PreprintPDF Available

MansourBeaudryNguyenGroncki Preprint forDistribution

October 2024

DOI:10.13140/RG.2.2.28494.88643

Authors:

Jamal Mansour

University of Lethbridge

Jennifer Lynn Beaudry

Swinburne University of Technology

Preprints and early-stage research may not have been peer reviewed yet.

Content uploaded by Jamal Mansour

Content may be subject to copyright.

IDENTIFICATION DECISION PROCESSES

This is the pre-peer reviewed version of the following article:

Mansour, J. K., Beaudry, J. L., Nguyen, M-T. & Groncki, R. (in press). Eyewitness decision

processes: A valid reflector variable. Applied Cognitive Psychology.

This article will be published in final form at

https://onlinelibrary.wiley.com/journal/10990720. This article may be used for non-

commercial purposes in accordance with Wiley Terms and Conditions for Use of Self

Archived Versions.

IDENTIFICATION DECISION PROCESSES 1

.Eyewitness Decision Processes: A Valid Reflector Variable

Jamal K. Mansour,1 Jennifer L. Beaudry,2 Mai-Tram Nguyen3, & Roy Groncki3

1 Department of Psychology, University of Lethbridge

2 Research Development and Support, Flinders University

3 Independent scientist

Author Note

Jamal K. Mansour https://orcid.org/0000-0001-7162-8493, Jennifer L. Beaudry

https://orcid.org/0000-0003-1596-6708, jen.beaudry@flinders.edu.au, Mai-Tram Nguyen,

maitramn23@gmail.com, Roy Groncki, super_roy@hotmail.com

Much of this work occurred while Jennifer L. Beaudry, Mai-Tram Nguyen, and Roy

Groncki were with the Department of Psychological Sciences; School of Health Sciences;

Faculty of Health, Arts and Design; Swinburne University of Technology and Jamal K.

Mansour was with Psychology, Sociology, and Education; Queen Margaret University.

We have no known conflicts of interest to disclose. The videos and images used are

available upon reasonable request to first author for research use. The questions asked, data,

and analysis scripts are available on the Open Science Framework (OSF):

https://osf.io/yj8mx/?view_only=7529a7ee6d944609b5e4c9b0b67963a5

Correspondence concerning this article should be addressed to Jamal K. Mansour,

Department of Psychology, University of Lethbridge, Lethbridge, Alberta, Canada, T1K 3M4

Phone: +001 (403) 329-2077. Email: jamal.mansour@uleth.ca

This research partially fulfilled the requirements for the degree of Bachelor of Arts

in Psychology (Honours) to the third author. This research was presented at the Society for

Applied Research in Memory and Cognition, Sydney, Australia and the American

Psychology-Law Society, Seattle, Australia in 2017. This research was partially supported by

an Australian Government Research Training Program Scholarship to the fourth author.

IDENTIFICATION DECISION PROCESSES 2

Abstract

Identification accuracy can be predicted from eyewitnesses’ self-reported decision processes

but the evidence of their ability to improve prediction when confidence and response time is

mixed and minimal. Typically, decision processes are measured via one or five self-report

questions; we explored whether a more nuanced questionnaire could improve prediction.

Participants viewed a mock-crime video, made a target-present or -absent lineup decision,

and completed 17 decision process items. An exploratory factor analysis on choosers’ (n =

391) responses revealed three correlated factors, broadly reflecting Automatic Responses,

Relative Judgment, and Absolute Judgment. The three-factor solution had good internal

reliability (McDonald’s ωs = .93, .89, and .74, respectively). Scores produced from the

questions loading on the Automatic Responses and Relative Judgment improved predictions

of accuracy compared to using confidence and response time alone. Self-reported decision

processes may be an easy-to-administer and useful reflector of identification accuracy.

Keywords: eyewitness identification; decision processes; absolute judgments; relative

judgments; automatic recognition; reflector variables

IDENTIFICATION DECISION PROCESSES 3

Eyewitness Decision Processes: A Valid Reflector Variable

Eyewitness identifications are a common source of evidence in criminal justice but

evaluating their reliability is not straightforward. Triers of fact and research participants tend

to believe eyewitnesses regardless of other evidence (Boyce, Beaudry, & Lindsay, 2007).

Unfortunately, eyewitnesses sometimes make errors. Criminal justice could benefit from

tools for judging identification evidence. Measuring behaviors that covary with accuracy (i.e.,

reflector variables) can help meet this need (Wells, 2020).

The most well-established reflector variable is eyewitness confidence, partly because

jurors intuitively rely on confidence (Cutler, Penrod, & Dexter, 1990; Cutler, Penrod &

Stuve, 1988; Penrod & Cutler, 1995; Slane & Dodson, 2022). Confidence is a valid reflector

under certain circumstances (Sauer, Brewer, & Palmer, 2019; Wixted & Wells, 2017).

Eyewitness accuracy has also been reliably associated with how long an eyewitness takes to

make their decision (Brewer et al., 2006; Dunning & Peretta, 2002; Nyman et al., 2019;

Quigley-McBride & Wells, 2023; Seale-Carlisle et al., 2019; Sporer, 1992; Weber, et al.,

2004) and their decision process (Dunning & Stern, 1994; Kneller, Memon, & Stevenage,

2001; Ross et al., 2007; Smith, Stinson, & Prosser, 2004; Wittwer et al., 2022). Other

potential reflector variables are less well established: memory for fillers (Charman & Cahill,

2012), metamemory (Saraiva et al., 2020a; 2020b), response bias (Baldassari, Kantner, &

Lindsay, 2019), and (perhaps) measurements of individual differences such as face

identification ability (Bindemann et al., 2012)1. We were concerned with decision processes.

Historically, researchers have considered eyewitness decision processes in two ways:

using Wells’ (1984) absolute/relative judgment conceptualization and Dunning and Stern’s

1 Wells (2020) did not provide a formal definition but rather outlined features of reflector variables. We believe

any formal definition could be expanded to include individual differences because 1) reflector variables covary

with whether the lineup includes the culprit, and different levels of specific individual difference variables may

be associated with a higher probability of identifying a culprit versus an innocent suspect and 2) a key function

of reflector variables is to assist the criminal justice system in predicting the likelihood that an individual

eyewitness’ identification is accurate, and measuring individual differences may do so.

IDENTIFICATION DECISION PROCESSES 4

(1994) automatic recognition/deliberation conceptualization. Wells defined relative

judgement as looking for the person in a lineup who looks most like the culprit. Eyewitnesses

who do this are highly likely to make an identification (cf. reject the lineup) and will choose

based on comparisons between lineup members. In contrast, absolute judgment—wherein the

eyewitness compares each lineup member to their memory—can improve accuracy by

reducing identifications of lineup members who poorly resemble the perpetrator (even while

being the closest match to their memory). Incorrect lineup decisions are reduced when

eyewitnesses report or appear to engage more in absolute than relative judgments (e.g., Clark,

Erickson, & Brenneman, 2011; Gronlund, 2005; Lindsay & Bellinger, 1999; Lindsay et al.,

1991; Smith, Lindsay, & Pryke, 2000; Smith et al., 2001, but see Smith, Stinson, & Prosser,

2004).

Dunning and Stern (1994) aimed to differentiate accurate and inaccurate lineup

decisions. Participants answered questions about their decision process in four eyewitness

identification experiments. A principal components analysis of the responses elicited two

components, though the second was judged unnecessary. Dunning and Stern considered the

two items loading positively on their first component as reflecting automatic recognition and

the three items loading negatively on their first component as reflecting a deliberative process

normally instantiated via a process of elimination. An almost identical structure was elicited

by Robinson and Johnson (1998) using the same questions. Dunning and Stern found that

accurate identifiers endorsed more automatic recognition items, fewer process of elimination

items, were more likely to state that non-chosen photos had little influence on their decision

and that their memories had a greater influence than the photos on their decision, and were

less likely to say that the photos had a greater influence than their memories on their decision

than inaccurate eyewitnesses.

One limitation, however, is that the scope of the self-reported decision processes used

IDENTIFICATION DECISION PROCESSES 5

as reflectors of eyewitness accuracy has been limited. The absolute/relative judgment

conceptualization was never intended to be predictive2, but rather was an explanation for why

(simultaneous) lineup decisions tend to elicit high rates of innocent suspect identifications.

Although the questions created by Dunning and Stern (1994) were aimed at differentiating

correct and incorrect eyewitnesses, they comprise only five items and a single dimension.

Furthermore, self-reported decision processes will be more practically useful to

criminal justice systems if they improve one’s ability to predict accuracy beyond what

existing measures (confidence, response time) already provide. Only a few researchers have

explored this. Sauerland and Sporer (2007) found that while Dunning and Stern’s (1994)

questions predicted identification accuracy, they did not improve predictability when

confidence and response time were also considered (see also Sauerland & Sporer, 2009).

Recently, Wittwer et al. (2022) examined two different 17-item questionnaires in the

context of confidence and response time.3 Multiple factors representing decision processes

interacted with target presence in models that also included confidence and response time.

Their Automatic Recognition factor was associated with more accurate than inaccurate target-

present decisions but more inaccurate than accurate target-absent lineup decisions. Lack of

Familiarity (Experiment 1) was associated with more accurate responses to target-absent

lineups only. In Experiment 2, Process of Elimination was associated with more inaccurate

than accurate responses regardless of target presence, but the effect was larger for target-

present than -absent lineups while Familiarity-Search was associated with more accurate than

inaccurate responses to target-absent lineups but fewer accurate than inaccurate responses to

target-present lineups. Wittwer et al. did not test the three-way interaction of Chooser

(whether participants made an identification or a rejection), Presence (target presence), and

2 Strictly speaking, any measure collected after the lineup decision is a postdictor rather than a predictor of

eyewitness performance. However, for simplicity and clarity we use the term predictive.

3 We were not aware of Wittwer et al.’s (2022) work when we designed this study as we collected our data

before their work was published.

IDENTIFICATION DECISION PROCESSES 6

each decision process or conduct a separate analysis of identifiers and test for the two-way

interaction of Presence and each decision process. However, inspection of their 95%

confidence intervals suggests that Lack of Familiarity in Experiment 1 and that Familiarity-

Search and Process of Elimination in Experiment 2 differentiated accurate from inaccurate

identifications. Wittwer and colleagues therefore demonstrated that self-reported decision

processes can reflect accuracy, even when confidence and response time are considered (cf.

Sauerland & Sporer. 2007; 2009). However, more research and explicit tests of the ability of

decision processes to predict identification accuracy alongside confidence and response time

would be informative.

Even fewer studies have examined rejectors, and those that did report mixed results

(Charmin & Cahill, 2012; Kneller et al., 2001; Robinson & Johnson, 1998; Sauerland,

Sagana, & Sporer, 2012; Sauerland & Sporer, 2007; 2009). For example, Kneller et al. (2001)

found that accurate rejectors were more likely to endorse items resembling Dunning and

Stern’s (1994) automatic recognition items but found no relationship for

elimination/deliberation items. In contrast, Charman and Cahill (2012) reported that

inaccurate rejectors endorsed more deliberative items but did not differ in their endorsement

of automatic recognition items. More research is needed.

Another consideration is that eyewitness decision processes are likely more complex

than unidimensional conceptualizations (i.e., Dunning & Stern, 1994; Wells, 1978) can

account for. For example, Mansour et al. (2009) used eye tracking to monitor eyewitnesses

making lineup decisions and noted that participants appeared to use a combination of relative

and absolute judgments (see also Flowe, 2011; Flowe & Cottrell, 2010; Josephson & Holmes,

2011). Indeed, Wittwer et al. (2022) found four factors in each of their experiments.

Likewise, more nuanced decision processes have been detected when participants are asked

to think aloud while making a lineup decision (Mansour & da Costa, 2015). Finally, the

IDENTIFICATION DECISION PROCESSES 7

WITNESS model, a mathematical model of eyewitness decision processes, adequately fits

identification data when absolute/relative judgments are represented as a continuous

parameter along with response criterion (and memory strength; Clark, 2003; 2008; Clark et

al., 2011).

Yet, not all evidence points to eyewitness decision processes as multidimensional.

Fife et al. (2014) argued that their modification to the WITNESS model (WITNESS-

Restricted) could account for eyewitness decision processes by holding the absolute/relative

judgment parameter from the original model constant and relying on response criterion as the

sole decision process parameter. However, they tested their model only under circumstances

where response criterion was very liberal. When eyewitnesses are highly likely to choose to

begin with, it is unsurprising that the contribution of decision processes would be minimal.

The findings from WITNESS and WITNESSR (Clark, 2003; Fife et al., 2014)

highlight that a more advanced understanding of eyewitness decision processes may not only

benefit practice, but also theory. Consider applied lineup theory (Charman & Wells, 2007)

which implies that an eyewitness’ decision process mediates the relationship between

memory quality and identification accuracy such that when memory quality is strong, they

are more likely to use an effective decision process (cf. a counterproductive one), which in

turn increases their likelihood of being accurate. Other perspectives, such as the Diagnostic-

feature-detection theory (DFD; Wixted & Mickes, 2014) and the differential filler siphoning

hypothesis (Smith et al., 2022) may also benefit from an understanding of the

phenomenology of eyewitness decision processes by providing evidence of the extent to

which different processes occur and how these affect other strategies. If eyewitness’ decision

can be understood in terms of multiple, differentiable decision processes, these could inform

refinements to these models.

Another reason to further explore eyewitness identification decision processes is to

IDENTIFICATION DECISION PROCESSES 8

elucidate the relationship between existing conceptualizations. The terms absolute and

relative judgment are often used interchangeably with automatic recognition and process of

elimination (e.g., Harvey et al., 2020; Fife et al., 2014; Kneller et al., 2001). Yet, there is

reason to differentiate between them. Consider that automatic recognition—which is fast and

accurate (Moors, 2016)—likely involves absolute judgment, whereby only a single

comparison between a lineup member and memory is made—it just happens to be made very

quickly. In contrast, it would not make sense to say that all absolute judgments are also

examples of automatic recognition. That is, it is easy to imagine a witness comparing

multiple lineup members to their memory but doing so slowly because they do not experience

a strong feeling of recognition as they look at the lineup members. Thus, automatic

recognition is likely better understood as a special case of absolute judgment or as an extreme

point on the decision process continuum.

Likewise, we can think of the process of elimination as a type of relative judgment.

An eyewitness could make comparisons between lineup members in a variety of ways,

including engaging in a process of elimination (Dunning & Stern, 1994) or surveying the

lineup for diagnostic features (Wixted & Mickes, 2014). Notably, a process of elimination

could be conducted in an “absolute” way whereby the eyewitness compares each lineup

member to her memory and eliminates them one at a time if they do not meet their response

criterion. As such, a process of elimination could be a type of deliberative judgment that more

or less resembles an absolute or relative judgment, depending on how the eyewitness uses it.

Relative judgments may take a variety of forms.

Although Wells (1984) conceptualized them on a single continuum, absolute and

relative judgments processes may represent separately measurable dimensions. Indeed, it is

reasonable to expect that most, if not all, eyewitnesses first examine lineups for a match to

their memory. If they do not find a strong match quickly (i.e., experience automatic

IDENTIFICATION DECISION PROCESSES 9

recognition), they may engage in either an absolute or a relative judgment process.

Whichever process they engage in, they may follow up with other processes. One can

imagine an eyewitness who looks carefully at each lineup member, comparing them to their

memory (absolute judgment) and when they do not find one who sufficiently matches their

memory, begins comparing between them (relative judgment). Likewise, one could imagine

an eyewitness who compares between lineup members until they come to a decision about

who is the closest to their memory (relative judgment) and then decides whether that person

is a sufficient match to their memory to identify them as the perpetrator (absolute judgment).

