, there was a 5-minute distraction task.
Participants were informed that the test phase consisted of
a series of six-person lineups, with some lineups containing a
previously seen person and in other lineups all six faces were
never seen before. Specifically, they were told, ‘You will
now go through a series of lineups in which your goal is to
determine whether or not one of the people you saw earlier
is present in each lineup. Lineups will consist of six faces
shown together. In each lineup, only one face may correspond
to someone you saw earlier, but be aware that the
photo will not be identical. It is possible for all six people
in a lineup to be ones you have not seen earlier. Either
way, focus on just the faces: all lineup faces will be shown
with identical clothing and a neutral facial expression. If
you recognize a person in a lineup, click on that face to
highlight it. If you do not recognize any of the people, click
on the “Not Present” option.’ Each line-up consisted of a
two-row by three-column grid of faces with a ‘Not Present’
option centered underneath it.
Upon making a response, a confidence scale appeared
underneath the line-up, and participants provided a confidence
rating about the likely accuracy of their response.
Participants encountered only one of the nine different
confidence scales (described later). Each scale was presented
as a horizontal line with the anchors ‘Not at All Confident’
and ‘Completely Confident’ at the left and right endpoints,
respectively. Each scale consisted of either six or 11 discrete
points that the participant could select for their confidence
rating, in addition to a numeric slider scale that contained
101 points. The different scales were as follows:
1. Numeric, 6 points, 50–100% range: 50, 60, 70, 80, 90,
100
2. Numeric, 6 points, 0–100% range: 0, 20, 40, 60, 80, 100
3. Numeric, 11 points, 50–100% range: 50, 55, 60, 65, 70,
75, 80, 85, 90, 95, 100
4. Numeric, 11 points, 0–100% range: 0, 10, 20, 30, 40, 50,
60, 70, 80, 90, 100
5. Numeric, continuous (via a slider), 0–100% range
6. Verbal, 6 points: Not at All Confident, Quite
Unconfident, Somewhat Unconfident, Somewhat Confident,
Quite Confident, Completely Confident
7. Verbal, 6 points, but only endpoints are labeled as Not at
All Confident and Completely Confident with no labels
for the intervening points
8. Verbal, 11 points: Not at All Confident, Extremely
Unconfident, Quite Unconfident, Rather Unconfident,
Somewhat Unconfident, As Confident as Unconfident,
Somewhat Confident, Rather Confident, Quite Confident,
Extremely Confident, Completely Confident
9. Verbal, 11 points, but only endpoints are labeled as Not at
All Confident and Completely Confident with no labels
for the intervening points
Finally, each participant encountered 12 lineups during
the test phase, one for each of the target faces seen at
encoding. Three white lineups and three black lineups were
randomly assigned as target-present lineups (i.e., contained
a previously seen face), and the rest were target-absent
lineups (i.e., did not contain a previously seen face). Presentation
of the lineups was random with the constraint that
there were (a) no more than three consecutive target-present
or target-absent lineups and (b) no more than three consecutive
black or white lineups.
RESULTS
Effect sizes are measured with Cohen’s (1988) d for t-tests
and partial eta-squared (i.e., η2p
) for ANOVAs.
Accuracy and response bias
We assessed accuracy with d′ scores that were based on the
correct identification rate to target-present lineups and the
false alarm rate from target-absent lineups.2 The advantage
of d′ over other measures of accuracy, such as diagnosticity
ratios or the correct identification rate on target-present
lineups, is that d′ is less susceptible than other measures to
the confounding influence of changes in decision criteria
(e.g., the tendency to select any face in a line-up or conversely
the tendency to respond ‘not present’). A 2 (participant race:
black, white) × 2 (line-up race: black, white)× 9 (confidence
scale type) ANOVA revealed a significant interaction between
race of participant and race of line-up faces, F(1, 1638)
=32.07, MSE=1.07, p<.001, η2p
= .02, 95% CI [0.01, 0.03].
There is a significant cross-race effect for both groups of
participants: black participants showed significantly higher d′
scores for black lineups (M=2.04, SD= 1.36) than for white
lineups (M=1.51, SD=1.39), t(173) = 4.75, p<.0001,
Cohen’s d = 0.38, 95% CI [0.22, 0.54], whereas white participants
showed significantly higher d′ scores for white lineups
(M=1.61, SD= 1.27) than for black lineups (M=1.48,
2 To compute d′, we divided the false alarm rate by six to compensate for the
greater number of foils in the TA line-up as compared to the single target in
the TP line-up (e.g., Meissner et al., 2005). Moreover, because values of d′
are undefined when hit rates or false alarm rates are equal to zero, all hit rates
and false alarm rates were transformed by adding .1 to the numerator and .2
to the denominator (see Dobolyi & Dodson, 2013, for details).
116 C. S. Dodson and D. G. Dobolyi
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
SD= 1.24), t(1481)= 3.46, p = .0006, Cohen’s d = 0.10, 95%
CI [0.03, 0.18]. This interaction qualifies the significant main
effects of participant race, F(1, 1638)= 7.07, MSE=2.15,
p<.01, η2p
= .01, 95% CI [0.00, 0.01] and line-up race, F(1,
1638) = 11.69, MSE=1.07, p<.001, η2p
= .01, 95% CI [0.00,
0.02]. There were no other significant effects in this analysis;
all Fs<1.30. Overall, then, we replicated the same-race versus
cross-race effect on identification accuracy and critically
showed that this CRE occurred for both black participants
andwhite participants. In addition, we observed no statistically
significant differences in accuracy across the different confidence
rating scale formats.
As for response bias, we examined the signal detection
score, C, in which higher values correspond to a bias to
respond ‘not present’ and lower values correspond to a bias
to choose a face. A 2 (participant race: black, white) × 2
(line-up race: black, white) × 9 (confidence scale type)
ANOVA revealed an effect of line-up race, F(1, 1638)
= 4.06, MSE= 0.24, p = .04, η2p
= .00, 95% CI [0.00, 0.01]
and no other significant effects, all Fs<1.65. Participants
were significantly more biased to respond ‘not present’ to
white lineups (M= 0.86, SD= 0.57) than to black lineups
(M= 0.78, SD = 0.55), but there was no cross-race effect on
response bias. Overall, the presence of a cross-race effect
on identification accuracy and the absence of one on
response bias are consistent with Jackiw et al. (2008).
