Posted: December 28th, 2021

Capstones, Theses and
Dissertations

The use of filler samples moderates the effect of
contextual information on forensic match decisions

TABLE OF CONTENTS
Page
LIST OF FIGURES ……………………………………………………………………………………… iv
LIST OF TABLES……………………………………………………………………………………….. v
ACKNOWLEDGMENTS …………………………………………………………………………….. vii
ABSTRACT……………………………………………………………………………………… viii
CHAPTER 1 INTRODUCTION ……………………………………………………………….. 1
The Problem of Contextual Bias in Forensic Contexts…………………………………. 1
How Contextual Information Influences Judgments ……………………………………. 2
Current Research Addressing Forensic Contextual Bias………………………………. 5
Is There a Solution to the Problem of Contextual Bias in Forensic Examination? 7
Evidence Lineups Versus Evidence Showups…………………………………………….. 8
Predictions Based on Eyewitness Identification and Contextual Bias Literatures 10
CHAPTER 2 METHOD …………………………………………………………………………… 14
Participants and Design……………………………………………………………………………. 14
Materials …………………………………………………………………………………………… 14
Procedure …………………………………………………………………………………………… 17
CHAPTER 3 RESULTS …………………………………………………………………………… 20
Overview of Analyses……………………………………………………………………………… 20
Overview of Results………………………………………………………………………………… 21
Analysis of the Full Multilevel Model……………………………………………………….. 23
Was There a Contextual Bias Effect in the Standard Procedure?…………………… 24
Was There a Contextual Bias Effect in the Filler-Control Procedure?……………. 26
Does the Filler-Control Procedure Decrease “False Alarms” Compared
With the Standard Procedure? ………………………………………………………………….. 28
Does the Filler-Control Procedure Reduce the Number of Correct
iii
Match Decisions Compared with the Standard Procedure?…………………………… 29
Does the Filler-Control Procedure Result in Better, Applied Outcomes? ……….. 31
Is the Increase in d´ from the Filler-Control Procedure due to
Differential Filler Siphoning? …………………………………………………………………… 34
Are the Contextual Bias Effects and the Effect of the Filler-Control
Procedure Reflected in the Confidence Measures? ……………………………………… 37
CHAPTER 4 DISCUSSION……………………………………………………………………… 39
REFERENCES ……………………………………………………………………………………………. 49
APPENDIX A FINGERPRINT SETS ………………………………………………………….. 62
APPENDIX B INSTRUCTIONS………………………………………………………………… 94
APPENDIX C CONTEXTUAL BIAS MATERIALS ……………………………………. 95
APPENDIX D IRB ETHICS APPROVAL…………………………………………………… 103
iv
LIST OF FIGURES
Page
Figure 1 A graphical representation of the multilevel logistic
regression models that used binary sample choice variables as
the dependent measure, with three predictors, and two higher
level grouping variables …………………………………………………………………… 54
Figure 2 A graphical representation of the multilevel model
used in the analyses with participant’s confidence in their decisions
as the dependent measure. ………………………………………………………………… 55
v
LIST OF TABLES
Page
Table 1 A table summarizing of the number of participants
in each between-subjects condition ……………………………………………………. 56
Table 2 Summary of the mean proportion of people in each
between-subjects condition who selected match or no match,
and the mean confidence for each decision in the pilot data………………….. 56
Table 3 A summary of the terminology for the dependent
measures in the logistic multilevel regression analyses. ……………………….. 56
Table 4 Summary of the mean proportion of people in each
between-subjects condition who selected match or no match………………… 57
Table 5 Summary of the mean proportion of people in each
between-subjects condition who selected match or no match,
separated by ambiguity condition………………………………………………………. 57
Table 6 Summary of the mean confidence level of people in
each between-subjects condition who selected match or no match,
separated by procedure and context presence ……………………………………… 58
Table 7 Summary of the mean confidence level of people in each
between-subjects condition who selected match or no match,
separated by procedure, ambiguity condition, and context presence ………. 58
Table 8 Table comparing the d´ values in each procedure, with
and without context, with all fingerprint materials, and then
separated by ambiguity condition………………………………………………………. 59
Table 9 Table showing the intraclass correlation (ICC) values
for Fingerprint Set and Participant grouping variables in the current data. 59
Table 10 A summary of the multilevel models assessing contextual
bias in the data obtained from participants who received the standard
procedure and more ambiguous materials…………………………………………… 60
Table 11 A summary of the two-level logistic multilevel model results
to assess the affects of predictors on each decision type……………………….. 60
vi
Table 12 A summary of the two level logistic multilevel model results
to what predictors influence the confidence level participants’
had in their decisions……………………………………………………………………….. 61
Table 13 A summary of the three-way ANOVA with Context Presence,
Ambiguity Level, and Procedure Type as factors with two levels,
and d´ as the outcome variable. …………………………………………………………. 61
vii
ACKNOWLEDGMENTS
I would like to thank my advisor, Dr. Gary Wells, and my committee members, Dr.
Christian Meissner, and Dr. Stephanie Madon, for their guidance and support throughout the
course of this research. I would also like to thank Dr. Andrew Smith for his help planning
and analyzing this project. Finally, I would like to thank all of the research assistants who
helped to run these experiments, and those who took the time to participate in my experiment
for course credit.
In addition, I would like to thank my partner, Johnie Allen, and my good friends here
at Iowa State University—Kimberley More, Curt More, Nicole Hayes, Rachel Dianiska, and
Dominick Atkinson—for their constant encouragement, offering advice, listening to my
crazy research ideas, and being there when I need some excitement or relaxation. Finally, I
would like to thank my parents, Dr. Neil Quigley and Norine McBride, and my brothers,
Robert and Ian, for putting up with my nonsense and for Skyping me all the way from New
Zealand to remind me that I’ve always been a smarty-pants and a know-it-all and, therefore,
built for graduate school.
viii
ABSTRACT
The criminal justice system is susceptible to errors that can lead to wrongful
conviction of innocent people, sometimes caused by faulty forensic evidence presented at
trial. Among the problems is the fact that contextual information can bias forensic examiners
to make “match” decisions when the materials are ambiguous (Dror, Peron, Hind, &
Charlton, 2005; Dror, Charlton, & Peron, 2006 – Write a paper; Professional research paper writing service – Best essay writers). It is unlikely that contextual information
could ever be eliminated from police investigations and the forensic examination procedure.
Instead, the current experiment suggests that providing examiners with evidence lineups—
analogous to eyewitness identification lineups where the suspect is embedded among similarlooking, known innocent fillers—can reduce the effect of contextual bias. This paper
describes the first experiment conducted to demonstrate the effectiveness of evidence
lineups, called the filler-control procedure (Wells, Wilford, & Smalarz, 2013). Participants
were trained and then examined eight sets of fingerprint materials. The materials were either
more ambiguous or less ambiguous, and some of the sets had an actual match present and
some did not. Furthermore, some participants received the filler-control procedure, and some
the standard procedure—only one comparison print to compare to the crime print, as is
standard in forensic examination procedures. The final manipulation was the presence or
absence of related contextual information, in the form of a police case report suggesting that
the suspect in the case is guilty. The results showed a contextual bias effect in the standard
procedure when the materials were more ambiguous, but only when there was no actual
fingerprint match present. So, the innocent suspect is in the most danger when the materials
are degraded or difficult to compare, and the innocent suspect’s print is the only print
presented to compare to the crime sample. The filler-control procedure, however, eliminated
ix
the effect of contextual information. Although the number of affirmative match decisions
increased when using the filler-control method, these match decisions were spread across the
lineup to the filler prints rather than loading onto the innocent suspect. These results mirror
the results found in eyewitness identification, and show promise for use in the real world as a
means to reduce wrongful conviction and improve forensic testing accuracy.
Keywords: forensics, fingerprints, contextual bias, heuristics, lineups, filler-control
method, evidence lineups.
1
CHAPTER 1. INTRODUCTION
Lana Canen was charged with murder in 2004. The main evidence supporting her
conviction was a latent fingerprint analysis matching her fingerprints to prints found at the
crime scene. A local detective with minimal training in fingerprint examination performed
the analysis and testified that her prints matched those found at the crime scene. This,
combined with confession evidence from another man implicating her as his accomplice,
lead to her eight-year imprisonment for a crime she did not commit. On appeal, the
fingerprints were re-examined and it was discovered that they did not match—even the
original examiner agreed that the prints did not match when the original examiner was
allowed to re-analyze the prints (CBS News, 2014: 2024 – Essay Writing Service. Custom Essay Services Cheap). How does a mistake like this occur? We
know that the criminal justice system is fallible, but law enforcement professionals and the
public view forensic science as reliable and credible. The Innocence Project (Innocence
Project, 2016: 2024 – Do my homework – Help write my assignment online) has exonerated 330 people who were wrongfully convicted and, of these, 155
have involved some form of forensic examination error. Furthermore, these numbers only
represent the cases that have been found and resolved—the problem is likely much more
prevalent (Charman, 2013). There is a need for a systematic investigation of forensic
techniques and potential solutions to the errors seen in forensic examination.
The Problem of Contextual Bias in Forensic Contexts
The National Academy of Sciences (2009) released a report highlighting the need for
more research into forensic examination error rates, their causes, and how to prevent error in
forensic science. Of particular concern in the National Academy report was the impact of
confirmation bias and contextual bias on forensic analysis, which the current study seeks to
address. There is already some literature that speaks to the nature of contextual bias effects
2
and how they arise. To date, most of the empirical research seeking to find a solution to
contextual bias has focused on finding the conditions under which contextual bias occurs,
and then attempting to shield examiners from contextual information (Dror, Peron, Hind, &
Charlton, 2005) or control when the contextual information is revealed (Dror et al., 2015 – Research Paper Writing Help Service;
Dror, 2016: 2024 – Do my homework – Help write my assignment online). The current work, in contrast, assumes that it is almost impossible to fully shield
forensic examiners from contextual information and therefore examines a method for
neutralizing or diluting the impact of contextual information for a class of forensic tests that
constitute “match” or “source” tests. In a match or source test, the examiner is typically
presented with a crime scene sample (e.g., a latent fingerprint, fibers, shoeprint) and a
suspect sample (prints from the suspect, fibers associated with the suspect, shoes of the
suspect) and asked if the suspect sample was the source for the crime sample or if they
“match.” The current study used fingerprints, but the same general principles and findings
should apply to other source or match tests as well.
How Contextual Information Influences Judgments
So, what does the literature tell us about why contextual information might bias
examiners to think that two fingerprints look alike when they are not? The answer lies in
ordinary cognition and decision-making processes. When people make decisions, two kinds
of cognitive processing are used. Bottom-up processing is a data-driven analysis where
details of a stimulus are analyzed in a systematic way, without drawing on any other
information (Chaiken & Maheswaran, 1994). For example, fingerprint examiners use
bottom-up processing when they analyze the pattern of ridges and pores in a fingerprint to
compare to another fingerprint.
3
But bottom-up processing is most useful when the stimuli provided are unambiguous
and there is sufficient time to undergo a detailed analysis of all the material available. As a
result, people often rely on top-down processing or heuristics—making a judgment based the
likelihood of each potential outcome when the resources available are inconclusive (Chaiken
& Maheswaran, 1994; Saks, Risinger, Rosenthal, & Thompson, 2003). Lack of a clear
answer is not the only reason why someone might start to rely on top-down processing, but
these conditions will push people towards heuristic processing. One way heuristics can
operate is by using prior knowledge, beliefs, or expectations to form a base-rate—an idea
about the relative frequency of an outcome within a given set of circumstances (Tversky &
Kahneman, 1974). This kind of processing occurs in situations of uncertainty, when the
available resources are limited, unclear, or there are time constraints, such as a rushed
analysis of a partial fingerprint (Dror, et al., 2005; Neth & Gigerenzer, 2015 – Research Paper Writing Help Service).
Heuristics can be most helpful in ambiguous situations and heuristics often lead to
efficient and accurate decision-making for everyday situations (Neth & Gigerenzer, 2015 – Research Paper Writing Help Service).
But, heuristics can also result in biased or erroneous decisions. For example, if the other
information we draw on to help inform our judgment is inaccurate; our final decision might
also be inaccurate. If people look for evidence to support an expected outcome and ignore the
evidence against that outcome, our final decisions will be biased towards our expectations
(confirmation and contextual bias; Einhorn & Hogarth, 1981; Saks et al., 2003; Tversky &
Kahneman, 1974). Contextual bias in a forensic setting can take on many forms, because
there are many kinds of information that can be interpreted as incriminating. Maybe the
examiner saw the crime described in the paper, with all the evidence against the suspect
described in detail, or the examiner saw the press conference put together by the police on
4
television. What if the examiner overheard at the local hangout after work that the suspect
tried to flee when they were approached initially for questioning? Or maybe the police officer
that brings the evidence to the examiner is highly respected and is fairly sure “this is their
guy”.
To make this idea more concrete, think about the case of Lana Canen again. When the
examiner performed the analysis of the fingerprints, he knew that another person had
confessed and named her as his accomplice. So, the examiner probably did not begin the
examination with a neutral starting point, open to being swayed equally by incriminating or
exculpatory evidence. Rather, the examiner likely began the examination with the
expectation that the prints would probably match, an expectation that could have been guided
by the contextual information about the confession. Subsequently, he might have been more
likely to look for aspects of the prints that confirmed his expectation, and ignore the aspects
of the prints that disconfirmed. In addition, much more disconfirming evidence would have
been required to override the examiner’s expectation that the prints should match once the
examiner had formed the idea (Nickerson, 1998). Expectations can be formed by any number
of different sources—police case reports, communication with police, and media can all
change a fingerprint examiner’s view about the likelihood that a set of prints should match.