With the current study, we aimed to better understand the extent to which the

eyewitness identification decision process can be understood as falling on a single continuous

dimension versus multiple correlated (or uncorrelated) dimensions. We hoped to produce a

tool for criminal justice but also to inform theory. Although our decision process questions

were aimed at identifications, those who rejected the lineup also answered these. Therefore,

we also examined their responses to add to that small literature. We hypothesized that

eyewitness decision processes would be better accounted for by multiple correlated

dimensions than a single dimension. Importantly, we also examined whether these

dimensions reflect (i.e., predict) eyewitness identification accuracy and hypothesized that

they would do so above and beyond what confidence and response time provide. Finally, the

nature of our method allowed us to directly test applied lineup theory (Charman & Wells,

2007), therefore we did so and hypothesized that we would find support for it.

Method

This study was conducted in compliance with the requirements of the ethical review

board at the second author’s university. This study was not preregistered, however our

questions asked, data, and analysis scripts are available on the Open Science Framework

(OSF): https://osf.io/yj8mx/?view_only=7529a7ee6d944609b5e4c9b0b67963a5.

IDENTIFICATION DECISION PROCESSES 10

Participants

First-year undergraduate psychology and criminology students from an Australian

university (N = 1038) participated in exchange for course credit. We removed those who did

not complete the study (n = 141), self-reported that they did not follow the mock-crime video

instructions (n = 214; i.e., they did not watch the video or watched it more than once), and

those who did not make a lineup decision (n = 7). The final sample (N = 676) ranged in age

from 18 to 68 years (M = 32.00, SD = 10.06) and was mostly female (83.43%) and Caucasian

(88.31%; 4.34% Asian, 1.18% Indigenous, and 6.21% Other).

We collected as many participants as possible while our participant pool was

available. Nunnally (1978, as cited in Boateng et al., 2018) recommended a minimum of 10

participants per question (i.e., 170 participants for a scale with 17 items). Comrey and Lee

(1992, as cited in Boateng et al.) considered a sample of 300 as good, 500 as very good, and

1000 or more as excellent. Our primary aim was to produce a scale for identifiers and our

final sample of identifiers for the scale development was 391, which is broadly considered a

good sample size for scale development.

Design

We randomly assigned participants in a 2 (Memory Strength: strong, weak) x 2

(Target Presence: present, absent) x 4 (Target: two, eight, twelve, fourteen) between-subjects

factorial design. Participants watched a mock-crime video that depicted one of four culprits

(Target). To create a strong or weak memory for the culprit, we manipulated both opportunity

to encode the mock-crime video and the retention interval between the video and lineup (see

Materials). The purpose of including Memory Strength and Target manipulations was to

increase variability in memory quality and therefore responding, which mitigates, to a degree,

stimulus sampling considerations (Wells & Windschitl, 1999). Participants attempted to

identify the culprit from a six-person photographic simultaneous lineup containing either the

IDENTIFICATION DECISION PROCESSES 11

target (target-present) or a replacement filler (target-absent).

Materials

We used Qualtrics (Provo, UT) to present the online study to participants. The videos

and lineups comprised a subset of those originally created for Mansour et al. (2020).

Mock-crime Videos

The target was recorded head-on from the shoulders up against a green screen, acting

out one of two mock-crime scenarios: discussing on the phone a plot to murder someone or

being questioned by an (off-screen) police officer after a robbery. We used four White male

targets. In the strong Memory Strength condition, the mock-crime video was 30 seconds long

and presented full-screen, with the lineup shown immediately after the video. In the weak

condition, the mock-crime video was 5 seconds long, the video was half as large as in the

strong condition, and there was a three-minute delay between the video and lineup during

which participants answered questions about a Where’s Waldo?4 image.

Lineups

The target-present lineup included the target from the mock-crime video and five

fillers who matched the general description of the target. The target-absent lineup included

the same five fillers and one additional filler. The photos (heads only) were organized in a 3 x

2 simultaneous array. The lineups were constructed using a modified match-to-description

procedure. Tredoux’s E estimates how many lineup members were considered reasonable

choices by the sample of eyewitnesses. The resultant Tredoux’s Es (Tredoux, 1998) with

95% bca confidence intervals were 3.25 [1.08, 3.31] for target two, 3.14 [1.08, 4.07] for

target eight, 2.90 [1.34, 3.17] for target twelve, and 3.16 [1.00, 4.39] for target fourteen.

Measures

Perceived Memory Strength - Manipulation Check

IDENTIFICATION DECISION PROCESSES 12

Five questions assessed participants’ perceived memory strength for the perpetrator

(attention, view, memory quality, feature visibility, confidence to identify) on 7-point scales

(e.g., 1 = very poor; 7 = very good). We also asked participants for how long they saw the

culprit’s face (seconds) and the distance (meters) between the camera and the culprit.

Lineup

Participants could select one of the six lineup members or “none of the above.” We

did not designate an innocent suspect. Qualtrics (Provo, UT) recorded the time between when

the lineup appeared and when the participant clicked a response, which took them to the next

screen. Participants rated their confidence in their decision on a scale from 0% = Not at all

confident to 100% = Very confident.

Decision Process Questions

We first asked, “How did you make your decision?” and “Approximately how long

(in seconds) did it take you to make your decision?” However, these questions were

exploratory and were not analysed. We also created two sets of 17 Likert-style decision

process items: One asked participants to predict how they would make their lineup decision

(pre-lineup questions; see the project’s OSF page) and one asked how they made their lineup

decision (post-lineup questions; see Appendix A). The pre-lineup questions were exploratory,

but we report their analysis in the supplementary materials. Finally, after explaining absolute

and relative judgements, we asked “Which type of judgment did you use more in your

decision?” (1 = Absolute to 7 = Relative).

Data Quality

Three items assessed whether participants followed the video instructions.

Participants first indicated if they had watched a video. If they reported that they had, two

open-ended questions asked how many times they watched the video and what the video

depicted. Participants were included only if they reported watching the video once.

IDENTIFICATION DECISION PROCESSES 13

Procedure

Participants consented to participate by clicking a button after reading the letter of

information. After reporting their age, sex, and ethnicity, participants were informed that they

were about to view a video (“You are about to view a video. The video you are about to see

might last for only a few seconds. Please resist the desire to watch the video more than once.

The validity of this study depends on you watching it only once.”). They then viewed a

mock-crime video. After the video, participants were informed that they were a witness to a

mock crime and would later view a lineup. Participants either advanced to the next section of

the survey (strong Memory Strength condition) or completed the Where’s Waldo task for

three minutes (weak Memory Strength condition). Participants then answered the memory

strength manipulation check questions and the pre-lineup decision process questions. All

participants next read unbiased lineup instructions, made their lineup decision, rated their

confidence in their decision, and answered the post-lineup decision process questions.

Finally, participants answered the data quality questions and were debriefed.

Results

We conducted all analyses in R (R Core Team, 2024). Packages used included boot

(Canty & Ripley, 2021; Davison & Hinkley, 1997), dplyr (Wickham et al., 2023), effectsize

(Ben-Shachar, Lüdecke, & Mackowski, 2020), ggplot2 (Wickham et al., 2016), here (Müller,

2020), Hmisc (Harrell, 2022), janitor (Firke, 2021), lavaan (Rosseel, 2012), multiUS (Aleš &

Marjan, 2023), readr (Wickham, Hester, & Bryan, 2022), psych (Revelle, 2022), r4lineuips

(Tredoux & Naylor, 2018), stringer (Wickham, 2019), and tidyr (Wickham & Girlich, 2022).

All confidence intervals are 95% confidence intervals and presented in square brackets.

Perceived Memory Strength - Manipulation Check

Multivariate analysis of variance (MANOVA) confirmed our memory strength

manipulation influenced participants’ perceived memory strength. The multivariate main

IDENTIFICATION DECISION PROCESSES 14

effect of Memory Strength was significant, Wilks’ λ = 0.59, F(7, 649) = 65.30, p < .001, ηp2 =

.41 [.36, .46] 5. The supplemental materials report the univariate results.

Lineup Decisions

First, we examined whether identification accuracy—that is, whether choosing the

culprit from a target-present lineup or a filler from a target-absent lineup—was predicted by

memory strength. We entered Memory Strength, Target Presence, and their interaction as

predictors into a probit binary logistic regression. The probit approach allows one to test

whether discriminability (per Signal Detection Theory; DeCarlo, 1998) differs as a function

of memory strength. Discriminability was significantly higher for the strong (d = 1.07) than

the weak Memory Strength condition (d = .12), z = 4.78, p < .001, as expected.

Table 1 illustrates the distribution of lineup decisions by Memory Strength in target-

present and -absent lineups. The association between lineup decisions and Memory Strength

was significant for target-present lineups, χ2(2, n = 340) = 28.08, p < .001. Participants in the

strong condition made more correct identifications than those in the weak condition, z = 4.91,

p < .001, while those in the weak condition made significantly more incorrect rejections, z =

2.21, p = .027, and filler identifications, z = 4.21, p < .001, than those in the strong condition.

There was no association between lineup response and Memory Strength for target-absent

lineups, χ2(1, n = 336) = 2.94, p = .086.

Scale Development

Identifications

We included only participants who chose someone from the lineup (n = 393) in these

analyses. Two additional participants were dropped because they did not respond to all the

scale questions. We first examined the data (n = 391) for its appropriateness for factor

analysis and found they were. Next, we used exploratory factor analysis to determine the

5 This analysis is based on 657 participants because 19 did not complete all the measures.

IDENTIFICATION DECISION PROCESSES 15

underlying structure of the 17 items used to probe eyewitnesses’ decision processes. The

supplementary materials provide the specific details of these analyses. We retained the three-

factor model as the most appropriate (see Table 2).

Table 1

Proportion of Lineup Decisions by Memory Strength for Target-present and Target-absent

Lineups

Memory

Strength

n Suspect Selection Filler Selection Rejection

Target-Present Lineups

Weak

165

.50 (.06)

.21 (.07)

.29 (.06)

Strong

175

.75 (.04)

.06 (.07)

.19 (.07)

Overall

340

.63 (.03)

.13 (.05)

.24 (.05)

Target-Absent Lineups

Weak

165

.45 (.06)

.55 (.05)

Strong

171

.35 (.06)

.65 (.04)

Overall

336

.40 (.04)

.60 (.03)

Note: Standard errors are provided in parentheses.

We named Factor 1 Automatic Response because the questions reflect a quick and

easy (effortless) decision with minimal conscious input (e.g., “I recognized the culprit

easily”). Factor 2 was named Relative Judgment because the questions suggest use of the

lineup photos to aid one’s decision via comparisons (e.g., “[The non-chosen pictures] had

some influence on my decision because more than one face had the feature/features I

remembered about the perpetrator”). Finally, we named Factor 3 Absolute Judgment because

the questions indicate conscious use of one’s memory trace (e.g., “I searched the faces for the

feature/features I remembered”).

That we found three factors is consistent with our expectation that the

automatic/deliberative and absolute/relative conceptualizations do not reflect identical

decision processes. The results are also consistent with Dunning and Stern (1994) who found

IDENTIFICATION DECISION PROCESSES 16

a single factor reflected automatic versus deliberative processing. Interestingly, our process

of elimination item was dropped because it loaded moderately on all three factors.

Table 2

Summary of the Final Factor Structure for Identifiers’ Post-lineup Decision Process Items

Item

Factor loadings

Auto

Relative

Absolute

I recognized the culprit easily.

.95

.02

-.02

.88

The culprit’s face just ‘popped out’ at me.

.94

.09

-.09

.79

To what extent did the perpetrator standout to

you in the lineup?

.89 .01 -.03 .78

How easy was it for you to make your decision?

.85

-.05

.04

.78

Rate your memory for the culprit.

.66

.05

.11

.43

[The non-chosen pictures] had little influence on

my decision. I knew straight away who to pick.*

.56 -.04 -.02 .34

[The non-chosen pictures] had some influence on

my decision because more than one face had the

feature/features I remember about the

perpetrator.

.06 1.00 -.02 .92

[The non-chosen pictures] had some influence on

my decision because more than one face seemed

familiar.

.05 .89 .02 .76

[The non-chosen pictures] confused me, and

made the task more difficult. I didn’t know

whom to pick.

-.24 .58 -.06 .54

I searched the faces for the feature/features I

remembered.

-.04 -.09 .82 .64

I compared the photos to my memory in order

to pick someone who was the closest match to

what I remembered.

-.05 .06 .67 .47

I searched the photos for a familiar face.

.004

.10

.59

.39

Unrotated sums of square loadings

4.91

1.55

1.26

Rotated sums of square loadings

4.13

2.16

1.50

Rotated proportion variance

.34

.18

.12

Factor correlations

1. Auto

2. Relative

-.55

3. Absolute

.08

.21

Note. Factor loadings > 0.30 are bolded. λ2 = communality. Auto = Automatic Recognition.

Relative = Relative Judgment. Absolute = Absolute Judgment.

IDENTIFICATION DECISION PROCESSES 17

The correlations between factors aligned with our expectations. Automatic Response

and Relative Judgement were strongly and negatively correlated (r = -.55) indicating

participants tended to endorse only one of these. Automatic recognition has often been

treated as equivalent to absolute judgment, but our Automatic Response and Absolute

Judgment factors correlated weakly (r = .08). We expected a proportion of absolute

judgments to also be automatic, so while this is lower than expected, it is in line with our

expectations. Finally, Absolute Judgment and Relative Judgment correlated moderately and

positively (r = .21), suggesting participants may engage in both processes but rely more on

one or the other, potentially as a function of other variables (such as memory strength).

Reliability. Table 3 provides internal consistency estimates for the subscales, item-

total correlations, and descriptives (DiStefano, Zhu, & Mîndrilă, 2009). All factors exceeded

the minimum acceptable value for internal consistency (≥ .70). Automatic Response met the

criterion for excellent reliability (≥ .90) and Relative Judgment met the criterion for good

reliability (≥ .80; George & Mallery, 2019; Nunnally, 1978, as cited in Petersen, 1994).

Rejections

We examined rejections, though our sample was relatively small for factor analysis (n =

283). Four participants were excluded from this analysis for not responding to all the items.

The data (n = 279) were not ideal for factor analysis but did meet the minimum requirements

for suitability (see supplemental materials). Therefore, we proceeded with the exploratory

factor analysis (see supplementary materials). Table 4 summarizes our results, which

indicated the two-factor solution was preferred.

We named the first factor Automatic Response because the items indicated a decision

made quickly and easily, despite being a rejection; the three items also appeared in the

Automatic Response factor for Identifiers. Our second factor we named Relative Judgment as

it comprised the same items as the Relative Judgment factor on our identifications scale.

IDENTIFICATION DECISION PROCESSES 18

Table 3

Internal Consistency, Item-total Correlations, and Descriptives for the Post-Lineup Identifier

Scale

Factors & Items M (SD) Item-total

Correlation

ɑ if item

deleted

Automatic Response

(ɑ = .91 95% CI [.90, .93]; ω = .93) 4.73 (1.44)

I recognised the culprit

easily

. 4.65 (1.86) .93 .88

The culprit's face just "popped out" at me. 4.82 (1.88) .88 .89

To what extent did the culprit stand out to you in

the lineup? 4.81 (1.69) .88 .89

How easy was it for you to make your decision? 4.70 (1.65) .88 .89

[The non-chosen pictures] had little influence on

my decision. I knew straight away who to pick 3.81 (1.89) .58 .93

Rate the extent to which your memory for the

culprit influenced your decision. 5.57 (1.31) .65 .92

Relative Judgment

(ɑ = .87 95% CI [.85, .89]; ω = .89) 3.35 (1.63)

[The non-chosen pictures] had some influence on

my decision because more than one face seemed

familiar.