Confidence and accuracy
Absolute measure of confidence and accuracy: Calibration
We assessed the calibration of confidence to accuracy with
calibration error scores, which measure how well confidence
ratings align with accuracy. For instance, excellent calibration
between accuracy and confidence is shown when individuals
exhibit perfect, moderate and chance performance for responses
that are assigned, respectively, high (e.g., 100), medium
(e.g., 60) and low (e.g., 0) confidence ratings. In this way,
the absolute value of the confidence ratings corresponds to
the absolute level of performance for items that are assigned
these ratings. Calibration error scores were derived for each
individual by taking the absolute difference between predicted
accuracy, as indicated by the confidence rating, and actual
accuracy at each level of confidence (see Koriat & Goldsmith,
1996). This difference is then further weighted by the
frequency of responses at each level of confidence.3
Figure 1 shows a calibration plot for choosers (Figure 1a) and
nonchoosers (Figure 1b) to same-race and cross-race lineups,
collapsed across the different confidence-scale conditions.
Chooser accuracy refers to the likelihood of making a correct
identification given that a face was chosen from a line-up, and
following others (e.g., Brewer & Wells, 2006), it is computed
by the following formula: (correct identificationtarget present) /
(correct identificationtarget present+false identificationtarget absent).
By contrast, nonchooser accuracy refers to the likelihood of
correctly responding ‘not present’ when an individual
responded ‘not present’ to a line-up and is computed with the
following formula: (correct rejectiontarget absent) / (correct
rejectiontarget absent+misstarget present). In both figures, the black
diagonal line shows perfect calibration in which the lowest point
of the confidence scale is associated with chance performance,
the highest point of the scale is associated with perfect
performance and the points in between are associated with
corresponding degrees of accuracy. Greater deviation from the
black diagonal line corresponds to less accurate calibration.
Figure 1 makes two points clear: (1) although there are
significant deviations from the diagonal, confidence and
accuracy are reasonably calibrated for choosers. For
nonchoosers, however, the flat lines show that confidence
has very little correspondence with accuracy; and (2)
although the lines for same-race and cross-race performance
are similar, cross-race identification for choosers is clearly
less calibrated (i.e., farther from the diagonal than the samerace
calibration curve).
Table 1 presents the calibration scores from each
confidence-scale condition for choosers and nonchoosers to
same-race and different-race lineups. A 2 (participant race:
black, white) × 2 (line-up race: black, white) × 9 (confidence
scale type) ANOVA of the calibration error scores when
individuals choose a face showed a significant interaction
between line-up race and race of participant, F(1, 1600)
= 21.26, MSE= 0.029, p<.0001, η2p
= .01, 95% CI [0.01,
0.02]. Black participants were less calibrated with white
lineups (M= 0.37, SD = 0.22) than black lineups (M= 0.30,
SD = 0.24), t(168) = 3.50, p<.001, d = 0.31, 95% CI [0.13,
0.49], whereas white participants were less calibrated with
black lineups (M= 0.38, SD= 0.20) than white lineups
(M= 0.36, SD= 0.20), t(1448) = 3.26, p<.01, d = 0.10, 95%
CI [0.03, 0.17]. This interaction between line-up race and
participant race qualifies the significant main effects of participant
race, F(1, 1600) = 6.95, MSE= 0.051, p<.01,
η2p
= .00, 95% CI [0.00, 0.01], and line-up race, F(1, 1600)
= 6.42, MSE= 0.029, p<.01, η2p
= .00, 95% CI [0.00, 0.01].
There were no other significant effects from this ANOVA,
all Fs<1.57. Overall, then, when individuals chose a face
from a line-up, they were better calibrated when making judgments
about same-race than different-race lineups. Moreover,
as seen in Table 1, there were no statistically significant differences
between the nine different confidence rating conditions
in the calibration of confidence to chooser accuracy.
With respect to the calibration of confidence to accuracy
when individuals do not choose a face, a 2 (participant race:
black, white) × 2 (line-up race: black, white) × 9 (confidence
scale type) ANOVA produced an effect of confidence scale
type, F(8, 1407) = 2.89, MSE= 0.044, p<.01, η2p
= .02, 95%
CI [0.00, 0.02], and no other significant effects, all
3 We transformed the data from all scales to a 6-point scale that ranged from
0% to 100% so as to remove a potential confound of computing scores based
on different-sized scales. But, when participants chose an individual from a
line-up (i.e., choosers), we obtained the identical pattern of effects for all
measures when the analyses were based on either the raw (untransformed)
or transformed data. Similarly, for nonchoosers (i.e., when participants
responded ‘not present’ to a line-up), the raw data and the transformed data
produced the identical pattern of effects for Somers’ D scores, although there
were slight differences for calibration and over/underconfidence scores. Specifically,
for calibration scores, both the raw and transformed data produced
effects of confidence scale type, participant race and line-up race, except
these latter two effects were marginally significant when the data from all
the scales had been transformed to 6 points. For over/underconfidence scores,
both the raw and transformed data showed an effect of participant race, but
the transformed data showed an additional effect of confidence format (see
the text for discussion of this effect). Overall, there is a consistent pattern
of results from both the raw and transformed data; our particular transformation
did not produce the particular pattern of results.
Confidence and eyewitness identifications 117
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
Fs<3.23. Student–Newman–Keuls tests showed that when
individuals used the 50–100, 6-point confidence scale, they
were significantly less calibrated than when individuals used
any of the other scales, except for the 50–100, 11-point scale.
Similarly, the 50–100, 11-point scale was associated with
significantly worse calibration than the 0–100, 11-point
scale, the labeled, verbal 6-point scale, or the labeled, verbal
11-point scale.
Absolute measure of confidence and accuracy:
Over/underconfidence
Overconfidence and underconfidence are shown in Figure 1
by points on the calibration curve that fall below and above
the diagonal, respectively. Nearly all points of the samerace
and cross-race chooser-calibration curves fall below
the diagonal and thus reflect overconfidence. For example,
consider chooser accuracy when participants use the topmost
confidence rating (e.g., 100%). Instead of showing
100% accuracy at this top-most rating, accuracy is much
worse. In other words, participants believe that they are
performing better than they actually are. The other key
pattern is that for choosers, the cross-race curve reflects
more overconfidence than the same-race curve. The formula
for computing over/underconfidence (OU) scores is
nearly identical to that for computing calibration. However,
OU scores are based on the difference between predicted
accuracy, as indicated by the confidence rating, and actual
accuracy at each level of confidence—in contrast to
calibration that is based on the absolute value of this difference.
All other aspects of deriving OU scores are identical
to those for calibration. Positive OU scores indicate
overconfidence (i.e., the confidence rating is greater than
the corresponding level of accuracy), and negative OU
scores reflect underconfidence.