This is problematic for the presentation of forensic evidence in court. Forensic
experts are hired to testify about their analysis of the prints using the bottom-up process only.
They are not hired to evaluate the credibility of a confession, the suspicions of the police
investigator, a media slant, or interpret suspect behavior. These are all aspects of the case that
will be analyzed and, if admissible, presented in court by people with that expertise, such as
psychologists, police investigators, and interrogators. If the forensic analyst uses contextual
5
information available to them to inform their decision, their testimony in court may involve
double counting of evidence or be based on inaccurate or inadmissible evidence. Of course,
contextual information can be accurate information, so contextual information could help the
examiner make a correct decision. Nevertheless, contextual information is not for the
examiner to weigh; another expert or a direct witness should be the one to present contextual
information in court if it is probative and admissible. Therefore, contextual information does
not need to be, and should not be, allowed to influence the forensic examiner’s evaluation.
So, we need to find a solution to the problem of contextual bias to protect the independence
of forensic expert testimony at trial.
Current Research Addressing Forensic Contextual Bias
There is research demonstrating that contextual bias is a problem in forensic
fingerprint examination, for ordinary people (Dror, Peron, Hind, & Charlton, 2005; Osborne
& Zajac, 2016: 2024 – Do my homework – Help write my assignment online) and forensic experts (Dror, Charlton, & Peron, 2006 – Write a paper; Professional research paper writing service – Best essay writers). Other forensic
materials have also been used, including handwriting (Kukucka & Kassin, 2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay), bitemarks
(Osborne et al., 2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay), shoe impressions (Kerstholt, Paahuis, & Sjerps, 2007), and ballistics
(Kerstholt et al., 2010 – Essay Writing Service: Write My Essay by Top-Notch Writer). However, the results of studies using materials other than fingerprints
have been mixed, maybe because forensic techniques with fewer known protocols are more
difficult to manipulate in a way that is appropriate for empirical testing, and replicate with
expert participants.
Dror and colleagues (2005) created a paradigm for testing contextual bias in
fingerprints with lay people. Participants were trained briefly to make fingerprint
comparisons. Then, participants determined whether the fingerprints matched. Sometimes the
fingerprints were ambiguous so whether they matched was very unclear, and sometimes the
6
fingerprints were clearer meaning that people could be more certain that they matched or did
not match. In addition, sometimes participants made the decision with the help of additional,
contextual information, and sometimes there was no extra information. There were four
contextual information conditions: people received no context, photos with low emotion
content (e.g. a hammer), photos with high emotional content (e.g. a bloody crime scene), and
subliminal priming of emotional content paired with high emotion photos. Dror and
colleagues found that participants made significantly more match decisions for pairs of
fingerprints that were accompanied by highly emotional images that suggested incrimination.
However, this pattern was only found when the fingerprints were poor quality, rendering the
decision more ambiguous and uncertain. Similar patterns have been found in more recent
studies with ordinary people making judgments about pairs of fingerprints (Langenburg,
Champod, & Wertheimer, 2009), and in a replication of Dror and colleagues’ study (Osborne
& Zajac, 2016: 2024 – Do my homework – Help write my assignment online).
There is an obvious concern that arises from using lay people rather than experts and
the concern relates to how comparable the results will be and whether undergraduate data is
generalizable to experts. Experts are better able to discriminate between similar fingerprints
(Thompson & Tangen, 2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay), but novice examiners tend to be no better than lay people at
matching fingerprints (Thompson, Tangen, & McCarthy, 2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay), and lay people can
discriminate between prints at an above-chance level (Vokey, Tangen, & Cole, 2009).
Although there may be differences between experts and novices in their ability to perform
fingerprint analysis and maybe even differences in contextual bias susceptibility, contextual
bias effects appear to be robust to expertise level. For example, Dror and colleagues (2006 – Write a paper; Professional research paper writing service – Best essay writers)
presented five experts with sets of fingerprints that they had determined were a match in
7
previous cases. However, this time the researchers told the experts that these prints were
from a high profile case involving the FBI and Brandon Mayfield. The experts were familiar
with the case and therefore knew that if the prints were from this case, they should not match.
Dror and colleagues found that only one expert determined the prints to be a match now, as
they had in the past. The remaining experts all said that the prints either did not match now,
or that the crime sample was too degraded to decide. These results also show that contextual
bias can work against finding a match when the context suggests that a match is unlikely.
Dror and Charlton (2006 – Write a paper; Professional research paper writing service – Best essay writers) demonstrated similar results with a different group of
experts, but this time there was a control group where examiners were shielded from
additional contextual information. Expert fingerprint examiners were asked to assess
fingerprint materials from eight past cases. The examiners had judged half of these past cases
as individualizations (they were a match), or exclusions (they were not a match). For the
study, the examiners either received no contextual information (4 cases), context suggesting
the prints should match (Incriminating evidence; 2 cases), or context suggest the prints
should not match (Exculpatory evidence; 2 cases) along with each set of prints. Exculpatory
evidence was found to influence fingerprint experts by making the examiner’s decisions
more conservative. In three cases where the examiners had said the prints matched in the
past, the exculpatory evidence lead to the examiners to conclude that the prints did not match
now, and in one case the examiner said the materials were inconclusive. However, there was
no effect of incriminating evidence on expert decision-making found in this study, and in two
cases the experts made a decision inconsistent with their past determination in the absence of
context. So, inconsistencies can occur even without the influence of context.
8
Is There a Solution to the Problem of Contextual Bias in Forensic Examination?
What have researchers recommended to combat contextual bias in forensic contexts?
In the empirical papers on this issue of forensically-relevant contextual bias, authors have
typically suggested shielding forensic examiners from contextual information (Dror et al.,
2005) or gradually introducing levels of contextual information to examiners (Dror et al.,
2015 – Research Paper Writing Help Service; Dror, 2016: 2024 – Do my homework – Help write my assignment online). But an examiner can never be totally insulated from contextual
information. Explicit exposure to contextual information through police communication or
case information included when the evidence is handed over is not the only form of biasing
information. Some forms of contextual information are almost impossible to prevent.
Consider, for example, that forensic examiners are members of the community and are likely
to be exposed to media reports on crimes in their area. Evidence from these cases may end up
on their desk for examination. In addition, forensic experts and police tend to socialize in the
same circles, as well as with each other. In fact, even the presentation of a single sample to
be compared with the crime sample suggests that there is good evidence that the prints from
this person should match the crime print (Wells, Wilford, & Smalarz, 2013).
Evidence Lineups Versus Evidence Showups
In this study, I tested a different type of potential solution to the problem of
contextual bias—one that is designed to moderate the effect of contextual bias, while
accepting that contextual information will always be available to examiners. Instead of
shielding examiners from contextual information, Wells, Wilford, and Smalarz (2013)
proposed the use of evidence lineups as a way to dilute the effect of bias. This idea draws on
the already well-developed research in eyewitness identification that seeks to reduce the
9
chances that innocent people who become suspects in an investigation will be mistakenly
identified by an eyewitness. One of the main ideas to come out of eyewitness research is the
idea that a lineup is more protective of innocent suspects than is a showup (Steblay, Dysart,
Fulero, and Lindsay, 2003). A showup is where an eyewitness is shown a single individual,
who is a suspect, and asked to determine whether they are the culprit. A showup would be
equivalent to the current, standard procedure for forensic examination—the crime sample is
presented with a sample obtained from a single suspect and the examiner is asked to decide if
the samples are a match. A lineup is different from a showup because a lineup embeds the
suspect among other people who, although known to be innocent, fit the description of the
culprit that was obtained from the eyewitness (Wells, 1993). These non-suspect lineup
members are called lineup fillers. So, now the test is not simply whether the eyewitness can
tell if the individual is similar to the culprit, but rather the eyewitness needs to be able to pick
the culprit out of a number of people who could plausibly be the culprit. Importantly, if the
eyewitness picks someone from the lineup who is known to be innocent (a filler), there are
no incriminating consequences of this incorrect identification for the filler. After all, filler are
known-innocents in a lineup.
How would lineups work in a forensic context such as with fingerprints materials? If
the suspect sample is embedded in a lineup of other highly-similar samples, contextual
information still cannot tell the examiner which of the samples is a match to the crime
sample. Although the contextual information can raise expectations that one of the samples
should be a match to the crime sample, contextual information cannot point the examiner to
any one sample if the filler-control method is used. Thus, the examiner cannot simply use the
contextual information and instead must perform a bottom-up analysis of the prints. In fact,
10
an expert examiner may simply decide not to rely on the contextual information at all
because it is largely useless with respect to the task at hand. This aspect of the lineup
procedure is what makes this solution qualitatively different from the other recommendations
that attempt to shield examiners from contextual information. An evidence lineup does not
rely on contextual information being hidden from examiners, or examiners using their
“willpower” to be objective. Instead, the use of fillers should tend to neutralize or dilute the
contextual information due to the fact that the contextual information is not specific to one of
the samples but instead applies to the set of samples as a whole.
To test the viability of the filler-control procedure, I had undergraduate participants
learn about fingerprint examination, and then decide whether a single fingerprint matched a
crime print (standard forensic procedure), or whether one of six fingerprints matched a crime
print (filler-control procedure). Sometimes, the prints were presented with an incriminating
police case report, and sometimes they were not.
Predictions Based on Eyewitness Identification and Contextual Bias Literatures
The first prediction was that the standard (i.e., no fillers) forensic match procedure
would show a contextual bias effect. That is, there would be significantly more affirmative
match decisions made by participants when they received incriminating contextual
information prior to examining the fingerprint materials than when they did not receive such
contextual information. Also, this effect of incriminating contextual information should be
most pronounced when the prints are more ambiguous. Under ambiguous situations the
bottom-up analysis does not give a clear answer, and so people are more susceptible to
influence from top-down processes (Chaiken & Maheswaran, 1994). This pattern of results
11
would conceptually replicate experiments completed by other research laboratories using
fingerprint materials (Dror et al., 2005; Zajac & Osborne, 2016: 2024 – Do my homework – Help write my assignment online).
A number of additional hypotheses were derived from the eyewitness identification
literature regarding lineups and showups because of their close analogy to the filler-control
procedure and the standard procedure, respectively, in a forensic match test. First, the
eyewitness identification literature shows more affirmative choosing for lineups than for
showups. This is due to the fact that there are more faces that could potentially resemble an
eyewitness’s memory of the culprit when viewing a lineup than when viewing a showup.
Similarly, it was predicted that there would be more affirmative match decisions for the
fingerprint lineup than for the fingerprint showup due to the fact that there are more possible
prints to resemble the latent print.
The eyewitness literature shows, however, that this higher rate of affirmative
responding for lineups than for showups does not result in more mistaken affirmative
responses on the innocent suspect. This is because, although there is more choosing for
lineups because there are more options, the innocent suspect is no longer the only plausible
choice. In fact, there are a number of other fillers that match the description of the culprit as
well. Because the innocent suspect is not actually the culprit and therefore not a great match
to the eyewitness’ memory, choosing will spread out to the fillers, thereby reducing the false
positives on the innocent suspect to a level that is significantly lower than the rate observed
for showups (filler siphoning; Wells, Smalarz, & Smith, 2015 – Research Paper Writing Help Service; Wells, Smith, & Smalarz,
2015 – Research Paper Writing Help Service). Fillers will also siphon some positive identifications away from the actual culprit, but
to a lesser extent because the culprit is a good match to memory. This phenomenon is called
differential filler siphoning as the fillers have a differential effect contingent on whether the
12
actual culprit is in the lineup. Good fillers will siphon away from an innocent suspect more
than they will from a guilty suspect, and this is the mechanism through which the ratio of
innocent suspect identifications and actual culprit identifications (as well as d´ values)
improves when lineups are used. Based on this consistent result in the eyewitness literature, it
was predicted that this same differential filler siphoning would occur when comparing
fingerprint lineups to fingerprint showups.
Another hypothesis for the current study was that contextual bias effects would be
diluted in the filler-control procedure when compared to the standard procedure. In effect,
this dilution prediction is closely related to the idea of filler siphoning. For example, if
incriminating contextual information increases false affirmative match decisions by 12%,
then the entire 12% increase would fall on the innocent suspect sample for the standard
(showup type) procedure. For the filler-control procedure, however, the 12% increase in false
affirmative responding that results from incriminating contextual information would dilute
(spread) across the six samples, producing (on average) a mere 2% increase in false
affirmative match decisions on the innocent suspect. An alternative hypothesis was that
contextual information would have little or no effect at all on affirmative match decisions
when using the filler-control method. This is because, although the contextual information
suggests to the examiner that there should be a match, contextual information does not assist
the examiner at all on being able to determine which of the six samples matches the crime
sample. Accordingly, the contextual information does not relieve any of the examiner’s
burden of relying as much as possible on the bottom-up approach. Hence, when given the
fingerprint lineup (rather than the fingerprint showup) the examiner might simply dismiss the
13
contextual information as being irrelevant or unhelpful and rely almost totally on
characteristics of the prints themselves.
Finally, predictions were made about the confidence that the examiners expressed in
their decisions. First, confidence should be overall lower in the filler-control procedure
because the task is more difficult. Furthermore, when people make an incorrect decision,
confidence should be lower compared with when they make a correct decision. This is
expected because there is typically a confidence-accuracy relation seen in eyewitness
identification studies (Wixted & Wells, 2017) as well as other tasks for which people
perform above chance levels. Also, the more ambiguous materials should result in lower
confidence in the decisions too—this hypothesis functions as a manipulation check as well.