3.63 (1.84) .86 .80

[The non-chosen pictures] had some influence on

my decision because more than one face had the

feature/features I remember about the

perpetrator.

3.55 (1.80) .91 .74

[The non-chosen pictures] confused me, and made

the task more difficult. I didn’t know whom to

pick.

2.88 (1.85) .69 .91

Absolute Judgment

(ɑ = .73 95% CI [.68, .78]; ω = .74) 5.52 (1.14)

I searched the faces for the feature/features I

remembered. 5.62 (1.31) .71 .60

I compared the photos to my memory in order to

pick someone who was the closest match to what I

remembered.

5.54 (1.46) .66 .65

I searched the photos for a familiar face. 5.41 (1.45) .60 .70

Note. ɑ = Cronbach’s alpha. ω = McDonald’s omega. In all cases, ratings ranged from 1-7.

Bolded words are shorthand terms for items.

IDENTIFICATION DECISION PROCESSES 19

Table 4

Summary of the Final Factor Structure for Rejectors’ Post-lineup Decision Process Items

Item

Factor loadings

Auto

Relative

The culprit’s face just ‘popped out’ at me.

.96

.08

.84

I recognized the culprit easily.

.94

.01

.87

To what extent did the perpetrator standout to you in

the lineup? .85 -.02 .74

[The non-chosen pictures] had some influence on my

decision because more than one face had the

feature/features I remember about the perpetrator.

.07 1.01 .94

[The non-chosen pictures] had some influence on my

decision because more than one face seemed familiar. .06 .89 .74

[The non-chosen pictures] confused me, and made the

task more difficult. I didn’t know whom to pick. -.25 .56 .53

Unrotated sums of square loadings

2.04

1.80

Rotated sums of square loadings

2.61

2.13

Rotated proportion variance

.44

.36

Factor correlations

1. Auto

2. Relative

.56

Note. Factor loadings > 0.30 are bolded. λ2 = communality. Auto = Automatic Response.

Relative = Relative Judgment.

Reliability. We next examined the internal consistency estimates for the subscales,

item-total correlations, and descriptive statistics (Table 5). Both factors met the criterion for

good reliability (≥ .80; George & Mallery, 2019; Nunnally, 1978, as cited in Petersen, 1994).

Scale Validity

Convergent Validity

As an indicator of convergent validity, we examined the correlations between

participants’ composite scores on each factor and participants’ responses to the question

IDENTIFICATION DECISION PROCESSES 20

about the extent to which they made an absolute versus relative judgment (higher values

reflected greater agreement that a relative judgment was made compared to an absolute

judgment). We first considered identifications. As expected, the absolute/relative item

correlated negatively with Automatic Response, r(389) = -.44, p < .001, and positively with

Relative Judgment, r(389) = .40, p < .001. Unexpectedly, the absolute/relative item did not

correlate with Absolute Judgement, r(389) = .06, p = .20. For rejections, neither Automatic

Response, r(276) = -.02, p = .73, nor Relative Judgment, r(276) = .12, p = .053, correlated

with the absolute/relative item, although Relative Judgment approached significance.

Table 5

Internal Consistency, Item-total Correlations, and Descriptives for the Post-lineup Rejector

Scale

Factors & Items M (SD) Item-Total

Correlation

ɑ if item

deleted

Automatic Response

(ɑ = .82 95% CI [.978, .86]; ω = .84) 1.92 (1.26)

The culprit's face just "popped out" at me. 1.81 (1.42) .82 .71

I recognised the culprit easily. 1.86 (1.41) .86 .68

To what extent did the culprit

stand out

to you in

the lineup? 2.09 (1.57) .61 .87

Relative Judgment

(ɑ = .81 95% CI [.77, .85]; ω = .84) 3.47 (1.68)

[The non-chosen pictures] had some influence on

my decision because more than one face seemed

familiar.

3.40 (1.92) .83 .69

[The non-chosen pictures] had some influence on

my decision because more than one face had the

feature/features I remember about the

perpetrator.

3.58 (1.93) .86 .65

[The non-chosen pictures] confused me, and

made the task more difficult. I didn’t know whom

to pick.

3.44 (2.06) .58 .88

Note. ɑ = Cronbach’s alpha. ω = McDonald’s omega. In all cases, ratings ranged from 1-7.

IDENTIFICATION DECISION PROCESSES 21

Bolded words are shorthand terms for items.

External Validity

We conducted additional analyses to test the external validity of the scale for

identifications (see supplementary materials). First, we considered the extent to which the

obtained factor structure generalized across our Memory Strength conditions for

identification. Confirmatory factor analysis indicated the factor structure was acceptable for

both conditions across multiple, though not all, fit indices and was a better fit to the strong

than the weak condition. Second, we considered how scores on the scale differed across

Memory Strength conditions. They differed in logical ways. Participants in the strong

condition (cf. the weak condition) scored higher on Automatic Response and lower on

Relative Judgment, though the latter depended on whether raw scores or regression factor

scores were used (with no difference when regression factor scores were used).

Predictive Validity

Identifications. We conducted logit logistic regressions to investigate whether

responses on the decision process scale were related to the accuracy of lineup selections. For

each model, we predicted identification accuracy such that suspect selections from target-

present lineups were considered accurate and all other selections were considered inaccurate.

First, we entered the composite scores from the questions comprising the three factors in our

questionnaire and found two were reliably predictive. Automatic Response, β = 0.84, SE =

0.14, z = 5.90, p < .001, OR = 1.10 [1.07, 1.14], Relative Judgment, β = -0.46, SE = 0.14, z =

3.36, p = .001, OR = 0.91 [0.86, 0.96], and Absolute Judgment, β = 0.18, SE = 0.12, z = 1.47,

p = .14, OR = 1.05 [0.98, 1.13].

Our scale will be more useful if it provides predictive power over and above that

provided by confidence and response time. We first confirmed these variables predicted

identification accuracy. Both confidence, β = 0.93, SE = 0.13, z = 7.14, p < .001, OR = 1.04

IDENTIFICATION DECISION PROCESSES 22

CI [1.03, 1.05], and response time, β = -0.27, SE = 0.12, z = 2.19, p = .028, OR = 0.98 CI

[0.96, 0.998], were reliable predictors. We then tested whether a model with Confidence,

response time, Automatic Response, Relative Judgment, and Absolute Judgment predicted

identification accuracy. Compared to the confidence and response time only model (AIC =

463.65), the current model (AIC = 442.51) was a significantly better fit to the data, χ2 (3) =

27.14, p < .001. This indicates that self-reported decision processes are a useful reflector

variable. Within the current model, Relative Judgment, β = -0.48, SE = 0.14, z = 3.41, p =

.001, OR = 0.90 [0.86, 0.96], and confidence, β = 0.48, SE = 0.17, z = 2.82, p = .005, OR =

1.02 [1.01, 1.03], were the strongest predictors, followed by Automatic Response, β = 0.42,

SE = 0.19, z = 2.22, p = .026, OR = 1.05 [1.00, 1.10]. Neither Absolute Judgement, β = 0.19,

SE = 0.12, z = 1.52, p = .13, OR = 1.06 [0.98, 1.13], nor response time, β = -0.21, SE = 0.13, z

= 1.59, p = .11, OR = 0.98 [0.96, 1.00], were significant predictors in the current model.

Given that Absolute Judgement was not a significant predictor, we ran a fourth model,

removing Absolute Judgement as a predictor. Response time was still not a significant

predictor (p = .13), but the significance and relative priority of the remaining predictors was

maintained. The fourth model was not a significantly better fit to the data than the third

model, χ2 (1) = 2.29, p = .13, although it is more parsimonious and therefore preferable.

These data support our expectation that a more nuanced self-report questionnaire about

eyewitness identification decision processes can improve our ability to predict eyewitness

identification accuracy.6

Rejections. As with identifications, we produced a composite score for each factor

6 Best practice is to use factor scores rather than a summed composite score; however, we wanted to produce a

tool that was as easy as possible for practitioners to use. Importantly, when we repeated the analyses in this

section using factor scores, there was no difference in our conclusions and the variance accounted for by the

final model was identical as measured by pseudo-R2 measures (Cox & Snell = .24; Nagelkerke = .32). However,

the relative priority of the predictors changed such that Automatic Recognition and Relative Judgments were

stronger predictors (βs = 0.80) than confidence (β = .49). Furthermore, we drew identical conclusions when we

produced factors with different oblique rotations (Quartimin and Oblimin).

IDENTIFICATION DECISION PROCESSES 23

directly from the scale responses. In a logit logistic regression, neither Automatic Response (p

= .35) nor Relative Judgment (p = .70) were significant predictors of rejection accuracy. This

was also true using regression factor scores as predictors rather than the composite scores.

Testing Applied Lineup Theory

Given that we successfully manipulated memory strength and obtained reliable

measures of decision processes, we examined whether our data supported Applied Lineup

Theory (Charman & Wells, 2007). Specifically, we examined whether decision processes—

operationalized by participants’ composite scores on the three factors derived from our factor

analysis—mediated the relationship between memory strength and identification accuracy.

We first constructed a model with our manipulation of memory strength as a predictor

of identification accuracy. As expected given our prior analyses, memory strength was a

reliable predictor, β = 0.45, SE = 0.10, z = 4.32, p < .001 OR = 2.46 [1.64, 3.71]. Next, we

examined whether Memory Strength predicted each of the three decision processes using

three separate models. The models for Automatic Response, β = 1.69, SE = 0.43, z = 3.92, p <

.001, OR = 29.08 [5.40, 156.48], and Relative Judgment, β = -0.49, SE = 0.25, z = 1.99, p =

.047, OR = 0.37 [0.14, 0.98], were significant, but the model for Absolute Judgment was not,

β = 0.11, SE = 0.17, z = 0.65, p = .52, OR = 1.25 [0.64, 2.46]. As such, we next tested

whether Automatic Response and Relative Judgment mediated the relationship between

Memory Strength and identification accuracy by constructing models with both Memory

Strength and the decision process as predictors of identification accuracy.

IDENTIFICATION DECISION PROCESSES 24

Figure 1

Path Models Predicting Identification Accuracy from Memory Strength and Decision

Processes

Note: path coefficients are standardized.

In the two-predictor model with Automatic Response, both Automatic Response, β = 1.02, SE

= 0.13, z = 7.76, p < .001, OR = 1.12 [1.09, 1.16], and Memory Strength, β = 0.34, SE = 0.12,

z = 2.94, p = .003, OR = 1.97 [1.25, 3.10], were significant. Notably, the effect of Memory

Strength (β = 0.34) was smaller than in the model with only Memory Strength (β = 0.45),

indicating partial mediation. The effect was tested by producing a 95% confidence interval

via bootstrapping with 1000 bootstrapped samples. The indirect effect of Memory Strength

Memory strength Ide nﬁcaon accuracy

Relave Judgment

a* = -0.10 b* = -0.34

c’ = 0.19

c = 0.22

Memory strength Ide nﬁcaon accuracy

Absolute Judgment

a* = 0.03 b* = 0.03

c’ = 0.22

c = 0.22

IDENTIFICATION DECISION PROCESSES 25

through Automatic Response was .08 [.04, .0.13], which is 36% of the overall effect of

Memory Strength. The 95% confidence interval did not include zero indicating the partial

mediation was reliably different from zero. Figure 1A depicts the mediation path model.

Both Relative Judgment, β = -0.78, SE = 0.12, z = 6.57, p < .001, OR = 0.85 [0.81,

0.89], and Memory Strength, β = 0.43, SE = 0.11, z = 3.86, p < .001, OR = 2.36 [1.53, 3.67],

were significant in the two-predictor mediation model. The regression coefficient for

Memory Strength was slightly lower than in the Memory Strength only model, again

suggesting partial mediation. The indirect effect of Memory Strength through Relative

Judgment was .03 [0, .07], which is 14% of the total effect—smaller than for Automatic

Response. Bootstrapping for the 95% confidence interval indicated that the effect did not

reach significance as it included zero. Figure 1B depicts this path model with standardized

path coefficients. Figure 1C depicts the path model for Absolute Judgement for completeness.

In summary, our results are consistent with applied lineup theory (Charman & Wells,

2007) regarding how memory strength, decision processes, and identification accuracy relate.

Discussion

Our goal was to develop a self-report decision process questionnaire more nuanced

than Dunning and Stern’s (1994) or asking eyewitnesses whether they made a relative versus

absolute judgment (e.g., Smith et al., 2000; 2001), and in doing so, enhance the ability of

decision processes to predict identification accuracy beyond what confidence and response

time already provide. We also hoped to inform theory: whether a unidimensional or

multidimensional approach to decision processes is more appropriate and the extent to which

applied lineup theory (Charman & Wells, 2007) can account for identification performance.

A secondary goal was to examine the decision processes of eyewitnesses who reject a lineup.

Critically, we demonstrated that self-reports of decision processes (Automatic

Response, Relative Judgment) explained variability in eyewitness identification decisions that

IDENTIFICATION DECISION PROCESSES 26

was not accounted for when only confidence and response time were used. This indicates that

administering our scale to eyewitnesses can improve predictions of whether the eyewitness is

accurate. Thus, consistent with Wittwer et al. (2022), we find that self-reports of decision

processes are a valid reflector variable.

Eyewitness decision processes were best accounted for using multiple, correlated

factors, consistent with our expectations. Building on Dunning and Stern (1994), we found

one dimension represented automatic responses, and—in the case of identifications—two

dimensions represented the type of deliberation used (only one deliberative dimension was

found for rejections). The nature of these dimensions reflected Wells’ (1984)

conceptualizations of how people approach lineups—using primarily relative or absolute

judgment.

Although researchers have often thought of eyewitnesses as using one or the other of

absolute versus relative judgment, research suggests eyewitnesses may use a combination

(e.g., Mansour et al., 2009). Our results support this conclusion: our Relative Judgment and

Absolute Judgment factors were positively correlated. Moreover, Relative Judgment was

strongly and negatively correlated with Automatic Response, as would be expected.

Surprisingly though, the correlation between Absolute Judgment and Automatic Response

was negligible. This indicates that automatic recognition and absolute judgments are not

interchangeable and that eyewitness decision processes are not best reflected by a single

continuous dimension, as suggested by prior research (e.g., Dunning & Stern, 1994; Fife et

al., 1994) and theory (e.g., Charman & Wells, 2007). Indeed, a one-factor solution was an

inadequate fit to the identification data.

Our Automatic Response factor seemed to reflect a quick and effortless process, like

Dunning and Stern’s (1994) automatic recognition and Wittwer et al.’s (2022) Automatic

factor. The questions associated with this factor asked the participant to indicate the extent to

IDENTIFICATION DECISION PROCESSES 27

which a face popped out or stood out for them, they found the decision easy, and relied on

their memory. High ratings on these questions align well with conceptions of automatic

responses as being uncontrolled (i.e., occurring regardless of the task one is asked to do),

unconscious (i.e., one need not think to produce the response), efficient (i.e., requiring little

or no attention), and fast (Moors, 2016). That is, our questions about whether the face popped

out or stood out get at the response being uncontrolled and fast while our questions about the

ease with which the identification was made reflect unconsciousness and efficiency. The

question about memory reflects the mechanism that leads to automatic responses—repeated

or good quality exposure. Consistent with the notion that automatic responses are fast, the

inclusion of the Automatic Response factor alongside confidence and response time as

predictors of identification accuracy noticeably reduced the power of these factors to predict

accuracy, based on standardized betas. The contribution of response time became non-

significant and the contribution of confidence became less than the contribution of Relative

Judgment. In contrast, Automatic Response accounted for the most variance when only the

decision process composite scores were used as predictors. It is notable that Automatic

Response is the factor on which the largest number of items from our questionnaire loaded,

even though its unique contribution may be more modest. Future research examining more

items related to other decision processes may produce a better questionnaire.