A 2 (participant race: black, white) × 2 (line-up race:
black, white) × 9 (confidence scale type) ANOVA of OU
scores when individuals choose a face yielded a significant
interaction between participant race and line-up race, F(1,
1600) = 20.93, MSE= 0.05, p<.0001, η2p
= .01, 95% CI
[0.01, 0.02]. As seen in Table 1, white participants were significantly
more overconfident when choosing a face from a
black line-up (M= 0.28, SD = 0.27) than from a white lineup
(M= 0.22, SD = 0.28), t(1448) = 6.28, p<.001, d = 0.22,
95% CI [0.15, 0.29], whereas black participants showed
the opposite pattern and were significantly more overconfident
when choosing a face from a white line-up (M= 0.27,
SD= 0.28) than from a black line-up (M= 0.21, SD = 0.29),
t(168) = 2.58, p = .01, d = 0.22, 95% CI [0.05, 0.40]. There
was also a main effect of scale type, F(8, 1600) = 3.09,
MSE= 0.10, p<.01, η2p
= .02, 95% CI [0.00, 0.02], and no
other significant effects in this analysis, all Fs<1.20. As
presented in Table 1, Student–Newman–Keuls tests
confirmed that individuals who used the 50–100, 6-point
confidence scale were significantly less overconfident than
individuals who used any of the other scales. The remaining
scales did not differ from each other.
Finally, when individuals responded ‘not present’ and did
not choose a face, a 2 (participant race: black, white) × 2
(line-up race: black, white) × 9 (confidence scale type)
ANOVA of OU scores showed a significant effect of participant
race, F(1, 1407) = 9.07, MSE= 0.15, p<.01, η2p
= .01,
95% CI [0.00, 0.02]. White participants (M=0.04,
SD= 0.35) were more underconfident than black participants
(M= 0.03, SD = 0.33), but both groups of participants were
close to the optimal value of 0, which represents the absence
of overconfidence or underconfidence. There was also a
main effect of confidence scale type, F(8, 1407) = 2.49,
MSE= 0.15, p = .01, η2p
= .01, 95% CI [0.00, 0.02], and no
other significant effects in this analysis, all Fs<1.96.
Student–Newman–Keuls tests showed that individuals who
used the 50–100, 6-point confidence scale were significantly
more underconfident than individuals who used any of the
(a) (b)
Figure 1. Calibration of confidence-level to accuracy for same-race and cross-race lineups when participants either choose a face from the
lineup (i.e., Chooser lineup) or respond “not present” (i.e., NonChooser lineup). Error bars represent 95% confidence intervals.
118 C. S. Dodson and D. G. Dobolyi
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
other scales, except they were not significantly different from
those who used the 0–100, 6-point scale. The remaining
scales were comparable to each other.
Relative measure of confidence and accuracy
Although Goodman–Kruskal gamma scores have been the
most popular method of measuring the relative confidence/
accuracy relationship, growing research demonstrates fundamental
problems with this measure (e.g., Masson & Rotello,
2008). We used Somers’ D, which is computed by a nearly
identical formula to that of gamma and does not suffer from
the same faults (e.g., Pannu & Kaszniak, 2005). Like
gamma scores, Somers’ D ranges from 1 to 1 and measures
the correlation between accuracy and confidence,
with a score of 0 indicating no relationship and scores
closer to 1 and 1 reflecting perfect positive and negative
relationships, respectively.
For choosers, a 2 (participant race: black, white)×2 (line-up
race: black, white)×9 (confidence scale type) ANOVA of
Somers’ D scores indicated no significant effects, all
Fs<1.21. There was no effect of either the type of confidence
scale or whether individuals were responding to same-race or
cross-race faces. White participants showed comparable
Somers’ D scores when responding to white lineups
(M=0.52, SD=0.58) and black lineups (M=0.56, SD=0.53).
Similarly, black participants’ Somers’ D scores were similar
for responses to white lineups (M=0.57, SD=0.49) and black
lineups (M=0.63, SD=0.49). But, it is important to highlight
that these chooser Somers’ D scores were well above 0, which
means that confidence is predictive of accuracy. As for
Table 1. Calibration and over–underconfidence scores as a function of participant race, line-up race and confidence scale format
Participant race Line-up race
Confidence scale format
Chooser Nonchooser
Calibration OU Calibration OU
Numeric/Verbal Range Points M SE M SE M SE M SE
White Black Numeric 50–100 6 0.35 0.01 0.19 0.02 0.42 0.02 0.12 0.03
Black Numeric 0–100 6 0.38 0.02 0.29 0.02 0.38 0.02 0.08 0.03
Black Numeric 50–100 11 0.39 0.02 0.26 0.02 0.42 0.02 0.10 0.03
Black Numeric 0–100 11 0.38 0.02 0.29 0.02 0.36 0.02 0.02 0.03
Black Numeric 0–100 101 0.37 0.02 0.27 0.02 0.38 0.02 0.04 0.03
Black Verbal* 6 0.39 0.02 0.28 0.02 0.38 0.02 0.03 0.03
Black Verbal 6 0.38 0.01 0.28 0.02 0.38 0.02 0.04 0.03
Black Verbal* 11 0.39 0.02 0.32 0.02 0.37 0.02 0.02 0.03
Black Verbal 11 0.38 0.01 0.29 0.02 0.35 0.02 0.01 0.03
Mean 0.38 0.01 0.28 0.01 0.38 0.01 0.05 0.01
White Numeric 50–100 6 0.33 0.01 0.14 0.02 0.42 0.02 0.14 0.03
White Numeric 0–100 6 0.37 0.02 0.24 0.02 0.36 0.02 0.07 0.03
White Numeric 50–100 11 0.35 0.02 0.23 0.02 0.39 0.02 0.03 0.03
White Numeric 0–100 11 0.36 0.02 0.26 0.02 0.33 0.02 0.03 0.03
White Numeric 0–100 101 0.35 0.02 0.23 0.02 0.33 0.02 0.03 0.03
White Verbal* 6 0.36 0.02 0.22 0.02 0.35 0.02 0.01 0.03
White Verbal 6 0.37 0.01 0.22 0.02 0.32 0.02 0.01 0.02
White Verbal* 11 0.38 0.02 0.26 0.02 0.39 0.02 0.01 0.03
White Verbal 11 0.35 0.01 0.22 0.02 0.31 0.01 0.04 0.02
Mean 0.36 0.01 0.22 0.01 0.36 0.01 0.03 0.01
Black Black Numeric 50–100 6 0.24 0.03 0.06 0.06 0.45 0.06 0.16 0.10
Black Numeric 0–100 6 0.29 0.05 0.23 0.06 0.40 0.05 0.02 0.10
Black Numeric 50–100 11 0.25 0.05 0.12 0.07 0.30 0.04 0.05 0.07
Black Numeric 0–100 11 0.28 0.06 0.21 0.07 0.31 0.04 0.10 0.07
Black Numeric 0–100 101 0.29 0.06 0.24 0.07 0.28 0.04 0.07 0.05
Black Verbal* 6 0.33 0.05 0.25 0.06 0.40 0.07 0.01 0.13
Black Verbal 6 0.26 0.05 0.14 0.06 0.33 0.03 0.05 0.07
Black Verbal* 11 0.32 0.07 0.25 0.08 0.32 0.06 0.09 0.08
Black Verbal 11 0.41 0.07 0.33 0.08 0.38 0.04 0.10 0.07
Mean 0.30 0.02 0.21 0.02 0.35 0.02 0.02 0.03
White Numeric 50–100 6 0.31 0.04 0.15 0.06 0.43 0.06 0.02 0.11