Furthermore, when people make a decision that is incongruent with the suggestion in the
contextual information (e.g. the context implies guilt but the participant says there is no
match), confidence should be reduced compared with when the contextual information agrees
with their match decision. Additionally, it was anticipated that there may be stronger
evidence of contextual bias in the confidence measure rather than the binary match decision
because the measure is much more sensitive (scores ranging from 0 to 100% confidence
compared with a two-option forced choice measure).
14
CHAPTER 2. METHOD
Participants and Design
A total of 244 undergraduate participants from a Midwestern University took part in
the study for partial course credit. All participants were fluent English speakers and over the
age of 18 years. Nine participants were excluded from analyses due to experimenter error, or
unusual participant behavior. There were four independent variables in this study: what
procedure was used, whether contextual information was provided, how difficult the task
was, and whether or not one of the samples actually did match the crime sample. The design
was a 2 (context: context vs. no context) x 2 (procedure: standard vs. filler-control method) x
2 (ambiguity: more ambiguous vs. less ambiguous) x 2 (match presence: match present vs.
match absent) mixed factorial model.
Context, procedure, and ambiguity were manipulated between subjects and match
presence manipulated within subjects. Participants were randomly assigned to ambiguity,
procedure, and context conditions, and half of the fingerprints that each participant saw
matched, and half did not, presented in a random order. Refer to Table 1 for a breakdown of
the numbers in the between subjects groups.
Materials
The session began with a training video on fingerprint analysis (Introduction to
Fingerprint Analysis, 8:00), created using information about FBI standards and training for
fingerprint analysis. The video consists of a series of informative Power Point slides with a
voice over and contains information about the background of fingerprint analysis, how an
analysis is performed, and then a number of working examples for the participants.
Participants watched the video on the computer screen with headphones on. The training
15
video and response questions were all presented on a desktop computer using MediaLab
software.
The fingerprint samples were from a previous study by Marcon (2009), using
fingerprints from 125 students at the University of Texas at El Paso. There were three
different levels of print quality. Rolled fingerprints include the entire print from the tip of the
finger and are rolled slowly and deliberately from one side of the finger to the other to ensure
the print is clear. Plain fingerprints are not rolled, so they do not include as much detail from
the sides of the finger and can sometimes be unclear or smudged. Partial fingerprints, or
latent fingerprints, occur when someone quickly touches a surface, not attempt to leave a
deliberate print. Partial fingerprints typically lack detail, are smudged, or unclear. Marcon
also had 60 undergraduates rate how distinct and typical each print is from all the other
prints. The final library of fingerprint sets that are rated consists of fingerprints from 113
undergraduates.
Crime scene samples in the more ambiguous condition were drawn from the less clear
plain prints, or partial print samples. In the less ambiguous condition, in contrast, the crime
prints were drawn from rolled or plain prints, and were complete, with sharper lines and no
smudging. Fillers samples for the filler-control procedure were selected based whether they
had the same fingerprint pattern (loop, whorl, or arch) and came from the same finger (index,
thumb, etc). In the more ambiguous condition, the filler samples were also pulled from
partial or less clear plain fingerprint samples, whereas the less ambiguous condition the
fillers are clearer. The fingerprints were selected so that the fingerprints were representative
of the range of typicality and distinctiveness ratings collected by Marcon (2009).
16
Each set of fingerprints consisted of one crime sample, and then a different
impression of the print from that same finger was used as the “match” sample. Then, six
other fingerprints were selected with the same type of pattern, from the same finger, with the
same level of clarity at the matching sample. Then, from these six fingerprints, one was
randomly selected to be the “innocent suspect” print. This print appeared in the standard
procedure for that fingerprint set as the “no match” sample. In the filler-control procedure,
the matching sample or the “innocent suspect” print was embedded amongst the other five
fingerprints which were ordered randomly in to a 2 by 3 lineup configuration. Refer to
Appendix A for all of the fingerprint materials used in this experiment. Prior to running the
full experiment, the filler-control materials were tested on a smaller sample of participants (N
= 68) with contextual information (the condition that would be the most time consuming) to
determine how much time participant’s would take to make match decisions concerning four
cases, ensure that there were no ceiling or floor effects, ensure that the filler samples were
plausible options, and test whether the more ambiguous condition was more difficult. See
Table 2 for a break down of these pilot data with the 16 sets of fingerprints in the fillercontrol method, eight of which were more ambiguous.
The average amount of time taken to run through the instructions and the training
video, complete four cases with evidence lineups and a police case report (contextual
information) to read, and receive the oral debriefing was 29 minutes (rounded to the nearest
minute). Therefore, it was concluded that it was feasible for each participant to analyze eight
cases in an hour to increase the power of the study. There were no ceiling or floor effects—
when there was an actual match present, participants matched that print to the crime sample
approximately half of the time (50% of the time for less ambiguous materials, and 44% for
17
more ambiguous materials). When no match was present, people were picking the innocent
suspect sample some of the time (15% of the time for less ambiguous materials, and 10% for
more ambiguous materials). These proportions were also consistent with expectations
regarding the ambiguity level—there were less correct matches in the more ambiguous
materials, and people’s confidence in their match decisions were lower in the more
ambiguous materials condition (72.5% confident for less ambiguous materials, and 65.5% in
the more ambiguous materials). Finally, participants were matching filler samples to the
crime print some of the time, and more often when there was no actual match present (27%
fillers when there was a match present, and 55% of the time when there was not), which is
what would be expected according to the eyewitness literature (Wells, Smith, & Smalarz,
2015 – Research Paper Writing Help Service; Wells, Smalarz, & Smith, 2015 – Research Paper Writing Help Service). When people picked fillers, choosing was spread
across the lineups, rather than one of the fillers standing out in the lineup. Therefore, these
materials appeared to be appropriate for testing the hypotheses in the current study.
Procedure
Participants signed up for the study using an online recruitment system, and came into
the laboratory to complete the experiment. The session began with the experimenter briefly
describing the task and providing a consent form that the participant could ask questions
about and sign. First, participants were instructed to put on their headphones and pay
attention to the training video presented on the computer screen. Next, participants read
through a series of instructions (Refer to the Appendix B for the complete instructions). They
were told that these materials are from real cases and that each case has a suspect, who may
or may not be the actual perpetrator. Participants were told that they would see materials
from eight real cases, to be analyzed one at a time. For each case, participants were asked to
18
make an assessment about whether or not the comparison print(s) matched the crime print,
and how confident they were in their decision. Participants were asked if they had any
questions about the task. If they did not, they were given a scientific-grade magnifying glass
with an LED light to aid their examination and more closely mimic a real expert’s
experience.
When the task began, participants would either see fingerprint sets that were all from
the more ambiguous materials (eight total), or fingerprint sets from the less ambiguous
materials (eight total), always randomly ordered. Half of these always had a match present,
and half always had a match absent, and this was also randomly ordered. Those who were
assigned to the condition where they were provided with contextual information were told in
the instructions that they would also receive extra information in the form of a police case.
Each case report was given to the participants for them to read prior to examining each set of
fingerprint materials. There were eight different police case reports containing details about
different types of crimes—kidnapping, extortion, armed robbery, bomb threat, homicide,
rape, arson, and identity theft. Each report was highly suggestive of guilt, for example one
reported that DNA evidence found under the victims nails was a match to the suspect. We
randomly ordered each of the eight police case reports to be presented with any one of the
eight fingerprint sets each person in the context condition received. Refer to Appendix C for
all of the case reports used in this experiment. If they were not assigned to the contextual
information group, they never saw a police case report.
The cases were handed to the participants one-by-one by the experimenter in a manila
folder, with the fingerprint materials enclosed, laminated, and labeled as if it were real
evidence. Participants responded to questions about the materials on the computer screen
19
about each case. Participants who were assigned to the standard procedure received one
comparison print for each case. For these participants, the instructions clearly stated that,
although there is a suspect in each case, this does not mean that they are the person who
committed the crime—the fingerprint may or may not match the crime print. They were
asked to respond to the question “Do the samples match?” with a “Yes” or a “No”. For the
participants who were assigned to the filler-control procedure, they received a lineup of six
comparison prints in each pack. Participants in the filler-control condition, like those in the
standard condition, were told that although there is a suspect in each case, this does not mean
that they are the person who committed the crime. However, participants were also told that
they would see six sample fingerprints for each case and that five of the samples they will see
were definitely not matches. So, if participants select one of the filler fingerprints then they
have selected an innocent person’s prints. Filler-control participants were asked “Does one of
these samples match the crime sample?” If participants responded yes, they needed to then
indicate which of the samples was a match (the samples were numbered on the sheet “Sample
1” through to “Sample 6”). Finally, for all cases, participants were asked “How confident are
you in your decision in this case?” and rated their confidence on an 11-point scale from 0%
(point “1”; not at all confident) to 100% (point 11; completely confident). The computer
prompted them after each confidence rating to ask for the next set of case materials from the
experimenter and the questions on screen were repeated. Once the participants had completed
their judgments for all eight cases, participants were told that is the end of the experiment
and the experimenter thanked the participant and gave them an oral debriefing.
20
CHAPTER 3. RESULTS
Overview of Analyses
Multilevel model analyses, using the LMER package in R Studio, were used to
analyze these data to account for the nested nature of the design—each observation was
nested within Fingerprint Set (N = 64) and each person assessed eight different fingerprint
sets, so observations were also nested within Participant (N = 234). These two grouping
variables were included at Level 2 in the multilevel models, and the relevant intraclass
correlations can be found in Table 9. The predictors entered in the model were Ambiguity
Level (more ambiguous = 1, less ambiguous = 0), Context Presence (context present = 1, no
context = 0), Procedure Type (standard procedure = 0, filler-control procedure = 1), and
Match Presence (match present = 1, match absent = 0). Finally, the Context Type was
entered as a categorical control variable to allow for any variation due to the eight different
case reports used as contextual information here. Each trial or case analyzed represented one
data point in these analyses, and each participant provided eight data points. The use of
Participant as a Level 2 grouping variable accounted for the repeated measures nature of
these data, and any variation that could be attributed to a single person’s unique biases and
perceptions was allowed for in the model (see Figures 1 and 2 for graphical representations
of the models).
For the full model of each analysis, a marginal R2 is reported, which is the amount of
variation in the outcome explained by the predictors in the model, and the conditional R2 is
also reported which represents the amount of variation in the outcome explained by all
variables in the model, including higher-level, grouping variables. The dependent measure
was determined by the research question being addressed. Any outcome variables associated
21
with participants’ match decisions were binary, so these models used logistic multilevel
regression models. A full breakdown of the terminology used to describe the different
outcome decisions can be found in Table 3. Affirmative “Match” decisions can be either
correct (“Hits”) or incorrect (“False alarms”), and different effects were hypothesized for
these two kinds of “match” decision. These were coded as binary variables—the presence of
a “hit” or “false alarm” is coded as “1” and absence coded as “0”. False alarms can also fall
on a filler sample in the filler-control procedure, but unless specifically mentioned, fillers are
not included in the analyses. “Choosing” refers to “Match” (coded 1) decisions, as oppose to
“No Match” (coded 0) decisions, and includes situations where people match the crime
sample to a filler sample in the filler-control procedure. “Suspect choosing” refers to a
“match” decision where the correct sample is selected (“hits”), or the innocent suspect’s print
is selected (“false alarms”). “No match” decision can be incorrect (a “miss”) or correct (a
“correct rejection”). The confidence measure will also feature as the dependent measure in
some analyses, and because confidence is a continuous measure, logistic regression analyses
were not used when analyzing the confidence measure.
Overview of Results
There were a number of interesting results in these data that were consistent with my
predictions. First, the context bias effects found in previous work (e.g. Dror et al., 2005) were
replicated under the specific conditions that were anticipated to produce the strongest effects.
Specifically, the presence of incriminating contextual information resulted in significantly
more false alarms when the materials were very ambiguous and the standard procedure (only
one sample was compared to the crime sample) was used. In contrast, there was no evidence
of contextual bias when hits were assessed, or when the fingerprint materials were less
22
ambiguous. Furthermore, no contextual bias effects were observed in the filler-control
conditions. That is, whether or not incriminating contextual information was presented to
participants, there was no significant change in hit rates or false alarm rates in the fillercontrol procedure.
Interestingly, the filler-control procedure appeared to be superior to the standard
procedure even for participants in the no-context conditions. Although the filler-control
procedure reduce the rate of hits compared to the standard procedure, the filler-control
procedure resulted in a much larger drop in false alarms compared with the reduction in hits.
In addition, there was no significant increase in “no match” decisions in the filler-control
procedure. In fact, there was no significant difference in either the number of misses or in the
number of correct rejections in the filler-control when compared to the standard procedure.
So, fingerprint lineups did not result in people to choosing to “back off” and refrain from
saying there was a match compared to the standard procedure. Instead, the filler-control
procedure led these match decisions to land on filler samples rather than suspect samples,
particularly when the evidence lineup did not contain an actual match. Therefore, these data
show evidence of the differential filler siphoning mechanism that occurs in eyewitness
lineups—good fillers surrounding a suspect will draw choices away from the suspect, but
will draw proportionately more choices away from an innocent suspect than from a guilty
one (Wells, Smalarz, & Smith, 2015 – Research Paper Writing Help Service; Wells, Smith, & Smalarz, 2015 – Research Paper Writing Help Service).
There was no support for the hypothesis that people’s confidence in their decisions
would be boosted by contextual information when that information was consistent with their
judgments (i.e. the participants says there is a match when the contextual information also
suggests that there should be a match). Nevertheless, the confidence measure did serve as a
23
useful manipulation check to see if people found the filler-control procedure more difficult
and whether they found the match in the more ambiguous materials less clear. Now, the data
analyses, beginning with a full multilevel model of all the data and predictors, will be
presented. I will explain how these results were obtained, and what they mean in a question
and answer format with each question representing the hypotheses of the current study.