We expected an automatic response to be negatively correlated or uncorrelated with

counterproductive or less efficacious deliberative processes. This was true in our data. The

Automatic Response factor was negatively and strongly correlated with the Relative Judgment

factor, which was associated with lower identification accuracy. However, it is perhaps

surprising that the correlation between Automatic Response and Absolute Judgment was as

weak as it was (r = .08). Earlier we noted that we do not consider automatic recognition and

absolute judgments as equivalent, but we do think that automatic recognition can be

IDENTIFICATION DECISION PROCESSES 28

considered a special case of absolute judgment. We expected a moderate correlation between

those processes but instead found a small positive correlation between Absolute Judgment

and Relative Judgment (r = .21). This suggests a tendency to sometimes engage in relative

judgments even when one can and does engage in absolute judgments.

Another reason for the low correlation between Automatic Response and Absolute

Judgments may be related to our identification procedure. We used simultaneous lineups in

this experiment; it is reasonable to expect that placing all lineup members side by side

encourages relative judgments even when they may be unnecessary. We can imagine that the

correlation may have been stronger had we used sequential lineups—and the correlation

between Absolute Judgments and Relative Judgments weaker. Future research should

consider the generalizability of our scale across identification procedures.

A third possibility is that what we should rethink our conceptualization of Factor 3.

We were surprised to find that identifiers’ scores for our Absolute Judgment factor did not

correlate with our Likert-style absolute/relative judgment question, although the question did

correlate with our Automatic Response and Relative Judgment factors. Although each

question that loaded on the Absolute Judgement factor implied a comparison between the

eyewitness’ memory and the lineup members (“I searched the faces for the feature/features I

remembered”, “I compared the photos to my memory”, “I searched for a familiar face”), this

factor may actally tap into participants’ perception of their memory strength. Yet, if that was

the case, we would expect the Absolute Judgment factor to predict identification accuracy and

to be related to memory strength, neither of which was the case, as demonstrated in our

consideration of applied lineup theory (Charman and Wells, 2007).

On the other hand, it seems logical that eyewitnesses who do not experience

automatic recognition will engage in a more extended absolute judgment process, that is a

search for a match to memory. Regardless, if an absolute judgment strategy fails,

IDENTIFICATION DECISION PROCESSES 29

eyewitnesses may turn to a relative judgment process—or they may start with a relative

judgment process and finish with an absolute judgment process. Thus, we might expect

accurate and inaccurate eyewitnesses to engage in absolute judgment. To that end, our third

factor may reflect a search that occurs regardless of memory strength and which is unrelated

to accuracy. Still, this explanation is unsatisfactory because we found a small positive

correlation between Automatic Recognition and Absolute Judgement. We would expect a

negative correlation in this case because eyewitnesses who experience automatic recognition

would be expected to engage in no or a very short search. We suggest that future researchers

consider asking questions about the temporal nature of the eyewitness’ decision process to

tease apart whether the eyewitness started with an absolute judgment process and then

followed up with a relative judgment process or vice versa.

To what extent do our items align with prior efforts to produce a decision process

questionnaire? Many of our items derived from Dunning and Stern’s (1994) seminal work

and our final set of items incorporated items that resonated with four of their five items.

Although we had an item that asked about the extent to which participants used an

elimination strategy, this item did not survive the exploratory factor analysis. This was

surprising, however, the item may have been double-barreled (See Appendix 1) which may

have reduced its utility. Nonetheless, our items were broadly consistent with Dunning and

Stern’s work.

Wittwer et al. (2022) also found evidence that people may engage in multiple types of

judgments. Like us they found an Automatic factor which was related to accuracy (though not

identification accuracy in their case7). That some eyewitnesses experience automatic

recognition and that this is relevant to performance appears ubiquitous. In both studies they

7 In two studies, Wittwer et al. (2022) reported a significant interaction between their Automatic Recognition

factor and target presence such that scores on the Automatic Recognition factor were higher for (correct)

identifications from compared to any other target-present decision but lower for (filler) identifications compared

to correct rejections from target-absent lineups.

IDENTIFICATION DECISION PROCESSES 30

labelled a factor Elimination and the items loading on this factor resonate with our Relative

Judgment factor. As argued earlier, an elimination process can be considered a type of

relative judgment and the questions on this factor were consistent with that notion (e.g., “I

compared the lineup members to each other, especially their facial features, looking for ways

in which they were dissimilar. This helped me make my decision.”). While we labelled our

remaining factor as Absolute Judgment, they labelled their remaining factors as Lack of

Familiarity and Difficulty Feeling (Experiment 1) and Familiarity-Search (Experiment 2;

their fourth factor labelled Facial Matching Detail was not considered sufficiently reliable to

maintain). The items on their “familiarity” factors resonate with the items on our Absolute

Judgment factor as they point to examination of features (e.g., “There were some features that

I remembered about the perpetrator that were missing from all the lineup members’ faces.”)

and the impact of non-chosen faces (e.g., “When I inspected the photographs, I immediately

noticed that no face match the culprit.”). Interestingly, their “familiarity” factors tended to be

associated with accuracy while their Elimination factor was not. This contrasts with our

finding that Relative Judgment but not Absolute Judgment was predictive. However, because

our scale and Wittwer et al.’s used different items, it is impossible to draw firm conclusions.

How can we contextualize our findings in relation to theory? First, these results are

contrary to existing conceptualizations of the lineup decision process as comprising a single

dimension (e.g., Charman & Wells, 2007). We found that multiple, correlated dimensions

accounted for our participants’ characterizations of their lineup decision, whether it was

identification or a rejection. Moreover, our Automatic Response factor was separable from

our Absolute Judgment factor, suggesting that automatic recognition and absolute judgments

are not interchangeable. Their very low correlation suggests they may reflect quite different

processes—or at least that people perceive them differently. That (for identifications) our

Relative Judgment factor correlated with the Absolute Judgment factor supports previous

IDENTIFICATION DECISION PROCESSES 31

findings and speculations that eyewitnesses engage in a mix of absolute and relative

judgments (e.g., Mansour et al., 2009) but adds to our understanding by highlighting their

relationship with automatic recognition.

The fact that we manipulated memory strength and measured both decision processes

and identification accuracy provided an opportunity to test a key tenet of applied lineup

theory (Charman & Wells, 2007). The theory suggests that decision processes mediate or

partially mediate the relationship between memory strength and identification accuracy. In

the current study, we found that memory strength and decision processes were strongly

related to identification accuracy, therefore we tested a mediation model for the decision

process factors we found. Automatic Response was a significant partial mediator, providing

support for applied lineup theory. As shown in Figure 1A, a stronger memory was associated

with greater endorsement of Automatic Response, which in turn increased the likelihood of

identification accuracy. Because mediation was partial, memory strength also directly

affected the likelihood of accuracy.

As a secondary question, we examined decision processes by those who rejected the

lineup. We were interested to find that a two-factor solution explained significant variance

(79%) in rejectors’ responses to the items and that the factors produced mirrored the factors

produced for identifiers: Automatic Response and Relative Judgment. Yet, these factors were

not predictive of rejection accuracy. Given that our items were constructed with identifiers in

mind, we are reluctant to make much of this result. Future research should take a more

systematic approach to producing items relevant to rejectors. Indeed, other research has

suggested that rejections may involve rather different decision processes from identifications

(Lindsay et al., 2013; Sporer, 1992).

Limitations

A few limitations are worth considering. Although we used a variety of mock-crime

IDENTIFICATION DECISION PROCESSES 32

videos, they were similar in content—with a strong and obvious focus on the culprit. If our

scale is to be useful, it will need to be shown to be effective across a broader range of

memory qualities. Similarly, our targets were all White males and our participants were

primarily White, so it is unlikely that many of our participants experienced a cross-race

effect, which may significantly influence one’s decision process. Another limitation is that, in

constructing our target-absent lineups, we simply replaced the culprit with another filler. It is

possible that this influenced our identification results such that the specific filler chosen may

have been a particularly good or poor filler, and therefore affected how the participants made

their decision. However, as we did not designate a specific innocent suspect for our analysis,

this concern is mitigated to some extent (i.e., the decision processes considered were for all

target-absent lineup identifications). Encouragingly, the rate of filler identifications and

correct rejections from our target-absent lineups almost perfectly align with the mean rates

reported by Steblay et al. (2011) in their meta-analysis (Steblay et al.: 43% filler

identifications, 57% correct rejections; Us: 40% filler identifications, 60% correct rejections).

A final limitation is that prior to the lineup decision, participants were asked to predict how

they would make their lineup decision using modified versions of the questions they were

asked after they made their lineup decision. Doing so may have affected their decision

process or responses to the subsequent decision process questions.

Implications

We have directly demonstrated that (12) self-report items improve the ability to

predict identification accuracy when confidence and response time are also considered,

providing further evidence that asking participants to self-report about their decision

processes can enhance predictions of eyewitness accuracy. Given how ubiquitous

identification evidence is, how difficult errors are to detect, and that many police jurisdictions

to not ask eyewitnesses to judge their confidence, more reflectors of eyewitness accuracy are

IDENTIFICATION DECISION PROCESSES 33

needed (Wells, 2020). Self-reports, which would require minimal resources to administer,

deserve further attention as reflectors of eyewitness identification accuracy. People are

relatively familiar with answering Likert-style questions. This short questionnaire can be

completed within a minute or two and, if needed, officers could even read out the questions

to people with literacy challenges. The use of scales as evidence is already commonplace

within criminal justice (e.g., for risk assessments, competency assessments) and therefore

there may be less resistance to their use than other tools. For example, many jurisdictions are

reluctant to ask eyewitnesses how confident they are because any response other than 100%

suggests the possibility that the eyewitness is inaccurate. In contrast, questions about how the

eyewitness made their decision may be less controversial because no single response

obviously indicates inaccuracy.

After sufficient replication and evidence of generalizability, this questionnaire could

be completed by eyewitnesses immediately after they express their confidence and provided

as part of the evidence presented in court. We could see this tool being used by investigators

to help them judge how confident they should be about an identification. If the eyewitness’

score on the scale (in combination with their confidence if possible) suggests a moderate or

low probability of accuracy, we would hope they would be more inclined to investigate

alternative suspects. Prosecutors might use this tool to inform their considerations of plea

deals. Lawyers might use this tool to better inform fact finders about how to weight

identification evidence. Of course, like any tool, this tool would have to be used carefully to

ensure it is not abused. It is certainly plausible that an eyewitness could be coached about

how to respond, which would mitigate the tool’s effectiveness. However, such concerns

should not be new to the criminal justice system: many assessment tools used regularly in

criminal justice can be abused similarly. It is important that best practices for administering

such a tool be studied and adhered to.

IDENTIFICATION DECISION PROCESSES 34

Conclusions

To summarize, the factors we found and the pattern of correlations amongst our

factors highlight the multidimensional nature of eyewitness identifications—which has

implications for theories of eyewitness decision processes, and we found support for applied

lineup theory (Charman & Wells, 2007). We found evidence that an Automatic Response

factor and two deliberative decision process factors: Relative Judgment and Absolute

Judgment reflected the decision processes of participants who made a lineup identification

and all three factors had adequate internal reliability. Importantly, scores based on these

factors not only predicted identification accuracy, but improved predictions of accuracy

compared to using only confidence and response time. Although validation and replication

are clearly necessary, it appears that eyewitness’ self-reports of their decision processes can

be an effective reflector variable—one that can easily be incorporated into extant

identification procedures—and that provides predictive power over and above confidence and

response time.

IDENTIFICATION DECISION PROCESSES 35

References

Aleš, Ž., & Marjan, C. (2023). multiUS: Functions for the courses multivariate analysis and

computer intensive methods (Version 1.2.3) [R package]. https://CRAN.R-

project.org/package=multiUS

Baldassari, M. J., Kantner, J., & Lindsay, D. S. (2019). The importance of decision bias for

predicting eyewitness lineup choices: toward a Lineup Skills Test. Cognitive

Research: Principles and Implications, 4, 1-13. https://doi.org/10.1186/s41235-018-

0150-3

Ben-Shachar, M., Lüdecke, D., & Makowski, D (2020). effectsize: Estimation of effect size

indices and standardized parameters. Journal of Open Source Software, 5(56), 2815.

https://doi.org/10.21105/joss.02815

Bindemann, M., Brown, C., Koyas, T., & Russ, A. (2012). Individual differences in face

identification postdict eyewitness accuracy. Journal of Applied Research in Memory

and Cognition, 1(2), 96-103. https://doi.org/10.1016/j.jarmac.2012.02.001

Boateng, G. O., Neilands, T. B., Frongillo, E. A., Melgar-Quiñonez, H. R., & Young, S. L.

(2018). Best Practices for Developing and Validating Scales for Health, Social, and

Behavioral Research: A Primer. Frontiers in Public Health, 6, 149.

https://doi.org/10.3389/fpubh.2018.00149

Boyce, M., Beaudry, J., & Lindsay, R. C. L. (2007). Belief of eyewitness identification

evidence. In The handbook of eyewitness psychology: Volume II (pp. 515-540).

Psychology Press.

Brewer, N., Caon, A., Todd, C., & Weber, N. (2006). Eyewitness identification accuracy and

response latency. Law and Human Behavior, 30(1), 31–50.

https://doi.org/10.1007/s10979-006-9002-7

Canty, A., & Ripley, B. (2021). boot: Bootstrap R (S-Plus) functions (Version 1.3-28) [R

IDENTIFICATION DECISION PROCESSES 36

package]. https://CRAN.R-project.org/package=boot

Charman, S. D., & Cahill, B. S. (2012). Witnesses’ memories for lineup fillers postdicts their

identification accuracy. Journal of Applied Research in Memory and Cognition, 1(1),

11-17. https://doi.org/10.1016/j.jarmac.2011.08.001

Charman, S., & Wells, G. L. (2007). Applied lineup theory. In R. C. L. Lindsay, D. F. Ross,

J. Read, M. P. Toglia (Eds.). The handbook of eyewitness psychology, Vol II: Memory

for people (pp. 219–254). Lawrence Erlbaum Associates, Publishers.

Clark, S. E. (2003). A memory and decision model for eyewitness identification. Applied

Cognitive Psychology, 17(6), 629-654. https://doi.org/10.1002/acp.891

Clark, S. E., Erickson, M. A., & Breneman, J. (2011). Probative value of absolute and

relative judgments in eyewitness identification. Law and Human Behavior, 35, 364–

380. https://doi.org/10.1007/s10979-010-9245-1

Cutler, B. L., Penrod, S. D., & Dexter, H. R. (1990). Juror sensitivity to eyewitness

identification evidence. Law and Human Behavior, 14(2), 185-191.

https://doi.org/10.1007/BF01062972

Cutler, B. L., Penrod, S. D., & Stuve, T. E. (1988). Juror decision making in eyewitness

identification cases. Law and Human Behavior, 12(1), 41-55.

https://doi.org/10.1007/BF01064273

Davison, A. C., & Hinkley, D. V. (1997) Bootstrap Methods and Their Applications.