White Numeric 0–100 6 0.40 0.04 0.29 0.06 0.23 0.04 0.04 0.05
White Numeric 50–100 11 0.36 0.04 0.22 0.05 0.36 0.06 0.01 0.08
White Numeric 0–100 11 0.32 0.07 0.29 0.08 0.28 0.05 0.16 0.06
White Numeric 0–100 101 0.36 0.05 0.24 0.06 0.38 0.05 0.00 0.08
White Verbal* 6 0.44 0.05 0.40 0.06 0.36 0.06 0.07 0.10
White Verbal 6 0.36 0.04 0.24 0.07 0.28 0.04 0.04 0.07
White Verbal* 11 0.37 0.06 0.24 0.08 0.38 0.05 0.03 0.09
White Verbal 11 0.41 0.06 0.36 0.06 0.38 0.04 0.18 0.08
Mean 0.37 0.02 0.27 0.02 0.34 0.02 0.04 0.03
Note; Numeric/Verbal refers to whether the confidence ratings are labeled with numbers or words. Range refers to whether the numeric confidence scale ranges
from either 0 to 100 or 50 to 100. Points refers to whether there are 6 or 11 different ratings on the confidence scale, in addition to the numeric slider scale that
had 101 points. Verbal* indicates that these scales had none of their points labeled except for the anchors at the opposite ends of the scale. Chooser and
Nonchooser refer to line-up decisions involving either the selection of one of the line-up faces or a response of ‘not present’, respectively. Calibration and
Over–Underconfidence (OU) measure the alignment of confidence to accuracy with higher scores indicating worse performance.
Confidence and eyewitness identifications 119
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
nonchoosers, the same ANOVA of Somers’ D scores also
showed no significant effects, all Fs<1.56. Somers’ D scores
for white participants were 0.09 (SD=0.61) and 0.07
(SD=0.65) for responses to white and black lineups, respectively.
And, for black participants, Somers’ D scores were
0.08 (SD=0.57) and 0.04 (SD=0.57) for responses to white
and black lineups, respectively. Consistent with the results of
others (e.g., Sauerland & Sporer, 2009; see Sporer, Penrod,
Read & Cutler, 1995), Somers’ D scores for nonchoosers were
much less accurate than the scores for choosers and were close
to a value of 0. When participants respond ‘not present’, confidence
has nearly no predictive value for determining accuracy.
Accuracy across changes in decision time and levels of
confidence
It is impossible to use the signal detection measure of accuracy,
d′, to measure changes in accuracy across different
decision times because there are too much missing data. This
measure is based on the combination of correct identifications
and false identifications, and an individual participant
rarely makes both of these responses with the identical
decision time and with the identical level of confidence—
both of which are required if one wanted to trace changes
in d′ across different decision times and different levels of
confidence. So, instead, following what others have done,
we measured (a) the correct identification accuracy: the frequency
of identifying the target when participants encountered
a target-present line-up; and (b) the correct rejection
accuracy: the frequency of responding ‘not present’ when
the line-up did not include a previously encountered person,
i.e., in a target-absent line-up.
Figure 2a–d is derived from the mixed effects analyses
(described later) and shows how identification accuracy
(Figure 2a and b) and rejection accuracy (Figure 2c and d)
change with (a) increasing decision time, (b) different levels
of confidence and (c) same-race and cross-race faces. Note
that the error shading for each line in these figures represents
a 95% confidence interval. Consistent with previous studies
(e.g., Sauerland & Sporer, 2009), we observed that rejection
accuracy—or the rate at which individuals correctly respond
‘not present’—is relatively constant across all levels of
confidence and all decision times. In other words, neither
confidence nor decision time is much associated with the
accuracy of saying that the target is ‘not present’ in the
line-up. In addition, there appears to be no substantive difference
in rejection accuracy for same-race and cross-race
faces. But, this is not the case for identification accuracy.
The accuracy of an identification is highest for responses
that are both fast and confident (e.g., see the top two confidence
ratings). As shown in Figure 2a and b, replicating
Sauerland and Sporer (2009), fast (e.g., 2.5 seconds) and
maximally confident line-up identifications are associated
with correct identification rates of roughly 90%. Although
accuracy is better overall for same-race than cross-race identifications,
there is little effect of line-up race on the overall
shape of the curve. In other words, the CRE appears to exert
a comparable effect on identification accuracy at all confidence
levels and all decision times.
We used logistic regression fitted via generalized linear
mixed effects models in the lme4 package (Bates, Maechler,
Bolker & Walker, 2014) within R (R Core Team, 2014) to
analyze accuracy (i.e., binary correct and incorrect
responses) via a five-way interaction of the following
fixed-effect factors: Decision Time (continuous, logtransformed),
4 Confidence (continuous, 0–100), Confidence
4 We used the Box–Cox transformation method to select the best correction
for reaction time to achieve optimal normality of model residuals (Box &
Cox, 1964). For these data, log transformations were preferable to inverse
reaction times (i.e., 1/RT). Of a total of 19 872 observations (i.e., 1656 participants
* 12), 20 were removed because they were not recorded because of
technical error. We also used a cutoff of 3 median absolute deviations to trim
log-transformed outliers. This ultimately led to the removal of 294 additional
observations, or 1.48% of the raw data (or 1.58% including the 20
missing observations, i.e., 294 + 20 = 314, leaving 19 558 for the final
sample).
(a) (b)
(c) (d)
Figure 2. Identification accuracy for Chooser lineups and rejection accuracy for Nonchooser lineups as a function of confidence-level,
decision-time to make a response, and same- vs. cross-race faces in the lineup. Error shading represents 95% confidence intervals.
120 C. S. Dodson and D. G. Dobolyi
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
Scale Type (nine different scales), Line-up Decision (Identification
vs. Rejection) and Same vs. Cross-Race Line-up.