Analysis of the Full Multilevel Model
It was predicted that contextual bias effects would be found in the standard procedure,
particularly when the materials were ambiguous, but these effects would be reduced in the
filler-control procedure. Statistically speaking, a contextual bias effect would be
characterized by a significant increase in “match” decisions, and this increase should be
largest for “false alarms” (incorrect “match” decisions). Ideally, to show support for this
hypothesis, a significant four-way interaction between Match Presence, Ambiguity Level,
Context Presence, and Procedure Type should be demonstrated.
To test for the four-way interaction, a model was run with “suspect choosing” as the
outcome variable (so that filler fingerprint samples were excluded from the analysis), and all
possible interactions between Match Presence, Ambiguity Level, Context Presence, and
Procedure Type included. In addition, Context Type was included as a control variable, and
the data was nested within Participant, and Fingerprint Set. The model failed to converge1
. In
an attempt to make the model merge, the model was reduced by removing Context Type as a
control variable, given that there was no significant impact of any types of contextual
information on suspect choosing in the full model. In addition, all the interactions that were
1 29% of the variation in “false alarms” is explained by the predictors and interactions (Marginal R2 = 0.285)
and 43% of the variation in “false alarms” is explained by all predictors in the model, including higher-level
grouping variables (Conditional R2 = 0.428). Goodness of Fit: χ
2 (8) = 24.785, p=0.002.
24
not approaching significance (p<.10) were removed from the analysis. But the model still did
not converge.
At this point, I concluded this full model was not going to be useful for answering the
research questions of this experiment. Clearly with this many predictors, a control variable
with eight categories, and a binary outcome, the model is not going to converge, so
interpreting any effects from the full model is not appropriate. The validity of the full model
is questionable and, therefore, should not be interpreted. Hence, the data analysis strategy
was driven by specific tests of the a priori hypothesis that motivated this research. These
predictions were sound, and based on previous literature in the forensic psychology and
eyewitness psychology domain. The following statistical analyses address each of the
hypotheses driving this research.
Was There a Contextual Bias Effect in the Standard Procedure?
The first hypothesis was that the contextual bias effects found in previous research
would be replicated (for example, Dror et al., 2005). When contextual information indicates
that a particular outcome is more likely, people are more likely to make decisions consistent
with the expectation created by that information (Chaiken & Maheswaran, 1994; Saks et al.,
2003). The case reports were highly suggestive of guilt, which should have created an
expectation that the prints should match. Accordingly I hypothesized that participants who
received the standard procedure would make more affirmative “match” decisions—hits and
false alarms—when an incriminating police case report was presented with the fingerprints.
This effect was expected to be most prominent when the materials were more ambiguous (as
was found in Dror et al., 2005), and when there was no actual match present (supported by
eyewitness identification research e.g. Steblay, Dysart, Fulero, and Lindsay, 2003).
25
All standard procedure data. Initially a model was run using the data from
participant who received the standard procedure. Ambiguity Level (more ambiguous = 1,
less ambiguous = 0) and Context Presence (context presence = 1, context absent = 0) were
entered into the model as predictors. The initial analysis with “Suspect choosing”
(culprit/innocent suspect print chosen = 1, any other decision = 0) as the outcome variable in
this model, and Match Presence (true match = 1, no match = 0) included as a predictor
suffered from similar issues to the full model and failed to converge. Accordingly, hits (hit =
1, any other decision = 0) and false alarms (false alarm = 1, any other decision = 0) were
assessed as outcome variables in separate models so that the Match Presence factor could be
removed from the analysis. In both models, the effect of Context Presence (hits: B = 0.099, p
= 0.767, false alarms: B = 0.054, p = 0.897), Ambiguity Level (hits: B = -0.727, p =0.519;
false alarms: B = -0.293, p = 0.767), and the interaction (hits: B = -0.076, p = 0.862, false
alarms: B = 0.916, p = 0.117) did not reach significance. This was not surprising, because
contextual bias effects were expected to be strongest for more ambiguous materials and for
false alarms, which is only a small subset of these data. Therefore, the data were split into
two subsets—people who received the less ambiguous materials, and people who received
the more ambiguous materials.
Separate analyses of more and less ambiguous materials. When the data from the
participants who received the less ambiguous materials were examined separately, there was
no effect of context on hits (B = 0.666, p = 0.355), or false alarms (B = 0.577, p = 0.531). So,
as predicted, I concluded that there is no evidence of contextual bias in the data from those
who received the less ambiguous materials. Next, the data from participants who received the
more ambiguous fingerprint materials were analyzed. Again, hits and false alarms were
26
assessed separately as outcome variable. As expected based on the proportions observed in
Table 5, there was no significant effect of context on hits (B = 0.032, p = 0.916). However,
there was a significant effect of context on false alarms (B = 0.769, p = 0.018)2
. Refer to
Table 10 for a summary of these results.
Conclusion. There was evidence of contextual bias in these data, but the effect of
contextual bias only occurred under very specific circumstances—the circumstances that
were predicted to be the most conducive to a contextual bias effect based on previous
literature. When materials were very ambiguous, and there was no actual match present,
contextual information influenced the number of match decisions in the standard procedure
for forensic examination. In other words, when people received incriminating contextual
information, there were significantly more incorrect match decisions (42% match decisions)
made compared with when participants received no contextual information (25% incorrect
match decisions). There was a striking 17% difference due to the presence of contextual
information, which was statistically significant. Refer to Tables 4 and 5 for a summary of
proportions of participants who made each type of decision separated by procedure,
ambiguity, and context conditions.
Was There a Contextual Bias Effect in the Filler-Control Procedure?
Having established the conditions under which contextual bias occurs in the standard
procedure, the next hypothesis to address was that the filler-control procedure would
moderate the contextual bias effect seen in the standard procedure. That is, it was anticipated
that there would be a smaller difference or no difference in affirmative “match” decisions
when there was context present versus context absent in the filler-control procedure. If there
2 1% of the variation in “suspect choosing” is explained by the predictors and interactions (Marginal R2 =
0.008) and 82% of the variation in “suspect choosing” is explained by all predictors in the model, including
higher-level grouping variables (Conditional R2 = 0.816). Goodness of Fit: χ
2 (8) =3.052, p=.931 (good fit).
27
was no significant main effect of context on “suspect choosing” decisions (hits and false
alarms), this hypothesis would be supported by these data.
All filler-control procedure data. As for the analysis in the previous section, a
model was run with “suspect choosing” as the outcome variable with only the data obtained
from participant who received the filler-control procedure. All possible interactions between
Match Presence, Context Presence, and Ambiguity Level were included, as well as Context
Type as a control variable. Fingerprint Set and Participant were included as higher-level
grouping variables. The full model failed to converge, and there were no significant
interactions3
. When, hits and false alarms were examined separately, there were no
significant effects (Context Presence, Ambiguity Level, and the two-way interaction).
However, because the contextual bias effects were only found under very specific conditions
in the standard procedure, the data were split into less and more ambiguous subsets to
confirm there was no evidence of contextual bias.
Separate analyses of more and less ambiguous materials. As anticipated, there
were no significant effects of Context Presence for hit rates (B = 0.339, p = 0.748), or false
alarm rates (B = -0.163, p = 0.721) when the materials were less ambiguous. Therefore,
contextual bias did not appear to have an impact on less ambiguous materials, whether or not
the standard procedure or an evidence lineup was used. But any contextual bias effects in the
filler-control procedure were most likely to be found under circumstances where the evidence
lineup decision is ambiguous, as was found in the standard procedure. The more ambiguous
materials were analyzed next to see whether there was any evidence of contextual bias. Was
there any evidence of an effect of context presence? Was the effect of context presence
3 26% of the variation in “suspect choosing” is explained by the predictors and interactions (Marginal R2 =
0.258) and 43% of the variation in “suspect choosing” is explained by all predictors in the model, including
higher-level grouping variables (Conditional R2 = 0.426). Goodness of Fit: χ
2 (8) = 11.52, p=0.174 (good fit).
28
reduced compared to the same conditions in the standard procedure? When the model was
run, there was no significant effect of context on hits (B = -0.221, p = 0.486) or false alarms
(B = 0.140, p = 0.753) when the fingerprint materials were very ambiguous.
Conclusion. There was no evidence of contextual bias in the filler-control procedure.
The hypothesis that evidence lineups would reduce the influence of contextual information
on affirmative “match” decisions was supported. In fact, the contextual bias affect appeared
to be totally eliminated for false alarms (10% false alarm rate for both context present and
context absent conditions), and hits actually appeared to reduce in the context condition
compared with the no context condition (40% versus 47% respectively; refer to Table 5).
Does the Filler-Control Procedure Decrease “False Alarms” Compared with the
Standard Procedure?
Some of the analyses have already indicated that the filler-control procedure reduces
the number of “false alarms” that occur, for example the filler-control procedure eliminated
the boost in false alarms that occurred in the presence of incriminating contextual
information. The observed proportions, found in Tables 4 and 5, also show that the fillercontrol procedure seems to consistently reduce the average false alarm rate to approximately
8 to 10%, regardless of ambiguity of the materials and whether context is present. False
alarm rates in the standard procedure are consistently higher in these data—an average of
33.5% of the time, which is 23.5% higher than that highest false alarm rate in the fillercontrol procedure. So, the following analyses examined the significance of this observed
difference.
First, a model was run with false alarms as the outcome variable (false alarm = 1, any
other decision = 0) and all possible interactions between Context Presence, Ambiguity Level,
29
and Procedure Type, and with data from both procedure types. The model failed to converge,
so the model was reduced to only the main effects. In this final, reduced model, there was
only one significant effect: Procedure Type (B = -1.905, p = 0.011)4
. That is, the standard
procedure resulted in significantly more false alarms overall (M = 33.5%) than did the fillcontrol procedure (M = 9.25%), as was hypothesized. None of the other variables in the
model (Ambiguity Level, or Context Presence) had a significant effect false alarms overall.
Refer to Table 11 for a summary of these results.
Does the Filler-control Procedure Reduce the Number of Correct Match Decisions
Compared with the Standard Procedure?
I expected that the filler-control method might also result in a reduction in correct
match decisions (hits) because this is the typical pattern observed in lineups when they are
compared to showups. Specifically, in eyewitness identification research showups result in
more innocent suspect identifications, but also more correct identifications, when compared
with lineups (Steblay, Dysart, Fulero, & Lindsay, 2003). Did this pattern also hold in the
current experiment? The full model run on “suspect choosing” could have helped to address
this question, but it failed to converge. Therefore, a multilevel model was run with hits (hit =
1, any other decision = 0) as the outcome variable, and data from both procedure types
included in the model. Procedure Type, Context Presence, and Ambiguity Level were
included as predictors and Fingerprint Set and Participant were, again, included as a higherlevel grouping variables. However, none of the predictors resulted in significant effects. In
particular, there was no evidence of a significant difference in hits based on procedure type
4 8% of the variation in “false alarms” is explained by the predictors and interactions (Marginal R2 = 0.081) and
71% of the variation in “false alarms” is explained by all predictors in the model, including higher-level
grouping variables (Conditional R2 = 0.713). Goodness of Fit: χ
2 (8) = 19.203, p=0.014.
30
(B = -0.990, p = 0.392), even when interactions and the control variable were excluded. Refer
to Table 11 for a summary of these results.
Even though the multilevel analysis did not show a significant difference in hit rates,
the observed proportions clearly show that the hit rate is reduced in the filler-control
procedure. In the filler-control procedure there was a mean percentage of 44.5% of
participants correctly matching the suspect print to the crime print, which was significantly
fewer than the 65% of participants correctly matching the suspect print to the crime print in
the standard procedure (20.5% decrease overall; %; χ
2 = 39.98, p<0.001, 95% CI [14.07%,
26.74%]). The presence of context did not change this pattern, but the ambiguity of materials
had a small impact. The more ambiguous materials resulted in 15.5% fewer correct match
decisions for the filler-control procedure (filler-control: 43.5%; standard: 59%; χ
2 = 11.61,
p=0.001, 95% CI [6.32%, 24.41%]) compared with the 26% decrease observed in the less
ambiguous materials (filler-control: 46%; standard: 72%; χ
2 = 32.37, p<0.001, 95% CI
[16.80%, 34.79]).
Although these differences were not significant in the multilevel model, there was a
drop in correct “matches” when the filler-control procedure was used. However, the observed
drop in hits appears to be far less than the drop in false alarms, seen in Tables 4 and 5. The
next section will specifically test the tradeoff between a small reduction in hits and a large
reduction in false alarms—is the reduction in false alarms proportionately larger than the
reduction in hits?
Does the Filler-Control Procedure Result in Better, Applied Outcomes?
The findings presented so far show that the filler-control procedure reduced the
number of times people were able to correctly identify the print that is actually a match
31
(“hits”). But, these results also indicated that the filler-control procedure vastly reduced the
number of incorrect match decisions on an innocent suspect’s print (“false alarms”).
However, the question remains—does the reduction in false alarms in the filler-control
procedure outweigh any loss of hits that also results? Forensic policy makers do not want to
reduce examiner’s chances of identifying a true match, but a slight reduction in hits might be
justifiable if the reduction in false alarms is large enough, as false alarms contribute to
wrongful conviction (equivalent to innocent suspect identifications in eyewitness literature).