Cambridge.

DeCarlo, L. T. (1998). Signal detection theory and generalized linear models. Psychological

Methods, 3(2), 186–205. https://doi.org/10.1037/1082-989X.3.2.186

Dunning, D., & Stern, L. B. (1994). Distinguishing accurate from inaccurate eyewitness

identifications via inquiries about decision processes. Journal of Personality and Social

Psychology, 67(5), 818–835. https://doi.org/10.1037/0022-3514.67.5.818

IDENTIFICATION DECISION PROCESSES 37

Fife, D., Perry, C., & Gronlund, S. D. (2014). Revisiting absolute and relative judgments in

the WITNESS model. Psychonomic Bulletin Review, 21, 479–487.

https://doi.org/10.3758/s13423-013-0493-1

Firke, S. (2021). janitor: Simple tools for examining and cleaning dirty data (Version 2.1.0)

[R package]. https://CRAN.R-project.org/package=janitor

Flowe, H. (2011). An exploration of visual behaviour in eyewitness identification tests.

Applied Cognitive Psychology, 25(2), 244-254. https://doi.org/10.1002/acp.1670

Flowe, H., & Cottrell, G. W. (2011). An examination of simultaneous lineup identification

decision processes using eye tracking. Applied Cognitive Psychology, 25(3), 443–451.

https://doi.org/10.1002/acp.1711

George, D., & Mallery, P. (2019). IBM SPSS statistics 25 step by step: A simple guide and

reference. Taylor & Francis.

Gronlund, S. D. (2005). Sequential lineup advantage: Contributions of distinctiveness and

recollection. Applied Cognitive Psychology, 19(1), 23-37.

https://doi.org/10.1002/acp.1047

Harrell Jr, F. (2022). Hmisc: Harrell miscellaneous (Version 4.7-0) [R package].

https://CRAN.R-project.org/package=Hmisc

Harvey, A. J., Shrimpton, B., Azzopardi, Z., O'Grady, K., Hicks, E., Hirst, E., & Atkinson-

Cox, K. (2020). The influence of alcohol and weapon presence on eyewitness memory

and confidence. Applied Cognitive Psychology, 34(2), 489-503.

https://doi.org/10.1002/acp.3636

Josephson, S., & Holmes, M. E. (2011). Selecting the suspect: An eye-tracking comparison

of viewing of same-race vs. cross-race photographs in eyewitness identification. Visual

Communication Quarterly, 18(4), 236–249.

http://doi.org/10.1080/15551393.2011.627280

IDENTIFICATION DECISION PROCESSES 38

Kneller, W., Memon, A., & Stevenage, S. (2001). Simultaneous and sequential lineups:

Decision processes of accurate and inaccurate eyewitnesses. Applied Cognitive

Psychology, 15(6), 659-671. https://doi.org/10.1002/acp.739

Lindsay, R. C. L., & Bellinger, K. (1999). Alternatives to the sequential lineup: the

importance of controlling the pictures. The Journal of Applied Psychology, 84(3), 315–

321. https://doi.org/10.1037/0021-9010.84.3.315

Lindsay, R. C. L., Kalmet, N., Leung, J., Bertrand, M., Sauer, J. D., & Sauerland, M. (2013).

Confidence and accuracy of lineup selections and rejections: Postdicting rejection

accuracy with confidence. Journal of Applied Research in Memory and Cognition, 2(3),

179–184. https://doi.org/10.1016/j.jarmac.2013.06.002

Lindsay, R. C. L., Lea, J. A., Nosworthy, G. J., Fulford, J. A., Hector, J., Le Van, V., &

Seabrook, C. (1991). Biased lineups: Sequential presentation reduces the problem.

Journal of Applied Psychology, 76(6), 796–802. https://doi.org/10.1037/0021-

9010.76.6.796

Mansour, J. K., Beaudry, J. L., Bertrand, M. I., Kalmet, N., Melsom, E. I., & Lindsay, R. C.

L. (2020). Impact of disguise on identification decisions and confidence with

simultaneous and sequential lineups. Law and Human Behavior, 44(6), 502–515.

https://doi.org/10.1037/lhb0000427

Mansour, J. K., & da Costa, J. R. (2015, June 24-27). Testing applied lineup theory using

metacognitions: The relationship between memory strength, decision strategy, and

identification accuracy [Conference presentation]. Society for Applied Research in

Memory and Cognition, Victoria, Canada.

Mansour, J. K., Lindsay, R. C. L, Brewer, N., & Munhall, K. G. (2009). Characterizing visual

behaviour on a lineup task. Applied Cognitive Psychology, 23(7), 1012–1026.

https://doi.org/10.1002/acp.1570

IDENTIFICATION DECISION PROCESSES 39

Moors, A. (2016). Automaticity: Componential, causal, and mechanistic explanations.

Annual Review of Psychology, 67, 263–287. https://doi.org/10.1146/annurev-psych-

122414-033550

Müller, K. (2020). here: A simpler way to find your files. R package version 1.0.1,

https://CRAN.R-project.org/package=here

Nyman, T. J., Lampinen, J. M., Antfolk, J., Korkman, J., & Santtila, P. (2019). The distance

threshold of reliable eyewitness identification. Law and Human Behavior, 43(6), 527–

541. https://doi.org/10.1037/lhb0000342

Penrod, S., & Cutler, B. (1995). Witness confidence and witness accuracy: Assessing their

forensic relation. Psychology, Public Policy, and Law, 1(4), 817–845.

https://doi.org/10.1037/1076-8971.1.4.817

Qualtrics software. (2015). Provo, UT, USA. http://www.qualtrics.com

Quigley-McBride, A., & Wells, G. L. (2023). Eyewitness confidence and decision time

reflect identification accuracy in actual police lineups. Law and Human Behavior,

47(2), 333–347. https://doi.org/10.1037/lhb0000518

R Core Team (2024). R: A language and environment for statistical computing. R Foundation

for Statistical Computing, Vienna, Austria. https://www.R-project.org/.

Revelle, W. (2022). psych: Procedures for personality and psychological research (Version

2.2.5) [R package]. https://CRAN.R-project.org/package=psych

Robinson, M. D., & Johnson, J. T. (1998). How not to enhance the confidence—accuracy

relation: The detrimental effects of attention to the identification process. Law and

Human Behavior, 22(4), 409-428. https://doi.org/10.1023/A:1025770926338

Ross, D. F., Benton, T. R., McDonnell, S., Metzger, R., & Silver, C. (2007). When accurate

and inaccurate eyewitnesses look the same: A limitation of the ‘pop-out’ effect and the

10- to 12-second rule. Applied Cognitive Psychology, 21(5), 677–690.

IDENTIFICATION DECISION PROCESSES 40

https://doi.org/10.1002/acp.1308

Rosseel, Y. (2012). lavaan: An R package for structural equation modeling. Journal of

Statistical Software, 48(2), 1-36. https://doi.org/10.18637/jss.v048.i02

Saraiva, R. B., van Boeijen, I., Hope, L., Horselenberg, R., Sauerland, M., & van Koppen, P.

J. (2020a). Eyewitness metamemory predicts identification performance in biased and

unbiased lineups. Legal and Criminological Psychology, 25(2), 111-132.

https://doi.org/10.1111/lcrp.12166

Saraiva, R. B., Hope, L., Horselenberg, R., Ost, J., Sauer, J. D., & van Koppen, P. J. (2020b).

Using metamemory measures and memory tests to estimate eyewitness free recall

performance. Memory, 28(1), 94-106. https://doi.org/10.1080/09658211.2019.1688835

Sauer, J. D., Palmer, M. A., & Brewer, N. (2019). Pitfalls in using eyewitness confidence to

diagnose the accuracy of an individual identification decision. Psychology, Public

Policy, and Law, 25(3), 147–165. https://doi.org/10.1037/law0000203

Sauerland, M., Sagana, A., & Sporer, S. L. (2012). Assessing nonchoosers' eyewitness

identification accuracy from photographic showups by using confidence and response

times. Law and Human Behavior, 36(5), 394–403. https://doi.org/10.1037/h0093926

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(6), 611–625. https://doi.org/10.1080/10683160701264561

Sauerland, M., & Sporer, S. L. (2009). Fast and confident: Postdicting eyewitness

identification accuracy in a field study. Journal of Experimental Psychology: Applied,

15(1), 46–62. https://doi.org/10.1037/a0014560

Seale-Carlisle, T. M., Colloff, M. F., Flowe, H. D., Wells, W., Wixted, J. T., & Mickes, L.

(2019). Confidence and response time as indicators of eyewitness identification

accuracy in the lab and in the real world. Journal of Applied Research in Memory and

IDENTIFICATION DECISION PROCESSES 41

Cognition, 8(4), 420-428. https://doi.org/10.1016/j.jarmac.2019.09.003

Slane, C. R., & Dodson, C. S. (2022). Eyewitness confidence and mock juror decisions of

guilt: A meta-analytic review. Law and Human Behavior, 46(1), 45.

https://doi.org/10.1037/lhb0000481

Smith, A. M., Smalarz, L., Wells, G. L., Lampinen, J. M., & Mackovichova, S. (2022). Fair

lineups improve outside observers’ discriminability, not eyewitnesses’ discriminability:

Evidence for differential filler-siphoning using empirical data and the WITNESS

computer-simulation architecture. Journal of Applied Research in Memory and

Cognition, 11(4), 534–544. https://doi.org/10.1037/mac0000021

Smith, S. M., Lindsay, R. C. L., & Pryke, S. (2000). Postdictors of eyewitness errors: Can

false identifications be diagnosed? Journal of Applied Psychology, 85(4), 542–550.

https://doi.org/10.1037/0021-9010.85.4.542

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(1), 153–169. https://doi.org/10.1037/1076-8971.7.1.153

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

Behavioural Science / Revue Canadienne des Sciences du Comportement, 36(2), 146–

154. https://doi.org/10.1037/h0087225

Sporer, S. L. (1992). Post-dicting eyewitness accuracy: Confidence, decision-times and

person descriptions of choosers and non-choosers. European Journal of Social

Psychology, 22(2), 157–180. https://doi.org/10.1002/ejsp.2420220205

Steblay, N. K., Dysart, J. E., & Wells, G. L. (2011). Seventy-two tests of the sequential

lineup superiority effect: A meta-analysis and policy discussion. Psychology, Public

Policy, and Law, 17(1), 99–139. https://doi.org/10.1037/a0021650

IDENTIFICATION DECISION PROCESSES 42

Tredoux, C. G. (1998). Statistical inference on measures of lineup fairness. Law and Human

Behavior, 22(2), 217-237. https://doi.org/10.1023/A:1025746220886

Tredoux, C., & Naylor, T. (2018). r4lineups: Statistical inference on lineup fairness (Version

0.1.1) [R package]. https://CRAN.R-project.org/package=r4lineups

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(3), 139–147. https://doi.org/10.1037/1076-

898X.10.3.139

Wells, G. L. (1984). The psychology of lineup identifications. Journal of Applied Social

Psychology, 14(2), 89–103. https://doi.org/10.1111/j.1559-1816.1984.tb02223.x

Wells, G. L. (2020). Psychological science on eyewitness identification and its impact on

police practices and policies. American Psychologist, 75(9), 1316–1329.

https://doi.org/10.1037/amp0000749

Wells, G. L., & Windschitl, P. (1999). Stimulus sampling and social psychological

experimentation. Personality and Social Psychology Bulletin, 25(9), 1115–1125.

https://doi.org/10.1177/01461672992512005

Wickham, H. (2019). stringr: Simple, consistent wrappers for common string operations

(Version 1.4.0) [R package]. https://CRAN.R-project.org/package=stringr

Wickham, H., Chang, W., Henry, L., Pedersen, T. L., Takahashi, K., Wilke, C., Woo, K.,

Yutani, H., Dunnington, D., & Posit, P.B.C. (2016). ggplot2: Elegant Graphics for Data

Analysis (Version 3.4.2) [R package]. https://cran.r-project.org/package=ggplot2

Wickham, H., Franҫois, R., Henry, L., Müller, K., Vaughan, D., & Posit Software, PBC.

(2023). dplyr: A grammar of data manipulation (Version 1.1.2) [R package].

https://cran.r-project.org/package=dplyr

Wickham, H., & Girlich, M. (2022). tidyr: Tidy messy data (Version 1.2.0) [R package].

IDENTIFICATION DECISION PROCESSES 43

https://CRAN.R-project.org/package=tidyr

Wickham, H., Hester, J., & Bryan, J. (2022). readr: Read rectangular text data (Version

2.1.2) [R package]. https://CRAN.R-project.org/package=readr

Wittwer, T., Tredoux, C. G., Py, J., Nortje, A., Kempen, K., & Launay, C. (2022). Automatic

recognition, elimination strategy and familiarity feeling: Cognitive processes predict

accuracy from lineup identifications. Consciousness and Cognition, 98, 103266.

https://doi.org/10.1016/j.concog.2021.103266

Wixted, J. T., & Mickes, L. (2014). A signal-detection-based diagnostic-feature-detection

model of eyewitness identification. Psychological Review, 121(2), 262–276.

https://doi.org/10.1037/a0035940

Wixted, J. T., & Wells, G. L. (2017). The relationship between eyewitness confidence and

identification accuracy: A new synthesis. Psychological Science in the Public Interest,

18(1), 10-65. https://doi.org/10.1177/1529100616686966

IDENTIFICATION DECISION PROCESSES 44

Appendix A: Post-lineup Decision Process Questions

Question

Response Options

How easy was it for you to make your decision?*

1 = extremely difficult –

7 = extremely easy

To what extent did the perpetrator standout to you in the

lineup?*†

1 = not at all distinct –

7 = extremely distinct

Rate the extent to which the following factors influenced your

decision.

1 = no influence –

7 = strong influence

The lineup pictures.

Your memory for the culprit.*

Please rate the extent to which you agree with each of the

following statements.

1 = strongly disagree –

7 = strongly agree

The culprit’s face just ‘popped out’ at me. †‡

I recognized the culprit easily. †‡

I searched the photos for a familiar face. †

I searched the faces for the feature/features I

remembered.

†

I compared the photos to my memory in order to pick

someone who was the closest match to what I

remembered. †

I first eliminated the ones that were definitely not the

culprit, and then chose from those remaining. I didn’t

recognize the culprit easily.

I looked for clues in the lineup as to whom I should pick.

Regarding the influence of the other (non-chosen) photos:

1 = strongly disagree –

7 = strongly agree

They helped confirm my decision, to reinforce my

decision after it was made.

They had little influence on my decision. I knew straight

away who to pick. †

They had some influence on my decision because more

than one face seemed familiar.

†‡

They had some influence on my decision because more

than one face had the feature/features I remember about

the perpetrator. †‡

They confused me, and made the task more difficult. I

didn’t know whom to pick.

†‡

I was able to pick the person because the other lineup

members weren’t really plausible.

Note. Items with a † were included in the final scale for identifiers. Items with ‡ were

included in the final scale for rejectors.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 1

Supplementary Materials

Demographic Questions

Participants were asked in three separate questions to “Please indicate your” ethnic

background, age, and gender. The responses options for ethnic background were Caucasian,

Indigenous, Asian, and Other (specify). Age was entered into a text box. Participants could

select Male or Female for gender8. Participants could also choose not to respond.

Perceived Memory Strength - Manipulation Check

To follow up the significant multivariate result, we examined the univariate results.