Random effects consisted of an intercept within participant
to group participants’ responses together.5 The full model
in Wilkinson–Rogers notation can be written as follows:
Accuracy ~DecisionTime * Confidence * ConfidenceScale-
Type *LineupDecision *SameVsCrossRace+(1 | Participant).
This analysis of accuracy scores produced many significant
effects and interactions, which were evaluated using
likelihood ratio tests (e.g., Barr, Levy, Scheepers & Tily,
2013). Although we will list all of the significant effects,
we only concentrate on the highest order interactions
because they moderate all lower order effects. There were
significant main effects of Decision Time, χ2(1) = 7.80,
p<.01, Confidence, χ2(1) = 1055.09, p<.0001, Line-up Decision,
χ2(1) = 1267.58, p<.0001, and Line-up Race, χ2(1)
= 17.23, p<.001, which were qualified by interactions between
(a) Decision Time and Confidence, χ2(1) = 34.95,
p<.0001, (b) Decision Time and Line-up Decision, χ2(1)
= 8.62, p<.01, (c) Confidence and Line-up Decision, χ2(1)
= 670.52, p<.0001, (d) Line-up Decision and Line-up Race,
χ2(1) = 12.15, p<.001, (e) Confidence and Confidence
Scale, χ2(8) = 22.14, p<.01, (f) Confidence Scale and
Line-up Decision, χ2(8) = 21.86, p<.01, and (g) Decision
Time, Confidence and Line-up Decision, χ2(1) = 24.19,
p<.0001.
The interaction between Decision Time, Confidence and
Line-up Decision is clearly visible in Figure 2a–d by comparing
the top two figures (Figure 2a and b), which show
the accuracy of a line-up identification, with the bottom
two figures, which show the accuracy of responding ‘not
present’—rejection accuracy (Figure 2c and d). Whereas
higher confidence and faster decision times are clearly associated
with higher accuracy of an identification (top figures),
they are not associated with accuracy of a ‘not present’ response
(bottom figures). Moreover, there is another important
part of the identification data: the top figures are clear
that it is not the case that all identification responses—at all
levels of confidence—are meaningfully related to changes
in decision time. It is primarily the two highest levels of
confidence that show a significant decrease in accuracy with
increases in decision time. For example, the accuracy of the
most confident identifications (i.e., the top line) drops nearly
in half from ~90% accuracy at the fastest decision times to
~50% accuracy at the slowest decision times. By contrast,
for the lowest three levels of confidence, there is very little
change in the accuracy of an identification across all decision
times.
The Line-up Decision and Line-up Race interaction occurred
because there was a cross-race effect on identification
accuracy (i.e., Same-Race accuracy M= 0.39, SE = 0.01 vs.
Cross-Race accuracy M= 0.33, SE = 0.01, χ2(1) = 30.60,
p<.0001) but not on rejection accuracy (i.e., Same-Race
accuracy M= 0.67, SE = 0.01 vs. Cross-Race accuracy
M= 0.66, SE = 0.01, χ2(1) = .22).
Lastly, there were two significant interactions involving
the type of confidence scale. Appendix A shows the interaction
between Confidence Scale Type and Line-up Decision,
which was produced by higher identification accuracy when
participants received the 50–100, 6-point confidence scale
than any of the other confidence-scale types. By contrast,
rejection accuracy was comparable across all confidence
scale types. We have no explanation for this difference in
identification accuracy with this particular confidence scale.
There was also an interaction between Confidence Scale
Type and Confidence Level. As seen in Appendix A, this
interaction is driven by the larger differences in overall
accuracy between the confidence levels for the (a) 0–100,
11-point scale, (b) verbal, 6-point labeled scale, and (c) the
verbal, 11-point labeled scale, as compared to the other
scales. Overall, we suspect that these interactions are mainly
driven by the high power of our design as a result of our
large sample size.
GENERAL DISCUSSION
In this study, participants encountered same-race and crossrace
faces at encoding and then completed a series of
line-up recognition tests that either contained a previously
encountered person (i.e., a different photo of the same person
seen at encoding) or did not contain a previously seen
person. Consistent with past studies, we observed (a) better
accuracy when participants responded to same-race than to
cross-race faces (e.g., Meissner & Brigham, 2001), (b) better
calibration of confidence to accuracy when participants
chose a face from a line-up (i.e., choosers) than when they
responded ‘not present’ (i.e., nonchoosers; e.g., Weber &
Brewer, 2003) and (c) that the combination of confidence
and decision time was a very powerful predictor of chooser
accuracy, with relatively fast and confident decisions associated
with roughly 90% accuracy (e.g., consistent with
Sauerland & Sporer, 2009). By contrast, confidence and
decision time were not substantial predictors of nonchooser
accuracy (e.g., Sauerland & Sporer, 2009).
There are four novel findings from this study. First, we
observed a cross-race effect on the calibration of confidence
to accuracy for choosers but not for nonchoosers. When
participants selected a face from a line-up, confidence was
more closely aligned with accuracy (e.g., high confidence =
high accuracy and low confidence = low accuracy) when
participants selected a same-race than a cross-race face.
Moreover, the reason for this CRE on calibration is that
participants were significantly more overconfident when
choosing a cross-race than a same-race face. These samerace
versus cross-race effects on calibration and on overconfidence
occurred in both our African-American and our
Caucasian-American participants. Given the importance of
eyewitness confidence on jury decision-making and the fact
that mistaken eyewitness testimony is one of the major
causes of false convictions, these results indicate that
witnesses are vulnerable to being significantly more
5 We attempted to fit multiple random effect models for multi-model selection
via Akaike Information Criterion (Burnham & Anderson, 2002) by including
additional grouping factors and slope effects consistent with
maximal models (see Barr, 2013; Barr et al., 2013; however, none of these
models successfully converged despite using multiple optimizers and
rescaling and centering of continuous variables, likely due to substantial
model complexity (e.g., the five-way interaction of the fixed effects
consisted of 2 * 2 * 9 (36) factor levels, in addition to the interaction of the
two continuous variables).
Confidence and eyewitness identifications 121
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
overconfident—perhaps causing more false convictions—
when they make cross-race than same-race identifications.