The observed reduction in false alarms was much larger than the reduction in hits,
particularly in the more ambiguous condition—there was a 15.5% reduction in hits compared
with a 23.5% reduction in false alarms overall. As mentioned previously, this difference was
even more pronounced when contextual information effects are considered. When context
was present, false alarms were reduced by 32% when the filler-control procedure was used,
compared with a 15% reduction in false alarms when context was not present.
Use of signal detection theory measures. One way to look at data where there is a
proportion of correct affirmative decisions, and a proportion of incorrect affirmative
decisions is Signal Detection Theory using a measure called d´. There was both a decrease in
incorrect match decisions (“false alarms”) and correct match decisions (“hits”) in the fillercontrol procedure. So, the filler-control method reduced the number of incorrect matches
made, but also reduced the correct matches people made. This is similar to the patterns
observed in showups and lineups (Steblay et al., 2003). Situations in which a manipulation
produces the same directional change in both hits and false alarms (i.e., both increase or both
decrease) can be analyzed using d´.
32
Whereas d´ is generally used as a measure of psychological discriminability, it is not
presumed to be a measure of psychological discriminability when it is used on lineup data
(see Wells, Smalarz, & Smith, 2015 – Research Paper Writing Help Service; Wells, Smith, & Smalarz, 2015 – Research Paper Writing Help Service). The reason that d´ is
not a measure of memory discriminability for lineup data is because it ignores a large share
of false positive responses (on fillers), thereby violating assumptions of Signal Detection
Theory. However, as an applied measure of performance, it has been argued that a d´ analysis
on suspect identifications in lineups is a reasonable way to assess whether the trade-off in
reduced identifications of the culprit is compensated by the reduction in mistaken
identifications of innocent suspects (see Mickes, et al., 2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay). See Table 8 for the d´ values
for each condition, separated by ambiguity, procedure, and context conditions. Higher d´
values indicate that the number of innocent suspect picks is low relative to the number of
correct match decisions, and lower d´ value indicate that the number of innocent suspect
picks is high when compared with the number of correct matches. So, when thinking about
the trade-off between hits and false alarms, the better procedure should yield a higher d´
value.
Analyses using d´ values. For the ambiguous materials, the lowest d´ value was
obtained in the context-present condition with the standard procedure (d´ = 0.691), followed
by the context-absent condition with the standard procedure (d´ = 0.939). However, the fillercontrol method obtained higher d´ values than both of the standard procedure conditions,
regardless of the context presence manipulation (context: d´ = 1.156; no context: d´ = 1.190).
Table 8 shows that the less ambiguous materials achieved higher average d´ in each condition
than the more ambiguous materials too, as expected. But in all conditions, the filler-control
method proved to be superior, even over and above simply removing contextual information
33
from the procedure. An analysis with d´ was run with the full dataset, to examine whether
there was a three-way interaction between Ambiguity Level, Procedure Type, and Context
Presence on d´ values. A three-way ANOVA was used to analyze these data (refer to Table
13 for a summary of all d´ inferential analyses). There was no significant three-way
interaction (F(1, 229) = 0.659, p = 0.418), but there was an interaction between Ambiguity
Level and Procedure Type (F(1, 229) = 16.447, p <.001). So, the filler-control procedure had
the most benefit when the materials were ambiguous.
To confirm, a two-way ANOVA with context and procedure as factors was used to
determine whether the d´ values for participants who received more ambiguous materials
only were significantly different from one another. The interaction between context and
procedure had a nonsignificant effect on d´values (F(1, 115) = 1.437, p = 0.233). There was
also no main effect of context, suggesting that shielding people from contextual information
had no significant impact on d´ or improving the ratio of incorrect and correct match
decisions (F(1, 115) = 2.310, p = 0.131). But there was a main effect of procedure, indicating
that the filler-control procedure significantly improved d´ (F(1, 115) = 34.661, p <0.001).
These data suggest that shielding examiners from contextual information as much as possible
would not be as effective as evidence lineups at maximizing the trade-off between incorrect
and correct match decisions in a forensic context (Dror et al., 2015 – Research Paper Writing Help Service; Saks et al., 2003; Wells,
Wilford, & Smalarz, 2013).
Is the Increase in d´ from the Filler-Control Procedure due to Differential Filler
Siphoning?
In the eyewitness identification literature, the reduction in innocent suspect
identifications that is observed when a lineup is used comes as a result of filler siphoning
34
(Wells, Smalarz, & Smith, 2015 – Research Paper Writing Help Service; Wells, Smith, & Smalarz, 2015 – Research Paper Writing Help Service). In other words, good
fillers effectively draw (siphon) false positives, or “choosing”, away from an innocent
suspect. These same fillers, however, draw fewer choices away from the culprit because the
culprit (having been the source of the witnesses’ memories) tends to be the best choice. But
when the actual culprit is not in the lineup, then all of the options in the lineup are equally
plausible with regard to matching the eyewitnesses’ memory. Hence, these fillers tend to
strongly compete with the innocent suspect for choices by the eyewitness.
The filler-control method was anticipated to mirror the patterns seen in eyewitness
lineups and show differential filler siphoning effects. Previous analyses reported in the
current paper showed that the filler-control method (compared to the standard method)
decreased false match decisions on the innocent suspect prints. But, there are two possible
ways that the presence of fillers can decrease false match decisions on the innocent suspect
samples. First, the presence of fillers might have increased correct rejections. Alternatively,
the presence of fillers might have spread the false match decisions to false match decisions
the fillers (filler siphoning). The previous analyses on incorrect match decisions on the
innocent suspect sample did not tell us which of these occurred. This same question can be
asked about the reduction in correct match decisions in the filler-control procedure versus the
standard procedure. Was this reduction the result of filler siphoning or was this reduction the
result of an increase in incorrect rejections?
Analysis of all “choosing”. If filler siphoning was underlying reason for the
reduction in incorrect match decisions observed in the filler-control procedure, then there
should be no increase in the number of “no match” decisions (“correct rejections”) as a
function of the introduction of fillers. Likewise, if filler siphoning was underlying reason for
35
the reduction in correct match decisions observed in the filler-control procedure, then there
should be no increase in the number of “no match” decisions (“misses”) as a function of the
introduction of fillers. A full model was run with “choosing” (match decision = 1, no match
decision = 2) as the outcome variable. This was different from the full model analysis run on
“suspect choosing” as now occasions where people said match, but selected a filler, are
coded as “match” decisions. Match Presence, Procedure Type, Context Presence, and
Ambiguity Level were included as predictors, but the model failed to converge. The model
was reduced to include only the interactions that were approaching significance (Match
Presence by Procedure Type, and Context Presence and Ambiguity Level). This model did
converge, and the results revealed a significant Match Presence by Procedure Type
interaction (B = -0.801, p = 0.041; Table 11)
5
.
The nature of this interaction can be seen by looking at the proportions reported in
Table 4—in the filler-control procedure, people said that there was a “match” to the crime
print significantly more often than people in the standard procedure. Furthermore, this
difference in the number of “match” decisions between procedures was more pronounced
when there was no match present. But, what is most interesting is that this increase in
“match” decisions in the filler-control procedure did not result in more incorrect match
decisions (false alarms), or more correct match decisions (hits). In fact, the analyses
presented earlier in this paper show that hits and false alarms were reduced in the fillercontrol procedure, even though the current analyses show that choosing is boosted overall in
the filler-control procedure. But were these decreases in hits and false alarms are due to filler
siphoning? People who would have chosen the culprit or innocent suspect print if they had
5 11% of the variation in “choosing” is explained by the predictors and interactions (Marginal R2 = 0.105) and
25% of the variation in “choosing” is explained by all predictors in the model, including higher-level grouping
variables (Conditional R2 = 0.251). Goodness of Fit: χ
2 (8) = 428.14, p<.001.
36
received the standard procedure must be either “backing off” now that they need to choose
the match from a lineup of fingerprints. Alternatively, people in the filler-control procedure
may be still choosing to say there was a match, but now incorrectly selecting filler print
samples. Specifically, I needed to show if people were making fewer hits and false alarms
because they were now “missing” true matches, or “correctly rejecting” match absent trials
more often, or they were selecting filler print samples that because they thought the fillers
were a true “match” to the crime print instead.
Separate analyses of “miss” and “correct rejection” rates. To address this
question, the “miss” rates and the “correct rejection” rates were assessed between procedures
in two separate models. Initially for both, full models were run with all possible interactions
between Procedure Type, Context Presence, and Ambiguity, and Context Type as a control
variable. The two higher-level grouping variables were included (Fingerprint Set and
Participant). Neither of these converged, so the interactions and control variables were
excluded. The final models (refer to Table 11) had no significant predictors, and showed that
there was no significant effect of Procedure Type, and therefore no difference between the
standard and filler-control procedure for either “misses” (B = -0.203, p = 0.7916
) or “correct
rejections” (B = -0.222, p = 0.8287
), suggesting that people incorrectly chose a “no match”
decision equally often in both procedures. This was the same pattern has been consistently
observed when comparing eyewitness lineups and showups. These results indicated that the
6 0.3% of the variation in “misses” is explained by the predictors and interactions (Marginal R2 = 0.003) and
69% of the variation in “misses” is explained by all predictors in the model, including higher-level grouping
variables (Conditional R2 = 0.691). Goodness of Fit: χ
2 (8) = 19.413, p=0.013.
7 0.2% of the variation in “correct rejections” is explained by the predictors and interactions (Marginal R2 =
0.002) and 85% of the variation in “correct rejections” is explained by all predictors in the model, including
higher-level grouping variables (Conditional R2 = 0.846). Goodness of Fit: χ
2 (8) = 30.041, p<.001.
37
decrease in incorrect and correct match decisions (hits and false alarms) observed here was
due to filler siphoning, not an increase in “no match” decisions.
Are the Contextual Bias Effects and the Effect of the Filler-Control Procedure
Reflected in the Confidence Measures?
It was expected that the confidence measures would show that the filler-control
method and more ambiguous materials would decrease confidence overall. The reasoning
behind this hypothesis was that lineups and ambiguity makes it more difficult to determine
the correct answer, and therefore should lower confidence in decisions. In addition, it was
anticipated that contextual information congruent with a person’s decision (i.e., when the
contextual information suggested the prints should match and the participant chose match),
the contextual information should act as additional evidence in favor of their decision and
boost their confidence. A full multilevel model was run initially with the data nested within
Fingerprint Set and Participant (Level 2 grouping variables), allowing the variation in
confidence that was due to fingerprint materials, and participants’ unique biases. Context
Presence, Ambiguity Level, Procedure Type, and Match Presence were added as predictors,
and Context Type as a control variable. Participants’ subjective confidence in their decision
was the outcome variable.
The model was first run with all possible interactions. No interaction terms were
significant, but there was a significant effect of Ambiguity Level (B =-1.457, p = 0.002), and
Procedure Type (B =-0.913, p = 0.050) was approaching significance. In addition, unlike in
the binary models in the rest of the analyses, there was a significant effect of some Context
Types so this control variable would be retained in future models. To help reduce the model,
the “step” program in the LMER package in R was used to retrieve the best fitting model.
38
The model with the best fit included Ambiguity Level and Procedure Type as predictors, and
no interactions. So, this reduced model was used to examine the confidence levels of
participants (refer to Table 12 for a summary of these results)8
.
As hypothesized, the filler-control procedure (B = -0.4681, p=0.031), more ambiguity
(B = -0.902, p<.001) resulted in significantly lower average confidence levels overall. The
confidence results were largely consistent with our expectations and fit well with the
contextual bias results described earlier. The materials that were less clear (more ambiguous)
resulted in lower confidence. The more difficult task (filler-control procedure) also led to
lower confidence levels. However, there was no consistent increase in confidence in
participant’s decisions when the contextual information was congruent with the decision
being made, indexed by a nonsignificant Context by Match Presence interaction. Refer to
Tables 6 and 7 for summaries of the confidence levels for the decisions.
8 6% of the variation in confidence is explained by the predictors and interactions (Marginal R2 = 0.121) and
44% of the variation in confidence is explained by all predictors in the model, including higher-level grouping
variables (Conditional R2 = 0.259).
39
CHAPTER 4. DISCUSSION
Consistent with previous literature, these data demonstrate that the standard forensic
examination procedures can lead to contextual bias effects when the samples are ambiguous.
Using the standard procedure with the more ambiguous fingerprint materials, people who
received information suggesting the suspect is guilty made more incorrect match decisions on
the innocent suspect sample compared with people who received no contextual information
(25% versus 42%). These results are consistent with previous literature showing contextual
bias effects in fingerprint analysis (Dror et al., 2005; Dror et al., 2006 – Write a paper; Professional research paper writing service – Best essay writers; Obsorne & Zajac,
2015 – Research Paper Writing Help Service) and add weight to the National Academy of Science’s concerns regarding forensic
examination procedures (National Academy of Science, 2009).
In addition, this paper breaks new ground. The research community and the National
Academy of Science (2009) have been calling for a solution to the effect of contextual bias in
forensic judgments. Previous literature has typically suggested shielding examiners from
contextual information (Saks et al., 2003), but Wells, Wilford, and Smalarz (2013) argued
that this was insufficient—examiners can never be completely shielded from contextual
information because even a single sample presented alone suggests that there is evidence to
indicate that this sample is from the true culprit. Instead, Wells and colleagues (2013) asked
whether contextual bias could be reduced or eliminated in a forensic context with an
evidence lineup procedure. The current experiment suggests that the answer is yes. When
people received evidence lineups rather than the standard procedure, people matched the
crime print to the innocent suspect print 10% of the time on average, whether they received
contextual information or not. So, there was a 32% reduction in people matching the crime
print to the innocent suspect when people received an evidence lineup as well as contextual
40
information compared with the standard procedure with no lineup (42% versus 10%). These
data also indicate that evidence lineups offer a benefit not only when contextual information
is present, but also outperform the standard procedure even in the absence of additional
contextual information. Without contextual information, there was a 15% reduction in people
incorrectly matching the crime print to innocent suspect prints in the filler-control procedure
compared with the standard procedure (25% versus 10%).