Participants in the strong condition reported a significantly better memory than participants in

the weak condition across four of the five manipulation check measures (see Table S1). A

non-significant effect emerged for distance. However, given that we manipulated distance

only by changing the size of the video on screen, we acknowledge that this item was a poor

measure of memory strength because a correct response would be the same across our

manipulation. Based on our analysis, we deemed our manipulation successful.

Table S1

Descriptive and Inferential Statistics for Perceived Memory Strength - Manipulation Check

Weak

(n = 315)

Strong

(n = 342)

M (SD) M (SD) F (1, 655) d [95% CI]

View

5.07 (1.65)

5.90 (1.27)

48.92***

0.55 [0.39, 0.70]

Features

4.61 (1.57)

5.65 (1.17)

89.34***

0.74 [0.58, 0.90]

Memory

4.33 (1.40)

5.34 (1.17)

98.59***

0.78 [0.62, 0.93]

Attention

4.46 (1.56)

5.36 (1.32)

60.22***

0.61 [0.45, 0.76]

Pick out

4.42 (1.56)

5.59 (1.22)

109.72***

0.82 [0.66, 0.98]

Time

6.40 (8.87)

26.6 (19.0)

343.40***

1.45 [1.28, 1.62]

Distance

2.01 (3.56)

2.06 (6.11)

0.02

0.01 [-0.14, 0.16]

Note. Time was reported in seconds and distance in meters. *** p < .0001

8 Note we acknowledge a greater number of options would have been more informative, however, these data

were collected in 2015.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 2

Factor Analysis

Identifications

Checking Appropriateness. Prior to conducting an exploratory factor analysis, it is

appropriate to check the suitability of the data for factor analysis. For identifications, we first

examined the post-lineup questions to determine if they elicited a normal distribution of

responses because significant departures from normality suggest transformations of those

variables may improve results. Inspection of the histograms of responses for each of the 17

items indicated some departure from normality but none were extreme. We examined

whether the two most non-normal items showed a curvilinear relationship; they did not,

therefore we concluded that curvilinearity was unlikely to be an issue in our sample.

Next, we examined our data for factorability. A correlation matrix with each of the 17

items indicated that of the 136 correlations, 55 were above .30 (i.e., factorable). Consistent

with our conclusion regarding factorability, Bartlett’s test of sphericity was significant, χ 2 =

3692.64, p < .001, and the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy

exceeded the recommended value of .60 (KMO = .86; Tabachnick & Fidel, 2013). A KMO

between .8 and .9 is considered meritorious (Kaiser & Rice, 1974). Finally, we examined the

anti-image matrix which provides information about how strongly pairs of items are related

when the influence of other items is removed. An item should be highly correlated with itself

if sampling is adequate; consistent with this conclusion, all 17 values on the diagonal were

above 0.5. Items should be less highly correlated with other items in order to conduct factor

analysis. Indeed, only four of the off-diagonal values were above 0.5 (indicating a lack of

redundancy in our items). These results further support the conclusion of factorability.

We returned to the correlation matrix to assess whether multicollinearity was an issue

in our sample. All 17 items correlated below .90. Indeed, the determinant of the correlation

matrix was .00006. A value greater than .00001 indicates a lack of multicollinearity.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 3

We next checked that the data set was not singular, which we achieved by conducting

an unrotated principal components analysis. That analysis ran successfully indicating a lack

of singularity. This analysis was followed up with an examination of communalities from an

unrotated principal axis factor analysis. High communalities (>.60) mitigate the effects of

small sample sizes (Tabachnick & Fidell, 2013). Communalities ranged from .44 to .98 with

eight items below .60. We concluded that the data were appropriate for factor analysis.

Exploratory Factor Analysis. Satisfied that our data were appropriate for factor

analysis, we produced conducted Horn’s parallel analysis of factors to assess how many

factors were appropriate for our data and found that three factors were appropriate. We thus

began by conducting our factor analysis for three factors using maximum likelihood

extraction with Promax rotation. The solution explained 53% of the total variance.

Per Tabachnik and Fidel (2013), we inspected the pattern matrix for weak items—

those loading below .32 and those that cross-loaded with other items (75% or greater loading

on the non-maximum loaded factor; Samuels, 2016), which indicates that an item does not

discriminate between factors. One item loaded below .32 (pictures9). Two items cross-loaded

above our criterion (eliminated, confirm). Inspection of the communalities indicated that

pictures and clues had communalities below .20 (Child, 2006). All four items were dropped.

We repeated the factor analysis using the remaining 13 items. This three-factor

solution explained 61% of the total variance and all loadings remained above .32 without

cross-loadings. Inspection of the communalities indicated one additional item had a low

communality (plausible). Thus, we repeated the factor analysis without this item. Doing so

resulted in a solution that explained 65% of the variance with all items loading above .32, no

cross-loadings, and all communalities above .20. We retained this solution as the final three-

factor solution.

9 Appendix 1 provides the full question text with the shorthand term (used here) in bold.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 4

Upon inspection of the magnitude of the three correlations amongst the factors, we

noted that Factors 1 and 2 correlated quite highly (r = -.55). This indicates that an

uncorrelated solution is not appropriate for our data. However, Field (2018) notes that when

two factors correlate above .30, researchers should consider combining them, therefore we

next considered a two-factor solution.

We conducted another factor analysis on the 17 items but constrained the solution to

two factors, allowing the factors to correlate (with Promax rotation). The two-factor solution

explained 42% of the sample’s variance with a lower absolute value for the correlation

between the factors (r = -.37) compared to the three-factor solution. In this solution,

remembered fell below our .32 loading criterion and confirm cross-loaded above our

criterion. Furthermore, inspection of the communalities indicated that pictures, familiar,

remembered, compared, clues, confirm, and plausible were all below .20. We removed these

seven items and repeated the factor analysis. The resultant solution explained 66% of the

variance and the two factors correlated similarly to our three-factor solution (r = -.59). Factor

loadings and communalities were all acceptable in this final two-factor solution.

The factors comprising the two-factor solution strongly resembled the first two factors

of our three-factor solution. Notably, Factor 1 of the two-factor solution included six of the

seven items loading on Factor 1 of the three-factor solution; only knew was dropped.

Likewise, Factor 2 of the two-factor solution resembled Factor 2 of the three-factor solution:

all three items from the three-factor solution were included as well as one additional item

(eliminated).

The high correlation between the factors in the final two-factor solution suggests that

combining the factors may be appropriate (Field, 2018). A one-factor solution was produced

following the same procedures used above. The result was a 9-item solution which explained

53% of the variance. This factor comprised the same six items as Factor 1 of the two-factor

EYEWITNESS IDENTIFICATION DECISION PROCESSES 5

solution (and were included on Factor 1 of the three-factor solution) and the three items that

appeared on Factor 2 of both the two- and three-factor solutions.

Choosing Between Solutions. To choose between the solutions, we examined model

fit. For each solution we compared the reproduced correlation matrix (i.e., correlations

predicted/implied by the factors loadings) to the correlation matrix from the (unrotated)

solutions (i.e., observed correlations). Less than 50% of the residual differences greater than

.05 is generally considered acceptable (Field, 2018). We also used Bikos’ (2022) fit measure

which involves calculating the sum of the squared residuals divided by the sum of the

squared correlations and subtracting this value from one. A value greater than .95 indicates

good fit.

For the three-factor solution, just 3% of the residuals were above .05, specifically:

easy-popped out10 (.054) and knew-confused (.075). Field (2018) suggests that the mean of

the residuals should be below .08, and it was (M = .02). Residuals should be normally

distributed and a histogram showed this to be the case. The three-factor solution (.954) was a

good fit to our data based on Bikos’ (2022) measure. For the two-factor solution, 11% of

residuals were above .05, the mean of the residuals was .03, the histogram of residuals was

normal, and fit was .945. Thus, the solution was acceptable but did not reach Bikos’ .95

criterion. Finally, for the one-factor solution, the proportion of residuals above .05 was

acceptable (28%) but the mean of the residuals was greater than .08 (.13) which, according to

Field, suggests more factors should be extracted. Furthermore, the residuals were positively

skewed (cf. normal) and fit was only .76. Thus, the three-factor solution was the preferred

solution.

Rejections

Checking Appropriateness. We started with a check of whether these data were

10 Appendix 1 provides the full question text with the shorthand term (used here) in bold.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 6

appropriate for factor analysis. Inspection of the histograms of responses for each of the 17

items indicated considerable departure from normality. When we examined whether the two

most non-normal items showed a curvilinear relationship; a curvilinear pattern was present,

indicating curvilinearity is an issue with these data. However, though normality is preferred,

factor analysis is robust despite non-normality, we did not transform our items.

Next, we considered factorability. A correlation matrix with each of the 17 items

indicated that of the 136 correlations, only 13 were above .30 which is a small number but

still potentially enough for factor analysis. Bartlett’s test of sphericity was significant, χ 2 =

1366.387, p < .001, and the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy

exceeded the recommended value of .60 (KMO = .70; Tabachnick & Fidel, 2013). A KMO

between .5 and .7 is considered mediocre while .7 to .8 is considered good (Kaiser & Rice,

1974). Finally, we examined the anti-image matrix. All 17 values on the diagonal were above

0.5 and only six of the off-diagonal values were above 0.5. These results support a conclusion

of factorability, despite the small number of items with correlations above .30.

There was no concern with multicollinearity for the items as all 17 items correlated

below .90. The determinant of the correlation matrix was .006, much higher than the

threshold of .00001 for a lack of multicollinearity.

An unrotated principal components analysis ran successfully, therefore the structure

was not singular. Communalities from an unrotated principal axis factor analysis ranged from

.32 to .85, with six items below .60. High communalities (>.60) mitigate the effects of small

sample sizes (Tabachnick & Fidell, 2013), a slight concern in this case.

In conclusion, although the rejection data are not ideal for factor analysis (non-normal

items, small number of items correlating with multiple other items above .30), they did meet

the minimum requirements for being suitable for factor analysis.

Exploratory Factor Analysis. Factor analysis was used to determine the underlying

EYEWITNESS IDENTIFICATION DECISION PROCESSES 7

structure of the 17-item questionnaire. Horn’s parallel analysis indicated a three-factor

solution was appropriate. Thus, we conducted a factor analysis for a three-factor solution

with Promax rotation. The solution explained 37% of the total variance in our data set.

Following Tabachnik and Fidel (2013), we again dropped items which loaded below

.32 or cross-loaded with other items based on Samuels’ (2016) rule. The pattern matrix

indicated that pictures loaded below .32 on all three factors (max loading = .20). In addition,

eliminated cross-loaded with factors 2 and 3. Finally the the communalities of pictures,

compared, eliminated, clues, confirm, knew, and plausible were below.

We repeated our factor analysis excluding the 7 items identified as problematic above.

This solution explained 69% of the variance. None of the remaining items loaded below .32,

cross-loaded with other items above our criterion, or had communalities below .20.

The next step was to consider the correlations amongst our factors. As with the scale

we developed for identifications, Factors 1 and 2 correlated quite highly (r = -.54), which

indicated that an uncorrelated solution (i.e., varimax rotation) would be inappropriate.

However, Field (2018) suggests combining factors that correlate above .30, therefore we

considered whether a two-factor solution would be more appropriate for these data.

We conducted a new factor analysis with Promax rotation, using all 17 items, but

constrained the analysis to two factors. This solution explained 28% of the sample’s variance.

We again applied our minimum loading criterion, cross-loading criterion, and minimum

communality criterion. Eight items: pictures, memory, familiar, remembered, compared,

eliminated, clues, and confirm failed to load above .32. No items cross-loaded. Inspection of

the communalities indicated 11 items could be dropped: easy, pictures, memory, familiar,

remembered, compared, eliminated, clues, confirm, knew, and plausible.

Repeating the factor analysis without the 11 items resulted in a solution that explained

79% of the variance. This two-factor solution resulted in no items with loadings below .32,

EYEWITNESS IDENTIFICATION DECISION PROCESSES 8

cross-loaded, or with communalities below .20. However, the stability of this 6-item solution

must be questioned as feature loaded above 1.00 (1.007 on Factor 2) which can signal a

problem. The correlation between the factors was relatively high (r = .56).

Given this high correlation, we considered a one-factor solution. This solution

explained 13% of the sample’s response variance but only three items met our loading

criterion (face, feature, confused). The one-factor solution explained 63% of the sample’s

response variance with all three items loading above.32 and communalities above .20.

Choosing Between Solutions. For the three-factor solution, 9% of residuals were

above .05 with a mean of .046, and a normal distribution. These values are acceptable based

on Field (2018); however, model fit was .83 so well below the acceptable threshold suggested

by Bikos (2022). The two-factor solution produced no residuals above .05, with a mean of

.012, and a normal distribution. In contrast to the three-factor model, fit of the two-factor

model was .97. The one-factor solution had no residuals above .05, a mean residual of .014,

and a non-normal (positively skewed) distribution of residuals with a fit of .97. The two- and

one-factor solutions were similar on all metrics except the distribution of residuals, where the

two-factor solution was preferable because its residuals were normal (vs. skewed). Thus, we

judged the two-factor solution as preferable.

Generalizability Across Memory Strength Conditions (Identifications)

We examined whether our memory strength manipulation affected eyewitness

decision processes. First, we tested the generalizability of the three-factor structure for

identifications across the two memory strength conditions. Second, we created composite

scores for each of the factors and compared them across the two memory strength conditions.

Generalizability of Factor Structure Across Memory Strength Conditions

We conducted a confirmatory factor analysis (CFA) to ensure our memory strength

manipulation did not interact with our observed factor structure. Following Brown (2006), we

EYEWITNESS IDENTIFICATION DECISION PROCESSES 9

calculated absolute fit, fit adjusting for model parsimony, and comparative fit for the CFA

models. As measures of absolute fit, we used χ2 and standardized root mean square residual

(SRMR). As a measure of fit adjusting for model parsimony, we used the root mean-square

error of approximation (RMSEA). Finally, as a measure of comparative fit, we used the

comparative fit index (CFI). Poor, acceptable, and good model fit parameters for the

indices—per Bikos (2022) and Awang (2015)—are provided in Table S2.

Table S3

Parameters Indicating Poor, Acceptable, and Good Model Fit by Fit Index

Fit Index

Model Fit Parameters

Poor

Acceptable

Good

Chi-square (p)

≤ .05

> .05

SRMR

≥ .10

< .10*

≤ .05

RMSEA

≥ .10

< .10*

≲ .05

CFI

≤ .90

> .90

≥ .95

Note. Based on Bikos (2022) and Awang (2015). SRMR = Standardized root mean square.

RMSEA = Root mean-square error of approximation. CFI = Comparative fit index. * Hooper

et al. (2008) and West, Taylor, and Wu (2012) consider ≤ .08 as the appropriate cutoff.

We used a Maximum Likelihood extraction method with a multi-sample approach to

determine if there was model invariance between the weak and strong Memory Strength

conditions (for a detailed description of this approach, see Brown, 2006). We set the first

model to be unconstrained and set the second with constraints on measurement weights

(factor loadings), measurement intercepts, and covariances (residuals and latent variables).