One potential underlying mechanism for this cross-race
effect on calibration and overconfidence involves the same
processes that produce a CRE on memory accuracy. Although
calibration is not necessarily tied to the accuracy of
a response (e.g., Busey, Tunnicliff, Loftus & Loftus, 2000;
Dodson, Bawa & Krueger, 2007; Palmer, Brewer, Weber
& Nagesh, 2013), it is frequently the case that conditions that
produce increased accuracy also produce increased calibration
(e.g., Deffenbacher, 2008). In our study, participants
showed a strong correlation between their d′ score (i.e., measure
of accuracy) and their chooser-calibration score: the correlation
between the same-race d′ and calibration scores was
r(1449) =.63, 95% CI [0.66, 0.60], z = 28.21, p<.0001
for our Caucasian-American participants, and it was r(169)
=.74, 95% CI [0.81, 0.67], z = 12.37, p<.0001 for
our African-American participants; for the cross-race identifications,
the correlation between the cross-race d′ and calibration
scores was r(1449) =.67, 95% CI [0.70, 0.64],
z = 30.67, p<.0001 for our Caucasian-American participants,
and it was r(169) =.69, 95% CI [0.76, 0.61],
z = 10.99, p<.0001 for our African-American participants.
Note that these correlation scores are negative because better
calibration scores are associated with lower values whereas
the opposite is true for d′ scores. Overall, for both same-race
and cross-race identifications, participants with better accuracy
showed better calibration.
Moreover, the magnitude of the CRE on memory accuracy
(i.e., worse d′ score for cross-race than same-race faces)
was related to the magnitude of the CRE on choosercalibration
scores. For Caucasian-American participants,
the correlation between their (a) change in d′ score for
same-race and cross-race faces and (b) change in choosercalibration
score for same-race and cross-race faces was
r(1449) =.57, 95% CI [0.60, 0.54], z = 24.71,
p<.0001; for African-American participants, this correlation
score was r(169) =.62, 95% CI [0.71, 0.52],
z = 7.79, p<.0001. So, the underlying mechanism that
causes a CRE on calibration is likely the same mechanism
that is producing a CRE on identification accuracy because
changes in both scores as a function of the CRE are highly
correlated. We assume that cross-race faces decrease the
kind and amount of information that is retrieved, which
reduces accuracy and in turn impairs the correspondence
between confidence and accuracy. But, more fundamentally,
the cause of the cross-race effect is not completely known.
Some researchers argue that the CRE is a case of a more
general phenomenon of in-group/out-group differences,
whereas other researchers argue for differences in perceptual
expertise to account for the CRE—and both of these
accounts may separately contribute to the CRE (see Rhodes,
2013, for a review).
The second novel finding from this study involves the
relationship between decision time, confidence and the
accuracy of an identification from chooser lineups. Although
we replicated the well-established finding that identification
accuracy is related to how long it takes participants to make
a decision (e.g., Dunning & Stern, 1994; Sporer, 1992;
Weber & Brewer, 2006), our findings place a strong
boundary condition on this relationship. The relationship between
decision time and accuracy is not the same for highly
confident, moderately confident and weakly confident identifications.
The top panels of Figure 2 clearly show that it is
only when participants are highly confident (i.e., the top
two confidence ratings) that there is a dramatic decrease in
identification accuracy from faster decisions to slower decisions
(e.g., for the highest level of confidence within the
same-race condition, there is a drop in performance from
91.63% to 50.19% for decision times of 2.77 to 41.17 seconds).
By contrast, when participants are moderately or
weakly confident in their identification, there is very little
change in accuracy from faster to slower decisions. From a
practical perspective, this means that a high confidence identification
will be much less effective at predicting accuracy
for slower responses than for faster responses, but there is little
change in accuracy for low and moderate confidence responses
across all decision times.
Theoretically, the pattern of data that needs to be
explained is why longer decision times are associated with
either dramatic or no changes in identification accuracy,
depending on the confidence in the identification (e.g., as
shown in Figure 2, increasing decision time is associated
with a large drop in the accuracy of high confidence
responses, but it is associated with no change in accuracy
for low confidence responses). To explain this differing relationship
between accuracy, confidence and decision time that
is clear in the top panels of Figure 2, we propose a Time–
Confidence–Criteria account. We argue that the criteria
about the kind and amount of memorial information that
justify a particular confidence rating change with decision
time. Longer decision times are associated with using
increasingly liberal criteria for making a response at all
confidence ratings. For example, high confidence ratings
for slower responses are based on less vivid and less plentiful
memorial information, as compared to high confidence
ratings for faster responses. Or, conversely, participants use
a stricter criterion for the kind and amount of memorial information
that is required to make a high confidence response at
faster than at slower decisions. Because identification accuracy
is likely poor when it is based on less vivid and less
plentiful memorial information, this account can easily
explain the severe loss of accuracy for high confidence
identifications with increasing decision time.
But, why is it that accuracy only shows a dramatic decline
with increasing decision time for high confidence responses
and not for low confidence responses (and only a small drop
in accuracy for moderate confidence responses) if, according
to our account, the same process of using a looser criterion
with increasing decision time also occurs for making moderate
and low confidence responses? The magnitude of the
decrease in accuracy is determined by the distance from
chance performance. Consider low confidence responses.
These are responses that are presumably based on little, if
any, diagnostic memorial information, which is why performance
is no different from chance, even for the fastest
responses. Loosening the criteria with increasing decision
time for what constitutes as a low confidence response has
little functional consequence because performance is already
at chance and performance cannot get worse than chance.
122 C. S. Dodson and D. G. Dobolyi
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
The third major finding from this study is that the crossrace
effect exerts a similar effect on the accuracy of an
identification, regardless of either the particular level of
confidence or the particular decision time. In other words,
we observed no evidence that suggests that the CRE on
identification accuracy is more powerful for either (1) low
versus high confident responses (or vice versa) or (2)
slower versus faster decision times. As mentioned earlier,
our large sample size provided us with a large amount of
power to detect even small-sized effects and so we argue
that the CRE is a phenomenon that affects all kinds of
memorial information.
Finally, the fourth significant finding involves the effect of
using different confidence scales. We replicated Weber,
Brewer and Margitich (2008) and conceptually similar findings
by Wallsten and colleagues (e.g., Wallsten et al., 1993):
(a) both our numeric and verbal confidence scales produced
comparable calibration scores and over–underconfidence
scores and (b) both numeric and verbal scales varied in the
same way when individuals either chose a face from a lineup
or responded ‘not present’. The novel aspect of our study,
however, is that no one has systematically examined so
many scales in order to answer the following questions about
measuring the relationship between confidence and accuracy:
(1) does it matter if the confidence scale consists of
few points (i.e., 6 points) or many points (i.e., 11 points or
101 points with the slider scale)?; (2) when using numeric
confidence scales, does it matter—at least for our identification
task—if the low end of the scale is fixed at 0% or 50%
(i.e., 0–100% vs. 50–100%)?; and (3) more generally, how
resistant is the confidence–accuracy relationship to changes
in the format of the confidence scale?