Furthermore, these results capstone project writing help U showed identical patterns to those described in the
eyewitness literature. Generally, showups are characterized by higher innocent suspect
identifications, but lower choosing rates than lineups. Eyewitness lineups result in higher
choosing rates than showups, but also lower innocent suspect identification rates (Steblay et
al, 2003). In the current study, there was a higher rate of affirmative match decisions in the
filler-control procedure (72% in match present, and 58% in match absent trials) compared
with the standard procedure (65% in match present, and 33.5% in match absent). But in
match-absent filler-control trials, the majority of these match decisions were filler picks
(48.5% of the time people selected a filler; see Table 2). Therefore, only 9.5% of the time
people selected the innocent suspect in the filler-control procedure compared with 33.5% of
the time in the standard procedure. This result can be attributed to the eyewitness
identification filler siphoning mechanism—when no actual match is present, good fillers can
spread a large share of false positive errors out to the fillers (Wells, Smalarz, & Smith, 2015 – Research Paper Writing Help Service;
Wells, Smith, & Smalarz, 2015 – Research Paper Writing Help Service). But there is a trade-off—the innocent suspect is more
protected in a lineup, yet there is also a loss in culprit identifications. As is found in
eyewitness literature, people selected fillers in match-present trials too (27.5% of the time),
but the observed reduction in hits is smaller than the magnitude of the benefit to the innocent
41
suspect, making lineups the better choice (see Wells, 1993 for a summary of the eyewitness
literature). Accordingly, there is a benefit of evidence lineups over evidence showups in a
forensic examination context, just as has been observed in an eyewitness context. Future
research on evidence lineups can draw on the theory and methods in the eyewitness literature
to move forward and further explain and develop these findings.
In addition, the patterns in the confidence judgments were consistent with my
hypotheses, the underlying theoretical framework of this project, and the other results
obtained in this study. People were less confident when the materials were more ambiguous,
indicating that they found the task more difficult, in line with the increase in incorrect
decisions when the materials were more ambiguous. It also makes theoretical sense that
contextual bias effects were found in the trials with ambiguous materials only. When the
information in a bottom-up process is ambiguous and seems to not provide a clear answer,
people look for other information to help them, such as contextual information, which will
influence decisions via top-down processing (Tversky & Kahneman, 1974; Saks et al., 2003).
People tended to have lower confidence when they use the filler-control procedure,
indicating that this task maybe also felt more difficult or made people less sure of their
selection because there were other, plausible options. In fact, the lower confidence is
probably due to a combination of these factors, given the high number of people deciding
that a filler was a plausible match.
But what is the mechanism by which fingerprint lineups reduce the impact of
contextual information on decision-making? When faced with a single sample, the question
is whether or not someone is sure that this print resembles the crime print enough to say they
came from the same person’s finger. When contextual information is available that suggests
42
that a match is likely, people start with an expectation that the prints are likely to match and
search for evidence of this. This is a different starting point to someone who begins with no
information to help the decision-maker assess likelihood of guilt—the search for confirming
and disconfirming evidence is unbalanced and therefore more likely to fall in favor of his or
her expectation.
However, as suggested by Wells, Wilford, and Smalarz (2013), even presenting a
single sample suggests that there is reason to believe this person committed the crime. So,
whereas I found evidence in this study that the additional information contained in the police
case report increased innocent suspect match decisions when the task was more difficult, this
does not mean that there was not some contextual bias occurring with the standard procedure
in the condition where participants received no contextual information. These data can speak
to the effect of the additional contextual information and how it increased the likelihood of
an incorrect match decision when the suspect sample was presented alone. Specifically,
because the evidence lineups reduced innocent suspect match decisions even when
comparing the procedures that were absent of contextual information, it is possible that even
the single sample presentation was having a small contextual bias effect in the standard
procedure that was removed with the presence of additional filler samples.
How did the evidence lineups eliminate the contextual bias effect? When faced with
six plausible samples, one of which may be a match, the decision is now more difficult. Is
there one that looks more like it matches than the others? If so, then the decision is whether
that sample is actually a match or just very similar? In addition, people know that if they do
not do a proper bottom-up analysis there is a five out of six chance that they will pick a filler,
and they will be incorrect. Even when provided with information suggesting that the suspect
43
sample should match the crimple sample, they still need to be able to pick out which one is
the suspect. In the evidence lineups, unless the person can perform the bottom-up processing
task of identifying the print that matches the crime print, the addition information cannot
assist them in their decision via top-down processing. Importantly, the filler-control
procedure is not only a test of the suspect’s guilty, it is also a test of examiners’ abilities to
perform the task they are trained to do. The task can be failed if the examiner picks a filler,
feedback that they would get very soon after performing the examination. If an examiner
cannot perform the fingerprint-matching task proficiently, then they will begin to accumulate
a high rate of filler picks during the course of their work.
To illustrate, let’s return to the example of Lana Canen again, introduced at the
beginning of this paper. If Lana Canen’s print had been embedded in a lineup of six
fingerprints, the examiner would have needed to pick out her fingerprint before he could use
the other information he had concerning another man’s confession, which implicated her in
the crime. As such, if the filler-control procedure had been used in Lana Canan’s case, the
examiner would have had a five out of six chance of picking a known-innocent filler rather
than picking Lana’s prints. Furthermore, the examiner would have known that picking a filler
was a possibility, which might have pushed the examiner into a more conservative decisionmaking style and diluted the effect of contextual information. Because her prints did not
actually match the prints at the crime scene, our data suggest that a procedure where her print
was presented alone for comparison put her in the most danger of being matched to the crime
print, especially if other contextual information was available. But, if her print had been
examined in a fingerprint lineup, that had good fillers, her prints would have had no more of
a chance of being selected as a match than would the prints of a filler. If one extrapolated
44
from the current data, the chance of Lana’s prints being picked out as matching the crime
print would have been 32% lower in the filler-control procedure (10% innocent suspect
match decisions in the filler-control procedure compared with 42% in the standard
procedure). So, the filler-control procedure might have revealed that the examiner was biased
in this case. Other reasons that an examiner might pick a filler are that they cannot
adequately perform the bottom-up task, or are lying about their fingerprint analysis abilities.
Lana Canen may have avoided 8 years in prison if the filler-control procedure had been used.
The obvious application for this research is to move this procedure to the field to help
prevent wrongful convictions, uncover fraudulent or incompetent examiners, and improve the
quality and independence of forensic evidence in court. It may be that this procedure need
not be utilized in every case, but it should certainly be used whenever a second opinion is
required, especially where the previous examiners decision is known, or for high-profile
cases where contextual information is readily available. In addition, if a police investigator
wishes to provide the examiner with additional information about the case besides the
samples themselves, then a filler-control procedure should be used to ensure that this extra
information does not impact the decision. Finally, this procedure could be used as a way to
calibrate individual examiners and score the reputation of laboratories. The filler-control
procedure allows examiners to receive feedback about their performance in real cases, which
is useful and currently not available. If an examiner makes in incorrect affirmative
determination, this error might never be discovered or not discovered until many years later
if the innocent suspect is exonerated. And, as Wells et al. (2013) noted, the frequency with
which analysts select fillers in real cases can inform the legal system about the competencies
of individual examiners and the reliability of various techniques based on actual cases
45
submitted to analysis. As such, the filler-control method can provide the legal system with
the ability to calculate error rates for examiners and their laboratories using the actual cases
that are submitted to their labs. Therefore, not only can we identify examiners performing
poorly, but we can also establish a general error rate in fingerprint examination based on the
actual cases that they are routinely analyzing.
There are likely to be practical difficulties involved in creating evidence lineups for
real cases. The selection of fillers would take time and man hours, and there would need to
be a clear selection strategy in place to ensure that each evidence lineup is fair. Evidence
lineups would need to be created in a timely fashion to ensure they do not exacerbate the
backlog that is already an issue in many forensic laboratories. So, before these findings can
be applied in the real world, they need to be tested further and policy-related questions would
need to be answered. Who will create the evidence lineups? How will they be created and
verified as fair? How will this additional task be funded? What forensic materials benefit
most from evidence lineups and under what situations is it most beneficial to use the fillercontrol procedure? For example, fiber analysis is a good example of a evidence material that
could benefit from the filler-control procedure—if an examiner cannot say with certainly that
one fiber in the lineup is the same as the fibers associated with a crime, then the examiner
cannot testify that these fibers are incriminating evidence, but they may corroborate other
evidence in the case.
It may be that evidence lineups are not appropriate for routine cases that come
through a laboratory. But maybe verification evaluations should use evidence lineups to
weed out incompetent or dishonest examiners, and add another level of protection to the
innocent suspect. In addition, it would be valuable to require the filler-control procedure to
46
be used for any new forensic techniques. Police investigators are frequently looking for new
kinds of evidence to present to help secure convictions, so new techniques emerge fairly
regularly, for example, matching the wear on a pair of jeans to a pair of jeans belonging to
the suspect. Until examiners for these techniques have demonstrated that they can reliably
discriminate between fillers and the suspect sample with these new techniques, the fillercontrol procedure should be required in all cases.
It is important to note, also, that our participants were not fingerprint examination
experts. Whereas studies have shown that laypersons and novice examiners are able to
distinguish between matching and non-matching prints (Tangen, Thompson, & McCarthy,
2011; Vokey, Tangen & Cole, 2009), experts show different patterns to novices and
laypersons. Experts consistently outperform novice examiners, particularly in situations
where the two fingerprints to be analyzed do not match, but are extremely similar (Experts
were correct 99.32% of the time and novices only 44.82% of the time). However, when the
prints do match, or when the prints do not match and are very dissimilar, experts typically
achieve more than 90% accuracy, and novices between 70 and 80% accuracy (Tangen,
Thompson, & McCarthy, 2011). Experts do not achieve perfect accuracy, but their error rate
is much lower than novices. The other significant difference between experts and novices is
that experts tend to make conservative errors more often, rather than the kinds of errors that
could lead to wrongful convictions. So, these data suggest that if the current study was run
with an expert population, the contextual bias effects may be slightly smaller, and the error
rates in the standard procedure not so high if experts tended to be conservative in their
evaluations. But, experts will still make these kinds of errors that endanger innocent suspect
on rare occasions. However, to this day, there has never been a systematic test to obtain an
47
actual measure of the error rate in the field for fingerprint examiners, so whether these
patterns translate to the field remains to be seen (Busey & Parada, 2010 – Essay Writing Service: Write My Essay by Top-Notch Writer; Cole, 2005). In
addition, some studies suggest that the technology available to experts now results in an
increased susceptibility to bias and error in real world cases, such as the use of the large
AFIS database that is used to find highly similar prints for later examination by forensic
experts (Dror & Mnookin, 2010 – Essay Writing Service: Write My Essay by Top-Notch Writer; Marcon, 2009). By its very design, AFIS creates very
ambiguous comparisons for experts to inspect, the circumstances under which contextual bias
was found in the current study. Although experts tend towards more conservative errors and
have a smaller error rate, they are not immune to contextual bias effects (Dror et al., 2006 – Write a paper; Professional research paper writing service – Best essay writers),
so it is anticipated that the filler-control procedure would be beneficial for experts too and
would still improve their already superior performance.
Because this finding is novel, the effects should be replicated with fingerprints, as
well as other forensic materials to determine how generalizable the evidence lineups solution
is. In addition, we should see if other forms of contextual information produce contextual
bias in a similar way too. Currently, my laboratory is investigating how having knowledge of
another person’s determination about the same set of fingerprints can influence people’s own
match decision. This procedure can be likened to an examiner who knows what another
examiner concluded. If knowing what another person thought was correct impacts people’s
decisions, then there should be significantly more decisions that are consistent with the
previous person’s evaluation. For example, if a participant has information telling them that
the previous participant determined that the prints matched, then this would create an
expectation that the prints will match and they should be biased to also find a match. People
should also make consistent decisions when they believe the previous participant thought the
48
prints did not match, biasing them to determine no match as well. In addition, we are looking
at the influence of exculpatory information on fingerprint decisions, such as a DNA
mismatch, or a solid alibi.
Furthermore, in the real world, experts are allowed to conclude that crime samples are
not of sufficient quality for them to determine a “match” or “no match”. Therefore, a study
should be run where participants are also given the option to say that the prints are “not
suitable for analysis”. Choosing this option would allow a participant, or fingerprint
examiner, to forgo making a match decision on the grounds that the prints are too degraded
or ambiguous to decide whether the prints came from the same person who left the print at
the crime scene. This option may change the pattern of results for lay people, but in the real
world there may be too much pressure to make a match decision from the police
investigators. Accordingly, the last goal would be to test the filler-control procedure (with
and without the “not suitable for analysis option”) with real examiners too. Once data with
experts has been obtaining, it would be possible to determine the situations in which using
this procedure would be most efficient.
Although the filler-control procedure may require some additional resources and
examiner time, it would be extremely beneficial to the legal system to make use of this to
improve forensics. Given that there are some practical drawbacks to constructing and
examining a lineup versus a single sample, future research should find the conditions under
which contextual bias is most prevalent, and the phases in an investigation when it would be
most useful to introduce an evidence lineups, such as a second opinion examination or when
there is a lack of other evidence in the case to corroborate the fingerprint examiners
conclusion.
