Access to this full-text is provided by Wiley.
Content available from Campbell Systematic Reviews
This content is subject to copyright. Terms and conditions apply.
Received: 15 September 2023
|
Accepted: 30 August 2024
DOI: 10.1002/cl2.1441
SYSTEMATIC REVIEWS
Crime and justice
Interview and interrogation methods and their effects
on true and false confessions: A systematic review update
and extension
Mary Catlin
1
|David Wilson
1
|Allison D. Redlich
1
|Talley Bettens
1
|
Christian Meissner
2
|Sujeeta Bhatt
3
|Susan Brandon
4
1
Criminology, Law and Society, George Mason
University, Fairfax, Virginia, USA
2
Department of Psychology, Iowa State
University, Ames, Iowa, USA
3
Institute for Defense Analyses, Alexandria,
Virginia, USA
4
New York, New York, USA
Correspondence
Mary Catlin, Criminology, Law and Society,
George Mason University, Fairfax, VA, USA.
Email: mcatlin@gmu.edu
Funding information
High‐Value Detainee Interrogation Group,
USA, Grant/Award Number:
15F06720C0001969
Abstract
Background: False confessions are often the product of an interrogation process,
and the method by which an interrogation is conducted likely affects both the rate
of truthful confessions and false confessions. An optimal interrogation method will
maximize the former and minimize the latter.
Objectives: The current study was a partial update and extension of Meissner and
colleagues' (2012) prior Campbell systematic review titled Interview and Interrogation
Methods and their Effects on True and False Confessions. Our objective was to assess the
effects of interrogation approach on the rates of true and false confessions for criminal
(mock) suspects.
Search Methods: PsycINFO, Criminal Justice Abstracts, and 15 other databases
were searched starting October 20, 2022, with the final search conducted on May
23, 2023; together with reference checking, citation searching, and contact with
authors to identify additional studies.
Selection Criteria: All eligible studies experimentally manipulated interrogation
approach (i.e., accusatorial, information‐gathering, or direct questioning) were con-
ducted with mock suspects accused of wrongdoing where ground truth was known,
and included information about confession rates.
Data Collection and Analysis: We used standard methodological procedures
expected by The Campbell Collaboration for our selection of studies and data
collection. However, we developed our own risk of bias items and analyzed our data
using network meta‐analysis methods. Data were synthesized via random‐effects
network meta‐analysis based on the logged odds ratio.
Main Results: Across the 27 research articles that provided statistical information
sufficient to calculate an effect size, 29 individual studies provided a total of
81 effect sizes. Most studies were conducted with college students in the United
Campbell Systematic Reviews. 2024;20:e1441. wileyonlinelibrary.com/journal/cl2
|
1of37
https://doi.org/10.1002/cl2.1441
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2024 The Author(s). Campbell Systematic Reviews published by John Wiley & Sons Ltd on behalf of The Campbell Collaboration.
OSF link: https://osf.io/cz7a2/
States. Overall, our risk of bias assessment indicated that authors generally
adhered to double‐blind procedures and avoided selective reporting of outcomes.
Of note, however, it was often unclear how violations of the randomization
process were dealt with.
For true confessions, there were 12 studies estimating the effect between accusa-
torial and direct questioning, five estimating the effect between information‐
gathering and direct questioning, and another five estimating the effect between
accusatorial and information‐gathering. Compared to information‐gathering, on
average, the accusatorial conditions observed fewer true confessions, although not
statistically significant (combined OR = 0.55, 95% CI 0.29, 1.05). The largest effects
were between information‐gathering and direct questioning, with the former pro-
ducing significantly more true confessions on average (combined OR = 2.43, 95% CI
1.29, 4.59). This model showed good consistency between the direct and indirect
effects.
For false confessions, there were 20 studies estimating the effect between accu-
satorial and direct questioning, 4 studies estimating the effect between information‐
gathering and direct questioning, and 7 estimating the effect between accusatorial
and information‐gathering. On average, accusatorial conditions yielded more false
confessions than direct questioning (combined OR = 3.03, 95% CI 1.83, 5.02) or
information‐gathering (combined OR = 4.41, 95% CI 1.77, 10.97), both of which are
statistically significant. In contrast, direct questioning and information‐gathering had
roughly similar rates of false confessions with nonsignificant and small effects that
slightly favored information‐gathering (combined OR = 0.69, 95% CI 0.27, 1.78). This
model showed good consistency between the direct and indirect effects.
For true confessions under a six‐node model, most of the direct, indirect, and
combined network estimated mean odds ratios were not statistically significant. The
only significant effects were for (1) information‐gathering versus direct questioning,
with the former resulting in more true confessions (combined OR = 2.57, 95% CI
1.38, 4.78); and (2) accusatorial‐evidence ploy versus information‐gathering with the
former resulting in fewer true confessions (combined OR = 0.37, 95% CI 0.16, 0.84).