In terms of measuring the relationship between confidence
and accuracy, the size (i.e., number of points on the scale) and
format of the scale do not seem to matter much. Our data indicate
that participants show a strong degree of flexibility in
mapping a feeling of confidence onto a provided confidence
scale so that they are generally comparably calibrated regardless
of the format of the confidence scale. We contrasted
scales with 6 points, 11 points and 101 points (i.e., slider)
and observed no substantial differences in either the effectiveness
of the scale at measuring the calibration of confidence
to accuracy or the sensitivity of the scale to cross-race
effects. Moreover, consider the effects of using a numerical
confidence scale that ranges from either 50–100% or 0–
100%: if participants had ignored our labels for the endpoints
of the scale and, instead, had interpreted the values of the
scale literally so that, for instance, a confidence rating of
50% means a 50% chance of being correct, then participants’
CA relationship would have been much more impaired when
using the 50–100% scale than the 0–100% scale. But, this
was generally not the case. When individuals chose a face
from a line-up, we observed no difference between the 0–
100% scale and the 50–100% scale in calibration scores,
although individuals tended to be more overconfident with
the 0–100% scale than the 50–100% scale. Overall, we
suspect that when individuals understand how the confidence
scale maps onto accuracy (e.g., that the endpoints correspond
to chance and perfect performance, respectively), then the
format of the scale generally does not matter.
In conclusion, we observed that individuals tend to be
overconfident when selecting a cross-race face from a line-up
that worsens the relationship between their confidence and
accuracy of an identification for cross-race than same-race
faces. In addition, high confidence in an identification is a
tremendously powerful predictor of accuracy when the decision
is made quickly. But, this predictor is extremely sensitive
to decision time and quickly loses its predictive power with
increases in decision time. By contrast, although moderate
and weakly confident identifications are less powerful predictors
of accuracy, they also generally maintain their respective
predictive power across changes in decision time.
ACKNOWLEDGEMENT
This research was supported by National Science Foundation
Grant SES 0925145.
REFERENCES
Barr, D. J. (2013). Random effects structure for testing interactions in linear
mixed-effects models. Frontiers in Psychology, 4, 328. doi: 10.3389/
fpsyg.2013.00328
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects
structure for confirmatory hypothesis testing: Keep it maximal. Journal of
Memory and Language, 68(3), 255–278. doi: 10.1016/j.jml.2012.11.001.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Linear
mixed-effects models using Eigen and S4 (R package version 1.1-7).
http://CRAN.R-project.org/package=lme4
Box, G. E. P. & Cox, D. R. (1964). An analysis of transformations (with
discussion). Journal of the Royal Statistical Society B, 26, 211–252.
Brewer, N., Caon, A., Todd, C., & Weber, N. (2006). Eyewitness identification
accuracy and response latency. Law and Human Behavior, 30, 31–50.
Brewer, N., & Wells, G. L. (2006). The confidence–accuracy relationship in
eyewitness identification: Effects of lineup instructions, foil similarity,
and target-absent base rates. Journal of Experimental Psychology:
Applied, 12(1), 11–30. doi: 10.1037/1076-898X.12.1.11
Burnham, K. P., &Anderson,D. R. (2002).Model selection and multimodel inference:
A practical information-theoretic approach. New York: Springer.
Busey, T. A., Tunnicliff, J., Loftus, G. R., & Loftus, E. F. (2000). Accounts
of the confidence–accuracy relation in recognition memory. Psychonomic
Bulletin & Review, 7(1), 26–48.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd
ed.). Hillsdale, NJ: Lawrence Erlbaum Associates.
Corenblum, B., & Meissner, C. A. (2006). Recognition of faces of ingroup
and outgroup children and adults. Journal of Experimental Child Psychology,
93, 187–206.
Deffenbacher, K. A. (2008). Estimating the impact of estimator variables on
eyewitness identification: A fruitful marriage of practical problem solving
and psychological theorizing. Applied Cognitive Psychology, 22, 815–826.
doi: 10.1002/acp.1485
Dobolyi, D. G., & Dodson, C. S. (2013). Eyewitness confidence in simultaneous
and sequential lineups: A criterion shift account for sequential
mistaken identification overconfidence. Journal of Experimental Psychology:
Applied, 19, 345–357. doi: 10.1037/a0034596
Dodson, C. S., Bawa, S., & Krueger, L. E. (2007). Aging, metamemory, and
high-confidence errors: A misrecollection account. Psychology and
Aging, 22(1), 122–133.
Dunning, D., & Perretta, S. (2002). Automaticity and eyewitness accuracy:
A 10- to 12-second rule for distinguishing accurate from inaccurate
positive identifications. Journal of Applied Psychology, 87, 951–962.
Dunning, D., & Stern, L. B. (1994). Distinguishing accurate from inaccurate
eyewitness identifications via inquiries about decision processes. Journal
of Personality and Social Psychology, 67, 818–835.
Evans, J. R., Marcon, J. L., & Meissner, C. A. (2009). Cross-racial lineup
identification: Assessing the potential benefits of context reinstatement.
Psychology, Crime & Law, 15, 19–28.
Confidence and eyewitness identifications 123
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A
flexible statistical power analysis program for the social, behavioral,
and biomedical sciences. Behavior Research Methods, 39, 175–191.
doi: 10.3758/BF03193146
Horry, R., & Wright, D. B. (2008). I know your face but not where I saw
you: Context memory is impaired for other-race faces. Psychonomic
Bulletin & Review, 15, 610–614. doi: 10.3758/PBR.15.3.610
Horry, R., Wright, D. B., & Tredoux, C. G. (2010). Recognition and context
memory for faces from own and other ethnic groups: A remember–know investigation.
Memory & Cognition, 38, 134–141. doi: 10.3758/MC.38.2.134
Jackiw, L. B., Arbuthnott, K. D., Pfeifer, J. E., Marcon, J. L., & Meissner, C.
A. (2008). Examining the cross-race effect in lineup identification using
Caucasian and First Nation samples. Canadian Journal of Behavioral
Science, 40, 52–57.
Juslin, P., Olsson, N., & Winman, A. (1996). Calibration and diagnosticity
of confidence in eyewitness identification: Comments on what can be inferred
from the low confidence–accuracy correlation. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 22, 1304–1316.
Koriat, A. & Goldsmith, M. (1996). Monitoring and control processes in the strategic
regulation of memory accuracy. Psychological Review, 103, 490–517.
Marcon, J. L., Susa, K. J., & Meissner, C. A. (2009). An examination of the
cross-race effect in a repetition-lag paradigm. Psychonomic Bulletin and
Review, 16, 99–103.
Masson, M. E. J., & Rotello, C. M. (2008). Sources of bias in the Goodman–
Kruskal gamma coefficient measure of association: Implications for studies
of metacognitive processes. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 35, 509–527. doi: 10.1037/a0014876.