49
REFERENCES
CBS News (November, 2014: 2024 – Essay Writing Service. Custom Essay Services Cheap). Lana Canen freed over bad fingerprint evidence after 8 years
in prison for Indiana murder. CBS News. Retrieved from
http://www.cbsnews.com/news/lana-canen-freed-over-bad-fingerprint-evidence-after8-years-in-prison-for-indiana-murder/
Chaiken, S., & Maheswaran, D. (1994). Heuristic processing can bias systematic processing:
Effects of source credibility, argument ambiguity, and task importance on attitude
judgment. Journal of Personality and Social Psychology, 66(3), 460-473.
doi:10.1037/0022-3514.66.3.460
Charman, S.D. (2013). The forensic confirmation bias: A problem of evidence integration,
not just evidence evaluation. Journal of Applied Memory and Cognition, 2, 56-58.
doi:10.1016/j.jarmac.2013.01.010
Dror, I. E. (2016: 2024 – Do my homework – Help write my assignment online). A Heirarchy of Expert Performance. Journal of Applied Memory and
Cognition, 5, 121-127. doi:10.1016//j.jarmac.2016: 2024 – Do my homework – Help write my assignment online.05.121
Dror, I.E., Charlton, D., & Peron, A.E. (2006 – Write a paper; Professional research paper writing service – Best essay writers). Contextual information renders experts
vulnerable to making erroneous identifications, Forensic Science International, 156,
74-78. doi:10.1016/j.forsciint.2005.10.017
Dror, I.E., Peron, A.E., Hind, S., & Charlton, D. (2005). When emotions get the better of us:
The effect of contextual top-down processing on matching fingerprints. Applied
Cognitive Psychology, 19, 799-809. doi:10.1002/acp.1130
Dror, I.E., & Charlton, D. (2006 – Write a paper; Professional research paper writing service – Best essay writers). Why experts make errors. Journal of Forensic
Identification, 56(4), 600-616.
50
Dror, I.E., Thompson, W.C., Meissner, C.A., Kornfield, I., Krane, D., Saks, M., & Risinger,
M. (2015 – Research Paper Writing Help Service). Context Get research paper samples and course-specific study resources under   homework for you course hero writing service – Manage ment Toolbox: dissertation help HongKong A Linear Sequential Unmasking (LSU)
approach for minimizing cognitive bias in forensic decision making. Journal of
Forensic Science, 60(4), 1111-1112. doi:10.1111/1556-4029.12805
Einhorn, H.J. & Hogarth, R.M. (1981). Behavioral decision theory: Processes of judgment
and choice. Annual Psychological Review, 32, 53-88.
Eisen, M.L., Smith, A.M., Olaguez, A.P., & Skerritt-Perta, A.S. (2017). Examination of
showups conducted by law enforcement using a field-simulation paradigm.
Psychology, Public Policy, and Law, 23(1), 1-22. doi:10.1037/law0000115
Innocence Project (2016: 2024 – Do my homework – Help write my assignment online). Misapplication of Forensic Science. Retrieved December 19, 2016: 2024 – Do my homework – Help write my assignment online.
https://www.innocenceproject.org/causes/misapplication-forensic-science/
Kerstholt, J.H., Paashuis, R., & Sjerps, M. (2007). Shoe print examinations: Effects of
expectation, complexity and experience. Forensic Science International, 165, 30-34.
doi:10.1016/j.forsciint.2006 – Write a paper; Professional research paper writing service – Best essay writers.02.039
Kerstholt, J.H., Eikelboom, A., Dijkman, T., Stoel, R., Hermsen, R., & van Leuven, B.
(2010 – Essay Writing Service: Write My Essay by Top-Notch Writer). Does suggestive information cause a confirmation bias in bullet comparisons?
Forensic Science International, 198, 138-142. doi:10.1016/j.forsciint.2010 – Essay Writing Service: Write My Essay by Top-Notch Writer.02.007
Kukucka, J., & Kassin, S.M. (2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay). Do confessions taint perceptions of handwriting
evidence? An empirical test of the forensic confirmation bias. Law and Human
Behavior, 38(3), 256-70. doi:10.1037/lhb0000066
Langenburg, G., Champod, C., & Wertheim, P. (2009). Testing for potential contextual bias
effects during the verification stage of the ACE-V methodology when conducting
51
fingerprint comparisons. J Forensic Sci, 54(3), 571-582. doi:10.1111/j.1556-
4029.2009.01025.x
Marcon, J.L. (2009). The distinctiveness effect in fingerprint identification: How the role of
distinctiveness, information loss, and informational bias influence fingerprint
identification (Unpublished doctoral dissertation). University of Texas at El Paso,
USA.
Mickes, L., Moreland, M. B., Clark, S. E., & Wixted, J. T. (2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay). Missing the information
needed to perform ROC analysis? Then compute d´, not the diagnosticity ratio.
Journal of Applied Research in Memory and Cognition, 3, 58–62.
doi:10.1016/j.jarmac.2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay.04.007Mickes,
National Academy of Sciences. (2009). Strengthening forensic science in the United States:
A path forward. Washington, DC: National Academies Press.
Nickerson, R.S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises.
Review of General Psychology, 2(2), 175-220.
Osborne, N.K.P, Woods, S., Kieser, J., & Zajac, R. (2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay). Does contextual information bias
bitemark comparisons? Science and Justice, 54(4), 267-273.
doi:10.1016/j.scijus.2013.12.005
Osborne, N.K.P. & Zajac, R. (2016: 2024 – Do my homework – Help write my assignment online). An imperfect match? Crime-related context influences
fingerprint decisions. Applied Cognitive Psychology, 30, 126-134.
Saks, M.J., Risinger, D.M., Rosenthal, R. & Thompson, W.C. (2003). Context effects in
forensic science: A review and application of the science of science to crime
laboratory practice in the United States. Science & Justice, 43, 77-90.
doi:10.1016/S1355-0306(03)71747-X
52
Smith, A.M., Wells, G.L., Lindsay, R.C.L., & Penrod, S.D. (2017). Fair lineups are better
than biased lineups and showups, but not because they increase underlying
discriminability. Law and Human Behavior, 41(2), 127-145. doi:10.1037/lhb0000219
Steblay, N., Dysart, J., Fulero, S., & Lindsay, R.C.L. (2003). Eyewitness accuracy rates in
police showup and lineup presentations: A meta-analytic comparison. Law and
Human Behavior, 27(5), 523-540.
Thompson, M.B., & Tangen, J.M. (2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay). The nature of expertise in fingerprint matching:
Experts can do a lot with a little. PLoS ONE 9(12): e114759.
doi:10.1371/journal.pone.0114759
Thompson, M.B., Tangen, J.M., & McCarthy, D.J. (2014: 2024 – Essay Writing Service | Write My Essay For Me Without Delay). Human matching performance of
genuine crime scene latent fingerprints. Law and Human Behavior, 38(1), 84-93.
doi:10.1037/lhb0000051
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases.
Science, 185(4157), 1124-1131. doi:10.1126/science.185.4157.1124
Vokey, J.R., Tangen, J.M. & Cole, S.A. (2009). On the preliminary psychophysics of
fingerprint identification. The Quarterly Journal of Experimental Psychology, 62(5),
1023-1040. doi:10.1080/17470210802372987
Wells, G.L. (1993). What so we know about eyewitness identification? American
Psychologist, 48(5), 553-571. doi:10.1037/0003-066X.48.5.553
Wells, G.L., Smalarz, L., and Smith, A.M. (2015 – Research Paper Writing Help Service). ROC analysis of lineups does not measure
underlying discriminability and has limited value. Journal of Applied Research in
Memory and Cognition, 4, 313-317.
53
Wells, G.L., Smith, A.M., and Smalarz, L., (2015 – Research Paper Writing Help Service). ROC analysis of lineups obscures
information that is critical for both theoretical understanding and applied purposes.
Journal of Applied Research in Memory and Cognition, 4, 324-328.
Wells, G.L., Wilford, M.W., & Smalarz, L. (2013). Forensic science testing: The forensic
filler-control method for controlling contextual bias, estimating error rates, and
calibrating analysts’ reports. Journal of Applied Research in Memory and Cognition,
2, 53-55. doi:10.1016/j.jarmac.2013.01.004
Wixted, J.T. & Wells, G.L. (in press). The relationship between eyewitness confidence and
identification accuracy: A new synthesis. Psychological Science in the Public
Interest. doi:10.1177/1529100616686966
54
Figure 1. A graphical representation of the multilevel logistic regression models that used
binary sample choice variables as the dependent measure, with three predictors, and two
higher level grouping variables.
55
Figure 2. A graphical representation of the multilevel model used in the analyses with
participant’s confidence in their decisions as the dependent measure.
56
Table 1. A table summarizing of the number of participants in each between subjects
condition.
Between Subjects Group Number of Participants
Less Ambiguous Materials 116
More Ambiguous Materials 119
Standard Procedure without Context 57
Standard Procedure with Context 53
Filler-control Procedure without Context 61
Filler-control Procedure with Context 64
Table 2. Summary of the mean proportion of people in each between-subjects condition who
selected match (suspect or filler) or no match, and the mean confidence for each decision in
the pilot data.
Ambiguity Actual
match
Selected correct print or
innocent suspect
Selected a filler Selected no match
Proportion Confidence Proportion Confidence Proportion Confidence
Less
ambiguous
materials
Match 0.50 74% 0.29 65% 0.21 70%
No Match 0.15 71% 0.54 65% 0.31 70%
More
ambiguous
materials
Match 0.44 65% 0.25 59% 0.31 60%
No Match 0.10 66% 0.56 61% 0.34 62%
Notes. Proportion refers to the number of people in that condition who selected that sample,
e.g. the innocent suspect when there was no match, for the correct print when there was a
match. Confidence refers to the average confidence level for that choice e.g. people who
were in the condition with less ambiguous materials and correctly identified the print that
matched the crime print were 74% confident in their decision on average.
Table 3. A summary of the terminology for the dependent measures in the logistic multilevel
regression analyses.
Procedure Match
Presence
Match decision – “Choosing” No match decision
Suspect sample Filler sample
Standard Match “Hit” – “Miss”
No Match “False Alarm” – “Correct Rejection”
FillerControl
Match “Hit” “False alarm on a
filler sample”
“Miss”
No Match “False alarm” “False alarm on a
filler sample”
“Correct Rejection”
57
Table 4. Summary of the mean proportion of people in each between-subjects condition who
selected match or no match.
Procedure Context
present or
absent
Actual
match
Selected true
match or
innocent
suspect
Selected
filler
Selected no
match
Standard No Context Match 0.65 – 0.35
No Match 0.29 – 0.71
Context Match 0.65 – 0.35
No Match 0.38 – 0.62
Filler
Control
No Context Match 0.45 0.25 0.30
No Match 0.10 0.48 0.42
Context Match 0.44 0.30 0.26
No Match 0.09 0.49 0.42
Notes. Proportion refers to the number of people in that condition who selected that sample,
e.g. the innocent suspect when there was no match, for the correct print when there was a
match.
Table 5. Summary of the mean proportion of people in each between-subjects condition who
selected match or no match, separated by ambiguity condition.
Ambiguity
condition
Context
present or
absent
Actual
match
Selected
true match
or innocent
suspect
Selected
filler
Selected
no match
Standard
Procedure
Less
ambiguous
materials
No Context Match 0.72 – 0.28
No Match 0.34 – 0.66
Context Match 0.72 – 0.26
No Match 0.34 – 0.66
More
ambiguous
materials
No Context Match 0.59 – 0.41
No Match 0.25 – 0.75
Context Match 0.59 – 0.41
No Match 0.42 – 0.58
FillerControl
Procedure
Less
ambiguous
materials
No Context Match 0.44 0.22 0.34
No Match 0.09 0.48 0.43
Context Match 0.48 0.31 0.21
No Match 0.08 0.48 0.44
More
ambiguous
materials
No Context Match 0.47 0.28 0.25
No Match 0.10 0.48 0.42
Context Match 0.40 0.30 0.30
No Match 0.10 0.50 0.40
Notes. Proportion refers to the number of people in that condition who selected that sample,
e.g. the innocent suspect when there was no match, for the correct print when there was a
match.
58
Table 6. Summary of the mean confidence level of people in each between-subjects condition
who selected match or no match, separated by procedure and context presence.
Procedure Context
present or
absent
Actual
match
Selected true
match or
innocent
suspect
Selected
filler
Selected no
match
Standard No Context Match 72% – 62%
No Match 67% – 68%
Context Match 72% – 63%
No Match 67% – 70%
Filler
Control
No Context Match 74% 60% 63%
No Match 61% 63% 65%
Context Match 69% 64% 63%
No Match 67% 62% 64%
Notes. Confidence refers to the average confidence level for that choice for that combination
of between-subjects factors e.g. people who were in the standard procedure condition with no
context, and also correctly identified the print that matched the crime print were 72%
confident in their decision on average.
Table 7. Summary of the mean confidence level of people in each between-subjects condition
who selected match or no match, separated by procedure, ambiguity, and context presence.
Ambiguity
condition
Context
present or
absent
Actual
match
Selected
true match
or innocent
suspect
Selected
filler
Selected
no match
Standard
Procedure
Less
ambiguous
materials
No Context Match 78% – 68%
No Match 71% – 78%
Context Match 73% – 66%
No Match 67% – 74%
More
ambiguous
materials
No Context Match 64% – 59%
No Match 63% – 61%
Context Match 71% – 61%
No Match 68% – 67%
FillerControl
Procedure
Less
ambiguous
materials
No Context Match 76% 62% 72%
No Match 54% 66% 72%
Context Match 72% 67% 67%
No Match 65% 64% 69%
More
ambiguous
materials
No Context Match 71% 54% 51%
No Match 67% 59% 57%
Context Match 64% 61% 61%
No Match 69% 61% 60%
Notes. Confidence refers to the average confidence level for that choice for that combination
of between-subjects factors e.g. people who were in the standard procedure condition with
less ambiguous materials, and no context, and also correctly identified the print that matched
the crime print were 78% confident in their decision on average.