The difference between the goodness-of-fit statistics for the two models was not significant,

∆χ2(21) = 29.19, p = .11, indicating that the constraints were equal across the groups and no

model invariance. Per Table S3, the χ2 for both the constrained and unconstrained models

was significant, indicating a cause for concern. However, this index is overly sensitive,

including when one’s sample size is greater than 200 as in this case (Awang, 2015; Babyak &

EYEWITNESS IDENTIFICATION DECISION PROCESSES 10

Green, 2010; Bikos, 2022; Hooper et al., 2008; West, Tayler, & Wu, 2012). Kline (2016)

notes that a failed exact-fit test (like the chi-square test) gives preliminary evidence against a

model while passing it provides preliminary support; therefore, local fit testing (vis-à-vis fit

approximate fit indices) is also necessary. Our other fit indices indicated adequate model fit,

although if .08 is used as the criterion for acceptable fit for SRMR and RMSEA, the

unconstrained model does not meet that criterion for RMSEA.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 11

Table S3

Model Fit Indices for the Confirmatory Factor Analyses for Multiple-Sample Models and for the Strong and Weak Memory Strength Conditions

Fit Index

Multi-Sample Models

Unconstrained Revised Models by

Memory Strength

Unconstrained

Original

(df = 102)

Constrained

Original

(df = 123)

Unconstrained

Revised

(df = 100)

Constrained

Revised

(df = 122)

Strong

(df = 50)

Weak

(df = 50)

Chi-Square

269.47*

298.67*

223.54*

251.73*

86.76*

136.78*

SRMR

.06

.07

.05

.06

.04

.06

RMSEA

.09

.08

.07

.06

.10

CFI

.94

.96

.98

.94

Note. SRMR = Standardized root mean square. RMSEA = Root mean-square error of approximation. CFI = Comparative fit index. * p ≤ .001.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 12

Given that fit may not be ideal and per Kline (2016), we considered why this may be

the case. We noted that the cross-loading for confused on Factor 1 was markedly higher than

for other items. Inspection of modification indices showed that the path between confused

and the latent variable represented by Factor 1 had the second largest index. Thus, we added

this path to our factor structure. Confused loads negatively on the Automatic Response factor

therefore inclusion of this item does make some sense as an eyewitness who automatically

recognizes the culprit is not going to be confused by the presence of other images, much like

recognition of one’s mother in a crowd of people is no more difficult than recognizing her on

her own. Adding this path had a significant effect on our fit indices as can be seen in the

Unconstrained Revised and Constrained Revised sections of Table S3. Although the chi-

square statistics remained significant, the remaining fit metrics reflected acceptable fit.

We next conducted separate unconstrained confirmatory factor analyses to test model

fit for each of the two memory strength conditions using our revised structure. As seen in

Table S3, the χ2 was significant for both conditions. The other model fit indices were

acceptable, except for the RMSEA for the weak memory strength condition which is outside

the acceptable range if .08 (cf. .10) is used as the criterion for acceptable fit. These results

suggest the weak memory strength condition has marginal model fit. Notably, the model fit

the strong memory condition data better than the weak memory condition data.

Composite Scores Across Memory Strength Conditions

To further investigate how memory strength affects decision processes and as a basic

check of criterion validity, we created composite scores for each factor by summing the raw

scores for each participant on the items comprising each factor. Independent samples t-tests

were then used to compare the composite scores for each Memory Strength group for each

factor. For Automatic Response, scores could range from 7-49 and for both Relative

Judgment and Absolute Judgment, scores could range from 3-21. Participants in the strong

EYEWITNESS IDENTIFICATION DECISION PROCESSES 13

condition scored higher on Automatic Response (M = 32.65, SD = 7.48, Range = 10-47),

t(383.16) = 3.70, p < .001, d = 0.38 [0.18, 0.58], and lower on Relative Judgment (M = 9.58,

SD = 4.66, Range = 3-21), t(379.91) = 1.98, p = .048, d = 0.20 [0.00, 0.41], than those in the

weak condition (M = 29.77, SD = 7.91, Range = 9-49 for Automatic Response; M = 10.57, SD

= 5.10, Range = 3-21 for Relative Judgement). There was no evidence of a difference in

Absolute Judgment scores for the strong (M = 16.67, SD = 3.24) and weak conditions (M =

16.44, SD = 3.59), t(378.13) = 0.64, p = .52, d = 0.07 [ -0.14, 0.27]. We also conducted this

analysis on the regression factor scores and found slightly different results: in this case, there

was no difference by memory condition for scores on the Relative Judgment factor (p = .89).

Exploratory Factor Analysis of Pre-lineup Questionnaire

Responses to our pre-lineup questionnaire were of secondary interest. Only choosers

were included in these analyses. One participant that did not respond to all the scale questions

were dropped so our sample comprised 392 participants. The 17 items we factor analysed

were similar to those used in the post-lineup questionnaire (see Appendix A); specific

wording can be found on the project’s OSF page.

The histograms of responses indicated minimal departure from normality. Inspection

of the scatterplot for the two most non-normal items confirmed that curvilinearity was not a

concern; the curvilinear and linear best-fitting lines were very similar.

Next, we examined our data for factorability. Fewer correlations between items were

above .30 (i.e., factorable) on this questionnaire (51) compared to the post-lineup

questionnaire (85). Nonetheless, Bartlett’s test of sphericity was significant, χ 2 (136) =

2715.24, p < .001 and the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was

greater than .60 (KMO = .81 or meritorious; Kaiser & Rice, 1974; Tabachnick & Fidel,

2013). All 17 values on the diagonal and only four of the off-diagonal values were above 0.5.

Thus, we concluded that our data were factorable.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 14

Multicollinearity was not a concern as no items correlated above .90 and the

determinant of the correlation matrix was much greater than .00001 (det = .001). Next, using

an unrotated principal components analysis we confirmed the data were not singular. An

unrotated principal axis factor analysis indicated that communalities ranged from .45 to .91

with five items below .60. We concluded that our data were appropriate for factor analysis.

Horn’s parallel analysis of factors suggested three factors. Given this, we conducted

our factor analysis for three factors using maximum likelihood extraction with Promax

rotation. The solution explained 44% of the total variance. Inspection of the factor loadings

showed eliminate cross-loaded (i.e., 75% or greater loading on the non-maximum loaded

factor; Samuels, 2016) but all items loaded above .32. Two items had communalities below

.20 (pictures and clues). All three items were dropped.

Factor analysis on the remaining items explained 50% of the total variance but

confirm was below the .32 loading criterion and the .20 communality criterion (no items

cross-loaded). Thus, we dropped confirm. All loadings in a factor analysis with the remainign

12 items were above .32 without cross-loadings and all communalities were above .20. At

this point, we checked whether the three-factor solution met the criteria outlined by Field

(2018) for maintaining more factors based on Kaiser’s criterion. Specifically, we looked at

whether either the communalities of all items were above .70 or the mean of the

communalities were above .60. Only four of the 12 items had communalities above .70 and

the mean of the communalities was .53. Thus, we concluded that our three-factor solution

was unacceptable and conducted an exploratory factor analysis with Promax rotation on the

original 17 items but constrained it to two factors.

The two-factor solution explained 39% of the sample’s variance. No items had factor

loadings below .32 but three items cross-loaded (familiar, compare, and confuse) and three

items had communalities below .20 (pictures, clues, and plausible). We removed these items

EYEWITNESS IDENTIFICATION DECISION PROCESSES 15

and repeated the factor analysis on the remaining 11 items. The resultant solution explained

48% of the variance with all loadings above .32, no cross-loading, and all communalities

above .20. There was a negligible correlation between the factors (r = -.03) indicating

orthogonal factors. Thus, we rotated the final two-factor solution with a Varimax instead of a

Promax rotation. Table S4 depicts our final factor structure and other relevant indices.

Table S4

Summary of the Final Factor Structure for Identifiers’ Pre-lineup Decision Process Items

Item

Factor loadings

Automatic

Relative

I expect to recognize the culprit easily.

.89

.06

.79

When viewing the lineup, I expect the culprit’s face to just

‘pop out’ at me.

.78 .08 .61

To what extent do you think the culprit will stand out to

you in the lineup?

.75 .01 .56

How easy do you think it will be for you to make your

decision?

.71 .03 .51

[The non-chosen pictures] will probably have little

influence on my decision. I will know straight away who

to pick.

.55 -.14 .32

Rate the extent to which you expect your memory for the

culprit influenced your decision.

.47 .11 .23

[The non-chosen pictures] will have some influence on my

decision because more than one face may have the

feature/features I remember about the perpetrator.

-.07 .93 .87

[The non-chosen pictures] will have some influence on my

decision because more than one face may seem familiar.

-.03 .87 .77

I expect to search the faces for the feature/features I

remember

.26 .42 .24

[The non-chosen pictures] will help confirm my decision,

to reinforce my decision after it is made.

.22 .39 .20

I expect I will first eliminate the ones that are definitely

not the culprit, and then choose from those remaining. I

don't expect to recognise the culprit easily

-.23 .38 .20

Unrotated sums of square loadings

3.17

2.13

Rotated sums of square loadings

3.16

2.14

Proportion variance (rotated)

.29

.48

Note. Factor loadings > 0.30 are bolded. λ2 = communality. Automatic = Automatic

Response. Relative = Relative Judgment.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 16

We also checked how well our final two-factor solution fit the data. Specifically, we

considered the extent to which the reproduced correlation matrix (i.e., the correlations

predicted by the factors loadings) and the correlation matrix from the (unrotated) solutions

differed; less than 50% of these residuals should be greater than .05 (Field, 2018). For our

solution, 29% of residuals were above .05. The mean of the residuals was .05 and the

histogram of residuals was normal, which also indicate good fit. Finally, we examined fit by

summing the squared residuals and dividing them by the sum of the squared correlations

(Bikos, 2022). We were well below the .95 criterion at .83. Thus, we found mixed results in

terms of model fit.

We also subjected the pre-lineup scale to a reliability analysis (see Table S5). The

extent to which each factor was reliable depended on the measure considered. Based on

Chronbach’s alpha, the Automatic Response factor met the criterion for good reliability (≥

.80; George & Mallery, 2019)while the Relative Judgment meets the criterion for acceptable

reliability (≥ .70). However, when McDonald’s omega is considered, Automatic Response

meets the criterion for excellent reliablity (≥ .90) while Relative Judgment meets the criterion

for good reliability.

The items on the final pre-lineup scale were similar to the items on the post-lineup

scale. The six items that comprised Factor 1 on the post-lineup questionnaire (three-factor

solution) comprised Factor 1 on the pre-lineup questionnaire, therefore we also labelled the

first factor on this scale Automatic Response. As discussed in the main paper, the items

reflect a quick and effortless decision process with minimal conscious input (e.g., “I expect to

recognize the culprit easily”). Factor 2 comprised all three items that loaded on Factor 2 of

the post-lineup scale but included two additional items, eliminate and confirm, which were

not included in the final post-lineup questionnaire. We labelled Factor 2 on the pre-lineup

scale as Relative Judgment because most of the items point to comparisons between lineup

EYEWITNESS IDENTIFICATION DECISION PROCESSES 17

members (e.g., “[The non-chosen pictures] will have some influence on my decision because

more than one face may have the feature/features I remember about the perpetrator”).

Table S5

Internal Consistency, Item-total Correlations, and Descriptives for Pre-lineup Identifier

Scale

Factors & Items M (SD)

Item-total

Correlation

ɑ if item

deleted

Automatic Response

(ɑ = .84 95% CI [.82, .86]; ω = .92)

4.47 (1.01)

I expect to recognize the culprit easily.

.86

.78

To what extent do you think the culprit will stand

out to you in the lineup?

4.19 (1.23) .80 .80

How easy do you think it will be for you to make

your decision?

4.24 (1.24) .76 .80

When viewing the lineup, I expect the culprit’s

face to just ‘pop out’ at me.

4.84 (1.41) .74 .81

[The non-chosen pictures] will probably have little

influence on my decision. I will know straight

away who to pick.

3.64 (1.57) .68 .84

Rate the extent to which you expect your memory

for the culprit influenced your decision.

5.26 (1.27) .47 .85

Relative Judgment

(ɑ = .71 95% CI [.66, .75] ; ω = .80)

4.93 (0.96)

[The non-chosen pictures] will have some

influence on my decision because more than one

face may have the feature/features I remember

about the perpetrator.

4.94 (1.28) .86 .58

[The non-chosen pictures] will have some

influence on my decision because more than one

face may seem familiar.

4.87 (1.24) .83 .60

I expect to search the faces for the feature/features

I remember

5.55 (1.17) .42 .70

[The non-chosen pictures] will help confirm my

decision, to reinforce my decision after it is made.

4.55 (1.41) .44 .69

I expect I will first eliminate the ones that are

definitely not the culprit, and then choose from

those remaining. I don't expect to recognise the

culprit easily

4.72 (1.83) .42 .73

Note. ɑ = Cronbach’s alpha. ω = McDonald’s omega. In all cases, ratings ranged from 1-7.

Bolded words are shorthand terms for items.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 18

Predicting Identification Accuracy from the Pre-lineup Questions

As we were able to develop a reliable scale from our pre-lineup questionnaire, we

wanted to determine whether these questions had any predictive ability and therefore

probative value. To that end, we considered whether composite scores derived from our two-

factor solution for the pre-lineup questionnaire predicted identification accuracy. We entered

the composite scores for Automatic Response and Relative Judgment into a binomial logistic

regression predicting identification accuracy. Automatic Response, β = 0.26, SE = 0.10, z =

2.51, p = .012, OR = 1.04 [1.01, 1.08], was a significant predictor, but Relative Judgment β =

0.08, SE = 0.10, z = 0.80, p = .42, OR = 1.02 [0.98, 1.06], was not.

Pre-lineup and Post-lineup Questionnaire Item-Pair Correlations

We examined our decision process questionnaire item-pair correlations to determine

whether the pre-lineup questionnaire primed participants before their lineup decision and the

post-lineup questionnaire. As seen in Table S6, ratings by participants who made an

identification tended to be moderately positively correlated, .15 ≤ rs ≤ .49, ps ≤ .002. For

those who made rejections, the picture was slightly more complex. For most items, the

correlations were moderate and positive, .13 ≤ rs ≤ .56, ps ≤ .02. However, three items

(standout, poppedout, and features) showed low, nonsignificant correlations.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 19

Table S4

Item-Pair Correlations from the Pre-lineup and Post-lineup Decision Process Questionnaires

Identifications (df = 389)

Rejections (df = 277)

Item

Pearson r

Sig.

Pearson r

Sig.

easy

.29

< .001

.20

< .001

standout

.15

.002

.06

.29

picture

.31

< .001

.35

< .001

memory

.37

< .001

.26

< .001

poppedout

.18

< .001

.07

.23

easily

.27

< .001

.17

.005

familiar

.39

< .001

.30

< .001

remembered

.41

< .001

.20

< .001

compared

.26

< .001

.16

.007

eliminated

.35

< .001

.38

< .001

clues

.49

< .001

.56

< .001

confirm

.27

< .001

.34

< .001

knew

.26

< .001

.28

< .001

face

.16

.001

.14

.02

features

.24

< .001

.09

.15

confused

.28

< .001

.23

< .001

plausible

.29

< .001

.20

.001

EYEWITNESS IDENTIFICATION DECISION PROCESSES 20

References

Awang, Z. (2015). Validating the measurement model: CFA. In SEM made simple: A Gentle

Approach to Learning Structural Equation Modelling (pp. 54-74). MPWS.

Babyak, M. A., & Green, S. B. (2010). Confirmatory factor analysis: An introduction for

psychosomatic medicine researchers. Psychosomatic Medicine, 72(6), 587–597.

https://doi.org/10.1097/PSY.0b013e3181de3f8a

Bikos, L. H. (2022). ReCentering psych stats: Psychometrics.

https://lhbikos.github.io/ReC_Psychometrics/index.html

Brown, T. A. (2006). Confirmatory factor analysis for applied research. Guilford Press.

Child, D. (2006). The essentials of factor analysis. A&C Black.

DeVellis, R. F. (2017). Scale development: Theory and applications (4th edition). Sage.