Meissner, C. A. & Brigham, J. C. (2001). Thirty years of investigating the
own-race bias in memory for faces: A meta-analytic review. Psychology,
Public Policy & Law, 7, 3–35. doi: 10.1037//1076-8971.7.1.3
Meissner, C. A., Brigham, J. C., & Butz, D. A. (2005). Memory for own- and
other-race faces: A dual-process approach. Applied Cognitive Psychology,
19, 545–567.
Meissner, C. A., Tredoux, C. G., Parker, J. F., & MacLin, O. H. (2005). Eyewitness
decisions in simultaneous and sequential lineups: A dual-process
signal detection theory analysis. Memory & Cognition, 33(5), 783–792.
doi: 10.3758/BF03193074
Palmer, M. A., Brewer, N., Weber, N., & Nagesh, A. (2013). The
confidence–accuracy relationship for eyewitness identification decisions:
Effects of exposure duration, retention interval, and divided attention.
Journal of Experimental Psychology: Applied, 19, 55–71.
Pannu, J. K., & Kaszniak, A. W. (2005). Metamemory experiments in neurological
populations: A review. Neuropsychology Review, 15, 105–130.
doi:10.1007/s11065-005-7091-6.
R Core Team (2014). R: A language and environment for statistical computing.
R Foundation for Statistical Computing, Vienna, Austria. http://
www.R-project.org/.
Rhodes, G. (2013). Face recognition. In D. Reisberg (Ed.), The Oxford
handbook of cognitive psychology (pp. 46–68). New York, NY: Oxford
University Press.
Rhodes, M. G., Sitzman, D. M., & Rowland, C. A. (2013). Monitoring and
control of learning own-race and other-race faces. Applied Cognitive Psychology,
27, 553–563. doi: 10.1002/acp.2948
Sauerland, M., & Sporer, S. L. (2007). Post-decision confidence, decision
time, and self-reported decision processes as postdictors of identification
accuracy. Psychology, Crime & Law, 13, 611–625.
Sauerland, M. & Sporer, S. L. (2009). Fast and confident: Postdicting eyewitness
identification accuracy in a field study. Journal of Experimental
Psychology: Applied, 15, 46–62.
Smith, S. M., Lindsay, R. C. L., Pryke, S., & Dysart, J. E. (2001).
Postdictors of eyewitness errors: Can false identifications be diagnosed
in the cross-race situation? Psychology, Public Policy, and Law, 7,
153–169.
Smith, S. M., Stinson, V., & Prosser, M. A. (2004). Do they all look alike?
An exploration of decision-making strategies in cross-race facial identifications.
Canadian Journal of Behavioral Science, 36, 146–154.
Sporer, S. L. (1992). Post-dicting eyewitness accuracy: Confidence,
decision-times and person descriptions of choosers and nonchoosers.
European Journal of Social Psychology, 22, 157–180.
Sporer, S. L. (1993). Eyewitness identification accuracy, confidence and
decision times in simultaneous and sequential lineups. Journal of Applied
Psychology, 78, 22–33.
Sporer, S. L. (1994). Decision-times and eyewitness identification accuracy
in simultaneous and sequential lineups. In D. F. Ross, J. D. Read, & M. P.
Toglia (Eds.), Adult eyewitness testimony: Current trends and developments
(pp. 300–327). New York: Cambridge University Press.
Sporer, S. L. (2001). Recognizing faces of other ethnic groups: An integration
of theories. Psychology, Public Policy, and Law, 7, 36–97.
Sporer, S. L., Penrod, S., Read, D., & Cutler, B. (1995). Choosing, confidence,
and accuracy: A meta-analysis of the confidence–accuracy relation
in eyewitness identification studies. Psychological Bulletin, 118,
315–327.
Tredoux, C. (1999). Statistical considerations when determining measures
of lineup size and lineup bias. Applied Cognitive Psychology, 13(S1),
S9-S26.
Wallsten, T. S., & Budescu, D. V. (1983). Encoding subjective probabilities:
A psychological and psychometric review. Management Science, 29,
151–173.
Wallsten, T. S., & Budescu, D. V. (1995). A review of human linguistic
probability processing—General-principles and empirical-evidence.
Knowledge Engineering Review, 10, 43–62.
Wallsten, T. S., Budescu, D. V., & Zwick, R. (1993). Comparing the calibration
and coherence of numerical and verbal probability judgments.
Management Science, 39, 176–190.
Weber, N., & Brewer, N. (2003). The effect of judgment type and confidence
scale on confidence–accuracy calibration in face recognition.
Journal of Applied Psychology, 88, 490–499.
Weber, N., & Brewer, N. (2006). Positive versus negative face recognition
decisions: Confidence, accuracy, and response latency. Applied Cognitive
Psychology, 20, 17–31. doi:10.1002/acp.1166.
Weber, N., Brewer, N., & Margitich, S. (2008). The confidence–accuracy
relation in eyewitness identification: Effects of verbal versus numeric
confidence scales. In K. H. Kiefer (Ed.), Applied psychology research
trends (pp. 103–118). Hauppauge, NY: Nova Science Publishers.
Weber, N., Brewer, N., Wells, G. L., Semmler, C., & Keast, A. (2004). Eyewitness
identification accuracy and response latency: The unruly 10–12
second rule. Journal of Experimental Psychology: Applied, 10, 139–147.
Wells, G. L., & Penrod, S. D. (2011). Eyewitness identification research:
Strengths and weaknesses of alternative methods. In B. Rosenfeld, &
S. D. Penrod (Eds.), Research methods in forensic psychology. Hoboken,
NJ: John Wiley and Sons.
Wright, D. B., Boyd, C. E., & Tredoux, C. G. (2001). A field study of ownrace
bias in South Africa and England. Psychology, Public Policy, and
Law, 1, 119–133.
Wright, D. B., Boyd, C. E., & Tredoux, C. G. (2003). Inter-racial contact
and the own-race bias for face recognition in South Africa and England.
Applied Cognitive Psychology, 17, 365–373.
Young, S. G., Hugenberg, K., Bernstein, M. J., & Sacco, D. F. (2012).
Perception and motivation in face recognition: A critical review of theories
of the cross-race effect. Personality and Social Psychology Review,
16, 116–142. doi: 10.1177/1088868311418987
124 C. S. Dodson and D. G. Dobolyi
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
APPENDIX A
Confidence and eyewitness identifications 125
Copyright © 2015 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 30: 113–125 (2016)
Copyright of Applied Cognitive Psychology is the property of John Wiley & Sons, Inc. and
its content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder’s