59
Table 8. Table comparing the d´ values in each procedure, with and without context, with all
fingerprint materials, and then separated by ambiguity condition.
Ambiguity
Condition
Procedure Context d´
All materials Standard No Context .939
Context .691
Filler control No Context 1.156
Context 1.190
Less ambiguous Standard No Context .995
Context .995
Filler control No Context 1.190
Context 1.355
More ambiguous Standard No Context .902
Context .420
Filler control No Context 1.206
Context 1.028
Notes. D prime = d´. The parameter d´ is not being used as a measure of discriminability,
rather it is being used as an indication of how much the innocent suspect is incorrectly
identified relative to correct match decisions. So, lower d´ values indicate more danger of
incorrect identification of the innocent suspect.
Table 9. Table showing the intraclass correlation values (ICC) for Fingerprint Set and
Participant grouping variables in the current data.
Notes. All values rounded to 2dp. ICC = intraclass correlation, which is the percentage of
variation in the outcome variable attributable to the higher-level grouping variable alone. For
example, an ICC of 0.16 or 16% for “choosing” means that 16% of the variation in
“choosing” is attributable to the Fingerprint Set grouping variable.
Outcome variable ICC—Fingerprint Set ICC—Participant
Mean 95% Confidence
Interval
Mean 95% Confidence
Interval
Choosing 16% [0.11, 0.23] 5% [0.02, 0.08]
Suspect Choosing 26% [0.20, 0.35] 6% [0.03, 0.09]
False Alarms 27% [0.20, 0.36] 4% [0.11, 0.07]
Misses 23% [0.17, 0.31] – –
Correct Rejections 47% [0.39, 0.57] – –
Confidence 9% [0.06, 0.14] 39% [0.34, 0.45]
60
Table 10. A summary of the multilevel models assessing contextual bias in the data obtained
from participants who received the standard procedure and more ambiguous materials.
Outcome Fixed effect Estimate S.E. z value p value
Hits Intercept -4.014 1.945 -2.064 0.039
Context 0.032 0.301 0.105 0.916
False Alarms Intercept -4.088 1.434 -2.851 0.004
Context Presence 0.769 0.324 2.369 0.018
Notes. All values rounded to 3 dp. S.E. = Standard Error of the mean. Level 2 grouping
variables: Participant (N = 234), and Fingerprint Set (N = 64).
Table 11. A summary of the two level logistic multilevel model results to assess the affects of
predictors on each decision type.
Outcome Fixed effects Estimate S.E. z value p value
Choosing Intercept -0.779 0.241 -3.228 0.001
Match Presence 1.513 0.283 5.352 <.001
Procedure Type 1.085 0.281 3.856 <.001
Context Presence 0.182 0.169 1.074 0.282
Ambiguity Level -0.102 0.229 -0.446 0.656
Match x Procedure -0.801 0.393 -2.040 0.041
Context x Ambiguity -0.031 0.239 -0.129 0.897
Hits Intercept -2.916 0.982 -2.970 0.003
Procedure Type -0.990 1.556 -0.857 0.392
Context Presence 0.032 0.146 0.219 0.827
Ambiguity Level -0.434 0.645 -0.672 0.502
False Alarms Intercept -3.330 0.680 -4.895 <.001
Procedure -1.889 0.749 -2.521 0.012
Ambiguity -0.075 0.307 -0.246 0.806
Context Presence -1.179 0.621 -0.289 0.773
Context x Ambiguity 0.609 0.431 1.412 0.158
Misses Intercept -3.352 0.687 -4.881 <.001
Context Presence -0.169 0.160 -1.057 0.291
Ambiguity Level 0.203 0.568 0.357 0.721
Procedure Type -0.203 0.768 -0.265 0.791
Correct
Rejections
Intercept -3.372 1.004 -3.358 <.001
Context Presence -0.170 0.174 -0.976 0.329
Ambiguity Level -0.282 0.576 -0.490 0.624
Procedure Type -0.222 1.022 -0.217 0.828
Notes. All values rounded to 3 dp. S.E. = Standard Error of the mean. Level 2 grouping
variables: Participant (N = 234), and Fingerprint Set (N = 64). Goodness of fit (misses): χ
2
(8) = 19.413, p<.0128. Goodness of fit (correct rejections): χ
2 (8) = 30.041, p<.001.
61
Table 12. A summary of the two level logistic multilevel model results to what predictors
influence the confidence level participants’ had in their decisions.
Fixed effect Estimate S.E. df t value p value
Intercept 8.093 1.059 241.7 7.642 <.001
Procedure -0.463 0.217 221.4 -2.135 0.034
Ambiguity -0.902 0.216 223.1 -4.172 <.001
Notes. All values rounded to 3 dp. S.E. = Standard Error of the mean. Context Type as a
categorical control variable. Level 2 grouping variables: Participant (N = 234), and
Fingerprint Set (N = 64).
Table 13. A summary of the three way ANOVA with Context Presence, Ambiguity Level, and
Procedure Type as factors with two levels, and d´ as the outcome variable.
Subset Parameter F value p value
Full dataset Context Presence 2.113 .147
Procedure Type 12.664 <.001
Ambiguity Level 4.616 0.033
Context x Procedure 0.545 0.461
Context x Ambiguity 0.262 0.609
Procedure x Ambiguity 16.447 <.001
Three way interaction 0.659 0.418
More Ambiguous
Materials only
Context Presence 2.310 0.131
Procedure Type 34.661 <.001
Two way interaction 1.437 0.233
Less Ambiguous
Materials only
Context Presence 0.379 0.539
Procedure Type 0.106 0.746
Two way interaction 0.002 0.962
Notes. All values rounded to 3 dp. Adj. R2 = 0.116.
62
APPENDIX A. FINGERPRINT SETS
Figure 1. These are materials in the Set 1 fingerprints for the standard procedure, which are
in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
63
Figure 2. These are materials in the Set 1 fingerprints for the filler-control procedure, which
are in the more ambiguous condition. The match is in the top lineups (Sample 2) and the no
match comparison sample is in the bottom lineup (Sample 2).
64
Figure 3. These are materials in the Set 2 fingerprints for the standard procedure, which are
in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
65
Figure 4. These are materials in the Set 2 fingerprints for the filler-control procedure, which
are in the more ambiguous condition. The match is in the top lineups (Sample 4) and the no
match comparison sample is in the bottom lineup (Sample 4).
66
Figure 5. These are materials in the Set 3 fingerprints for the standard procedure, which are
in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
67
Figure 6. These are materials in the Set 3 fingerprints for the filler-control procedure, which
are in the less ambiguous condition. The match is in the top lineups (Sample 3) and the no
match comparison sample is in the bottom lineup (Sample 3).
68
Figure 7. These are materials in the Set 4 fingerprints for the standard procedure, which are
in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
69
Figure 8. These are materials in the Set 4 fingerprints for the filler-control procedure, which
are in the less ambiguous condition. The match is in the top lineups (Sample 6) and the no
match comparison sample is in the bottom lineup (Sample 6).
70
Figure 9. These are materials in the Set 5 fingerprints for the standard procedure, which are
in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
71
Figure 10. These are materials in the Set 5 fingerprints for the filler-control procedure, which
are in the more ambiguous condition. The match is in the top lineups (Sample 5) and the no
match comparison sample is in the bottom lineup (Sample 5).
72
Figure 11. These are materials in the Set 6 fingerprints for the standard procedure, which are
in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
73
Figure 12. These are materials in the Set 6 fingerprints for the filler-control procedure, which
are in the more ambiguous condition. The match is in the top lineups (Sample 4) and the no
match comparison sample is in the bottom lineup (Sample 4).
74
Figure 13. These are materials in the Set 7 fingerprints for the standard procedure, which are
in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
75
Figure 14. These are materials in the Set 7 fingerprints for the filler-control procedure, which
are in the less ambiguous condition. The match is in the top lineups (Sample 2) and the no
match comparison sample is in the bottom lineup (Sample 2).
76
Figure 15. These are materials in the Set 8 fingerprints for the standard procedure, which are
in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
77
Figure 16. These are materials in the Set 8 fingerprints for the filler-control procedure, which
are in the less ambiguous condition. The match is in the top lineups (Sample 3) and the no
match comparison sample is in the bottom lineup (Sample 3).
78
Figure 17. These are materials in the Set 9 fingerprints for the standard procedure, which are
in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
79
Figure 18. These are materials in the Set 9 fingerprints for the filler-control procedure, which
are in the more ambiguous condition. The match is in the top lineups (Sample 4) and the no
match comparison sample is in the bottom lineup (Sample 4).
80
Figure 19. These are materials in the Set 10 fingerprints for the standard procedure, which
are in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
81
Figure 20. These are materials in the Set 10 fingerprints for the filler-control procedure,
which are in the more ambiguous condition. The match is in the top lineups (Sample 6) and
the no match comparison sample is in the bottom lineup (Sample 6).
82
Figure 21. These are materials in the Set 11 fingerprints for the standard procedure, which
are in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
83
Figure 22. These are materials in the Set 11 fingerprints for the filler-control procedure,
which are in the less ambiguous condition. The match is in the top lineups (Sample 1) and the
no match comparison sample is in the bottom lineup (Sample 1).
84
Figure 23. These are materials in the Set 12 fingerprints for the standard procedure, which
are in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
85
Figure 24. These are materials in the Set 12 fingerprints for the filler-control procedure,
which are in the less ambiguous condition. The match is in the top lineups (Sample 5) and the
no match comparison sample is in the bottom lineup (Sample 5).
86
Figure 25. These are materials in the Set 13 fingerprints for the standard procedure, which
are in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
87
Figure 26. These are materials in the Set 13 fingerprints for the filler-control procedure,
which are in the more ambiguous condition. The match is in the top lineups (Sample 4) and
the no match comparison sample is in the bottom lineup (Sample 4).
88
Figure 27. These are materials in the Set 14 fingerprints for the standard procedure, which
are in the more ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
89
Figure 28. These are materials in the Set 14 fingerprints for the filler-control procedure,
which are in the more ambiguous condition. The match is in the top lineups (Sample 3) and
the no match comparison sample is in the bottom lineup (Sample 3).
90
Figure 29. These are materials in the Set 15 fingerprints for the standard procedure, which
are in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
91
Figure 30. These are materials in the Set 15 fingerprints for the filler-control procedure,
which are in the less ambiguous condition. The match is in the top lineups (Sample 4) and the
no match comparison sample is in the bottom lineup (Sample 4).
92
Figure 31. These are materials in the Set 16 fingerprints for the standard procedure, which
are in the less ambiguous condition. The match is on the left and the no match comparison
sample is on the right.
93
Figure 32. These are materials in the Set 16 fingerprints for the filler-control procedure,
which are in the less ambiguous condition. The match is in the top lineups (Sample 3) and the
no match comparison sample is in the bottom lineup (Sample 3).
94
APPENDIX B. INSTRUCTIONS
Instructions given to participants in the standard procedure:
“You will receive eight folders from eight different criminal cases.
Each folder will contain all the materials you will need for your analysis.
For each case there will be a fingerprint that was lifted from the crime scene.
In addition to the print lifted from the crime scene, there will be one other fingerprint labeled
“Comparison sample”.
The comparison sample is the suspect. Keep in mind that the suspect might or might not be
the culprit – that is what you are trying to help determine.
Your task, for each case, is to decide if the comparison print matches the one from the crime
scene.”
Instructions given to participants in the filler-control procedure:
“You will receive eight folders from eight different criminal cases.
Each folder will contain all the materials you will need for your analysis.
For each case there will be a fingerprint that was lifted from the crime scene.
In addition to the print lifted from the crime scene, there will be six other fingerprints labeled
“Sample 1”, “Sample 2” and so on.
Only one of the six sample prints is from someone who is the suspect in that case. For
comparison purposes, the other five prints are from people who we know did not commit the
crime in question. You are not told which of the six prints is from the suspect. Keep in mind
that the suspect might or might not be the culprit – that is what you are trying to help
determine.
Your task, for each of the eight cases, is to decide if any one of those six prints matches the
one from the crime scene and, if so, which one matches.”
Additional instructions for those in the Context-present condition:
“You will also receive a police case report containing details and some background of each
case to help with your decision.”
95
APPENDIX C. CONTEXTUAL BIAS MATERIALS
Figure 33. Contextual information in the form of a mock police case report describing
incriminating information relating to a kidnapping.
96
Figure 34. Contextual information in the form of a police case report describing
incriminating information relating to a rape/sexual assault.
97
Figure 35. Contextual information in the form of a police case report describing
incriminating information relating to an extortion case.
98
Figure 36. Contextual information in the form of a police case report describing
incriminating information relating to a bomb threat.
99
Figure 37. Contextual information in the form of a police case report describing
incriminating information relating to an arson case.
100
Figure 38. Contextual information in the form of a police case report describing
incriminating information relating to an identity theft case.
101
Figure 39. Contextual information in the form of a police case report describing
incriminating information relating to a homicide.
102
Figure 40. Contextual information in the form of a police case report describing
incriminating information relating to an armed robbery.
103
APPENDIX D. IRB ETHICS APPROVAL

Order | Check Discount