DiStefano, C., Zhu, M., & Mîndrilă, D. (2009). Understanding and using factor scores:

Considerations for the applied researcher. Practice Assessment, Research &

Evaluation, 14, 1–11. https://doi.org/10.7275/da8t-4g52

Field, A. (2018). Exploratory Factor Analysis. In Discovering statistics using IBM SPSS

Statistics (pp. 989-1060). Sage.

George, D., & Mallery, P. (2019). IBM SPSS statistics 25 step by step: A simple guide and

reference. Taylor & Francis.

Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modelling: Guidelines

for determining model fit. Electronic Journal of Business Research Methods, 6(1), 53-

60.

Kaiser, H. F., & Rice, J. (1974). Little jiffy, mark IV. Educational and Psychological

Measurement, 34(1), 111-117. https://doi.org/10.1177/001316447403400115

Kline, R. B. (2016). Principles and practice of structural equation modeling (4th ed.).

Guilford Publications.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 21

Samuels, P. (2016). Advice on exploratory factor analysis (Working Paper).

https://doi.org/10.13140/RG.2.1.5013.9766

Tabachnick, B. G., & Fidel, L. S. (2013). Using multivariate statistics (6th ed.). Allyn and

Bacon.

West, S.G., Taylor, A.B., Wu, W. (2012). Model fit and model selection in structural

equation modeling. In R. H. Hoyle (Ed.). Handbook of structural equation modeling (p

209-231). The Guilford Press.

EYEWITNESS IDENTIFICATION DECISION PROCESSES 22

Appendix B: Parameters Indicating Poor, Acceptable, and Good Model Fit by Fit Index

Fit Index

Model Fit Parameters

Poor

Acceptable

Good

Chi-square (p)

≤ .05

> .05

SRMR

≥ .10

< .10*

≤ .05

RMSEA

≥ .10

< .10*

≲ .05

CFI

≤ .90

> .90

≥ .95

Note. Based on Bikos (2022) and Awang (2015). SRMR = Standardized root mean square.

RMSEA = Root mean-square error of approximation. CFI = Comparative fit index. * Hooper

et al. (2008) and West, Taylor, and Wu (2012) consider ≤ .08 as the appropriate cutoff.

ResearchGate has not been able to resolve any citations for this publication.

Eyewitness Confidence and Decision Time Reflect Identification Accuracy in Actual Police Lineups

Article

Full-text available

Feb 2023

Objective: Although there are many lab-based studies demonstrating the utility of confidence and decision time as indicators of eyewitness accuracy, there is almost no research on how well these variables function for lineups in the real world. In two experiments, we examined confidence and decision time associated with real lineups that had been conducted using research-based recommendations. Hypotheses: We expected that how confident an eyewitness sounded and how quickly that eyewitness made their identification would be associated with whether that eyewitness identified a suspect or a filler. We also hypothesized that people's interpretations of eyewitness confidence could be easily influenced by additional, biasing information. Method: Using audio recordings of these lineups, we examined (a) participants' subjective ratings of how confident an eyewitness sounded at the time of the identification and (b) objective data regarding how quickly the eyewitness made the identification decision. We also manipulated what additional information, if any, eyewitnesses received in Experiment 2. Results: In both experiments, decision time and confidence predicted whether the eyewitnesses identified the suspect or a known-innocent filler, and when decision time and confidence diverged, it is likely that the eyewitness identified a filler. In Experiment 2, we found that people's interpretations of eyewitness's confidence statements could be biased. When observers believed that the witness picked a filler rather than a suspect, or vice versa, this changed how confident they thought the witness sounded. Conclusions: Confidence and decision time should both be collected when administering real lineups, but objective decision time data may be the most useful because people's perceptions of confidence are easily altered. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Fair Lineups Improve Outside Observers' Discriminability, Not Eyewitnesses' Discriminability: Evidence for Differential Filler-Siphoning Using Empirical Data and the WITNESS Computer-Simulation Architecture

Article

Full-text available

Mar 2022

General Audience Summary There is consensus among psychological scientists that fair lineups lead to a better trade-off between guilty-suspect identifications and innocent-suspect identifications than do biased lineups. Yet, debate over the mechanism that leads to this fair-lineup advantage abounds. On the one hand, some propose that fair lineups increase the ability of eyewitnesses to discriminate between the guilty and the innocent. On the other hand, others have proposed that fair lineups do not increase eyewitness discriminability, but simply spread choice preferences to fillers, and that this process is more pronounced when the suspect is innocent than when the suspect is guilty. In our first experiment, we show that fair lineups lead to (a) fewer perpetrator identifications, (b) fewer forced-choice perpetrator identifications, and (c) a similar rate of correct rejections, when compared to biased lineups. An improvement in eyewitness discriminability must manifest as an increase in perpetrator identifications, an increase in correct rejections, or both. Because fair lineups result in fewer perpetrator identifications and no more correct rejections than do biased lineups, it follows that they do not improve eyewitness discriminability. In our second experiment, we used computational modeling to demonstrate that a spreading effect we call differential filler-siphoning produces a fair-lineup advantage even when the psychology of simulated witnesses is held constant across fair and biased lineups.

Eyewitness Confidence and Mock Juror Decisions of Guilt: A Meta-Analytic Review

Article

Full-text available

Feb 2022

Objective: We investigated the impact of eyewitness confidence on the following dependent variables: (a) guilty or not-guilty verdict; (b) judgments of guilt as measured on a scale; and (c) mock jurors' perception of the accuracy of an eyewitness's identification. In addition, we examined two potential moderators of the effects of eyewitness confidence: (a) whether the eyewitness expressed confidence at trial versus during the initial lineup identification and (b) whether the eyewitness provided a numerical versus a verbal statement of confidence. Hypotheses: We expected all analyses to reveal that highly confident eyewitnesses are more persuasive to mock jurors than are eyewitnesses with lower confidence (Hypothesis 1). We expected eyewitness confidence at trial (relative to at identification) to be more persuasive to mock jurors (Hypothesis 2). We expected numerical expressions of confidence to be more persuasive to mock jurors than verbal confidence expressions (Hypothesis 3). Method: We conducted a meta-analysis of 35 studies from 20 published papers and seven theses or dissertations to quantify the effect of eyewitness confidence on juror judgments and investigated the influence of two primary moderator variables, time of confidence and format of confidence expression. Results: All analyses revealed an effect of eyewitness confidence on mock juror decisions (gs = .21-.36). Our moderator analysis showed that the timing of the confidence statement (identification vs. trial) did not affect the influence of eyewitness confidence on mock jurors' judgments of guilt or accuracy. The influence of eyewitness confidence was not moderated by verbal versus numerical expressions of confidence. Conclusions: Although eyewitness confidence is persuasive to mock jurors, the size of this effect is modest. Moreover, verbal and numerical expressions of confidence have similar persuasive effects, and mock jurors do not appear to be sensitive to the likely difference in evidentiary strength of eyewitness confidence expressed at the initial identification versus at trial. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Impact of Disguise on Identification Decisions and Confidence With Simultaneous and Sequential Lineups

Article

Full-text available

Dec 2020

Objective: Prior research indicates disguise negatively affects lineup identifications but the mechanisms by which disguise works have not been explored and different disguises have not been compared. We investigated how two different types of disguise, four levels of varying degrees of coverage, and lineup type influence eyewitnesses' identification decisions, accuracy, and confidence. Hypotheses: We predicted that identification accuracy would decrease as the disguise covered more of a perpetrator's face. We also predicted that type of disguise-stocking mask versus sunglasses and/or toque (i.e., knitted hat)-would influence identifications, but we had conflicting predictions about which disguise would impair their performance more. Method: In two experiments (Ns = 87 and 91) we manipulated degree of coverage by two different types of disguise: a stocking mask or sunglasses and toque. Participants viewed mock-crime videos followed by simultaneous or sequential lineups. Results and conclusions: Disguise and lineup type did not interact. In support of the view that disguise prevents encoding, identification accuracy generally decreased with degree of disguise. For the stocking disguise, however, full and 2/3 coverage led to approximately the same rate of correct identifications-which suggests that disrupting encoding of specific features may be as detrimental as disrupting a whole face. Accuracy was most affected by sunglasses and we discuss the role meta-cognitions may have played. Lineup selections decreased more slowly than accuracy as coverage by disguise increased, indicating witnesses are insensitive to the effect of encoding conditions on accuracy. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Psychological Science on Eyewitness Identification and Its Impact on Police Practices and Policies

Article

Full-text available

Dec 2020

Gary L. Wells

The development of forensic DNA testing has led to the discovery of hundreds of cases of mistaken eyewitness identification in which innocent people were convicted. Although these discoveries of wrongful convictions from mistaken identification based on DNA testing have been a surprise and shock to the legal system and the public, psychological scientists have been less surprised. This is because psychological scientists were “blowing the whistle” on the eyewitness identification problem for decades prior to forensic DNA testing. Today, most law enforcement agencies in the United States have adopted reformed policies and procedures on eyewitness identification that are based on research by experimental social and cognitive psychologists. This article describes core aspects of this research and how the research has managed to have this impact on the U.S. legal system.

effectsize: Estimation of Effect Size Indices and Standardized Parameters Aims of the Package

Article

Full-text available

Dec 2020

In both theoretical and applied research, it is often of interest to assess the strength of an observed association. This is typically done to allow the judgment of the magnitude of an effect (especially when units of measurement are not meaningful, e.g., in the use of estimated latent variables; Bollen, 1989), to facilitate comparing between predictors' importance within a given model, or both. Though some indices of effect size, such as the correlation coefficient (itself a standardized covariance coefficient) are readily available, other measures are often harder to obtain. effectsize is an R package (R Core Team, 2020) that fills this important gap, providing utilities for easily estimating a wide variety of standardized effect sizes (i.e., effect sizes that are not tied to the units of measurement of the variables of interest) and their confidence intervals (CIs), from a variety of statistical models. effectsize provides easy-to-use functions, with full documentation and explanation of the various effect sizes offered, and is also used by developers of other R packages as the back-end for effect size computation, such as parameters (Lüdecke et al., 2020), ggstatsplot (Patil, 2018), gtsummary (Sjoberg et al., 2020) and more. Comparison to Other Packages effectsize's functionality is in part comparable to packages like lm.beta (Behrendt, 2014), MOTE (Buchanan et al., 2019), and MBESS (K. Kelley, 2020). Yet, there are some notable differences, e.g.: • lm.beta provides standardized regression coefficients for linear models, based on post-hoc model matrix standardization. However, the functionality is available only for a limited number of models (models inheriting from the lm class), whereas effectsize provides support for many types of models, including (generalized) linear mixed models, Bayesian models, and more. Additionally, in additional to post-hoc model matrix standardization, effectsize offers other methods of standardization (see below). • Both MOTE and MBESS provide functions for computing effect sizes such as Cohen's d and effect sizes for ANOVAs (Cohen, 1988), and their confidence intervals. However, both require manual input of For t-statistics, degrees of freedom, and sums of squares for the computation the effect sizes, whereas effectsize can automatically extract this information from the provided models, thus allowing for better ease-of-use as well as reducing any potential for error.

Eyewitness metamemory predicts identification performance in biased and unbiased line‐ups

Article

Full-text available

Feb 2020
LEGAL CRIMINOL PSYCH

Purpose Distinguishing accurate from inaccurate identifications is a challenging issue in the criminal justice system, especially for biased police line‐ups. That is because biased line‐ups undermine the diagnostic value of accuracy post‐dictors such as confidence and decision time. Here, we aimed to test general and eyewitness‐specific self‐ratings of memory capacity as potential estimators of identification performance that are unaffected by line‐up bias. Methods Participants (N = 744) completed a metamemory assessment consisting of the Multifactorial Metamemory Questionnaire and the Eyewitness Metamemory Scale and took part in a standard eyewitness paradigm. Following the presentation of a mock‐crime video, they viewed either biased or unbiased line‐ups. Results Self‐ratings of discontentment with eyewitness memory ability were indicative of identification accuracy for both biased and unbiased line‐ups. Participants who scored low on eyewitness metamemory factors also displayed a stronger confidence–accuracy calibration than those who scored high. Conclusions These results suggest a promising role for self‐ratings of memory capacity in the evaluation of eyewitness identifications, while also advancing theory on self‐assessments for different memory systems.

The influence of alcohol and weapon presence on eyewitness memory and confidence

Article

Full-text available

Feb 2020
APPL COGNITIVE PSYCH

In this quasi‐experimental field study, bar drinkers (0.00–0.23% blood alcohol content) viewed a photographic sequence in which a male took a laptop from a helpdesk assistant, either on loan or at gunpoint. Following a brief retention period, participants answered 20 multiple‐choice questions about the male, his actions, and details of the scene, then attempted to identify him from a simultaneous target‐present or target‐absent line‐up. Alcohol was associated with a reduction in correct identifications and an increase in false identifications. Surprisingly, the presence of a weapon in the scene enhanced identification accuracy, though wider scene memory was not influenced by alcohol or the weapon. Findings offer some support for the view that alcohol restricts face encoding, perhaps through the narrowing of attention to salient external features (e.g., hair). We also suggest that curiosity about mock‐crime perpetrators may produce weapon focus reversals, although the factors that might elicit such curiosity remain unclear.

Using metamemory measures and memory tests to estimate eyewitness free recall performance

Article

Full-text available

Oct 2019
MEMORY

Using a mock witness methodology, we investigated the predictive value of metamemory measures and objective memory tests as indicators of eyewitness free recall performance. Participants (n = 208) first completed a metamemory assessment that included assessments of self-rated memory capacity, memory development and use of strategies. In a separate session, participants watched a mock-crime video and provided a free recall account, followed by one out of four independent memory tests (i.e., free recall, cued recall, face recognition and general knowledge). Accuracy, amount of details reported, confidence and over/underconfidence in the eyewitness free recall were the main dependent variables. Results indicated three main findings: (1) subjective assessments of memory capacity were not related to eyewitness free recall performance; (2) although individual confidence and over/underconfidence was somewhat stable across different memory tests, accuracy was less stable; and (3) individuals with higher self-rated memory capacity had a slightly stronger confidence-accuracy relation in free recall. These results are discussed with respect to metamemory assessments and performance stability across memory tests of different domains.

Automatic recognition, elimination strategy and familiarity feeling: Cognitive processes predict accuracy from lineup identifications

Article

Feb 2022
CONSCIOUS COGN

Identifying a suspect is critical for successful criminal investigations. Research focused on two decision processes during lineup identification, namely ‘automatic recognition’ and ‘elimination’ strategy, and their relation to identification accuracy. In this article, we report two experiments conducted in France and South Africa, which further examine strategies in eyewitness decision-making. We first used a modified-RSA (Retrospective Self Awareness; Kassin, 1985) interview method to construct questionnaires to have finer-grained measures of cognitive processes involved in lineup identification. Studies 1 and 2 tested the relevance of the questionnaire in each of the countries, and factor analysis yielded three common factors among the countries, namely the expected ‘automatic recognition’, and ‘elimination’ strategies, as well as an additional familiarity related factor. Logistic regressions showed that witnesses who reported using a familiarity feeling and/or an ‘elimination strategy’ to conduct their decision, were less likely to make correct decisions than witnesses who reported using ‘automatic recognition’.

MansourBeaudryNguyenGroncki Preprint forDistribution

Recommended publications

Eyewitness Decision Processes: A Valid Reflector Variable

Reflector variables in augmented reality lineups: Assessing eyewitness identification reliability in...

Eye tracking and eyewitness memory

An Examination of Simultaneous Lineup Identification Decision Processes Using Eye Tracking