For false confessions under a six‐node model, we found significant effects for (1)
accusatorial‐evidence ploys versus direct questioning, with the former resulting in
more false confessions (combined OR = 2.98, 95% CI 1.59, 5.59); (2) accusatorial‐
evidence ploys versus information‐gathering, with the former resulting in more false
confessions (combined OR = 4.47, 95% CI 1.46, 13.68); (3) accusatorial‐other versus
direct questioning, with the former resulting in more false confessions (combined
OR = 3.12, 95% CI 1.37, 7.10); (4) accusatorial‐other versus information‐gathering,
with the former resulting in more false confessions (combined OR = 4.67, 95% CI
1.61, 13.55); and (5) information‐gathering versus minimization, with the latter
resulting in more false confessions (combined OR = 0.25, 95% CI = 0.08, 0.83). No
other combined effects were significant. This model should be interpreted cau-
tiously, however, as the Qstatistics raised concerns regarding model consistency.
Authors' Conclusions: Overall, results support calls for reforming policies related
to interviewing and interrogation practices to prohibit the use of accusatorial ap-
proaches and require the adoption of approaches that are science‐based.
2of37
|
CATLIN ET AL.
1|PLAIN LANGUAGE SUMMARY
Information‐gathering approaches increase true confessions and
decrease false confessions.
1.1 |The review in brief
Information‐gathering approaches increase the number of true
confessions and decrease the number of false confessions when
compared to both accusatorial and direct questioning interrogation
approaches.
1.2 |What is this review about?
Ideally, interrogation approaches should maximize the number
of true confessions obtained while minimizing the number of
false confessions. Problematically, however, some interrogation
approaches, such as accusatorial approaches, have been tied to
false confessions which have contributed to approximately 12% of
known wrongful convictions in the United States. An alternative
approach, known as information‐gathering, has considerable
evidence showing its utility in reducing false confessions. For the
purposes of this review, we compared accusatorial, information‐
gathering, and direct question interrogation approaches. This
review examined whether interrogation approaches have different
effects on the rates of true and false confessions. The review
considered evidence regarding the debate about whether the
United States should move towards more information‐gathering
interrogation approaches.
1.3 |What is the aim of this review?
This Campbell systematic review examined the effects of accusa-
torial interrogations on true and false confessions, compared to the
effects of information‐gathering and direct questioning interroga-
tions. The review summarizes evidence from 29 high‐quality
experimental laboratory studies.
1.4 |What are the main findings of this review?
1.4.1 |What studies were included?
This review included studies that evaluate the effects of interrogation
approach on true and false confessions. A total of 27 studies were
identified. One article reported three independent studies, thus
leading to 29 independent studies included in analyses. The studies
span the period from 1996 to 2023 and were conducted largely in
the United States.
1.4.2 |What interrogation approach produces the
most true confessions?
Information‐gathering approaches, on average, produce more true
confessions than both accusatorial and direct questioning interroga-
tion approaches. The largest difference is between information‐
gathering and direct questioning approaches.
1.4.3 |What interrogation approach produces the
most false confessions?
Accusatorial approaches, on average, produce more false confessions
than both information‐gathering and direct questioning interrogation
approaches. The rates of false confessions between information‐
gathering and direct questioning interrogation approaches were
similar.
1.5 |What do the findings of this review mean?
Overall, results support calls for practice to continue moving away
from accusatorial interrogation approaches. Future research should
continue directly comparing accusatorial and information‐gathering
approaches; furthermore, future research should consider a more
nuanced exploration of accusatorial and information‐gathering tactics
and their effect on true and false confessions.
1.6 |How up‐to‐date is this review?
The review authors searched for studies up to 2023.
2|SUMMARY OF FINDINGS
SUMMARY OF FINDINGS (Table 1).
3|BACKGROUND
3.1 |Description of the problem
In 1999, Victoria Bell Banks convinced law enforcement to release
her from custody based on a claim that she was not receiving ade-
quate prenatal care while in jail. Later, when the Choctaw County
Sheriff came looking for a baby, Ms. Banks claimed she had a mis-
carriage. The sheriff was suspicious and eventually began questioning
Ms. Bank's estranged husband, Medell Banks, about the infant. After
several days of intense interrogation, Mr. Banks falsely confessed to
the child being born alive and buried near his property. It was not
until medical doctors examined Victoria and determined that she was
CATLIN ET AL.
|
3of37
physically incapable of being pregnant that Medell was exonerated, 2
years after being convicted and imprisoned (National Registry of
Exonerations & Banks, 2021). It may be surprising that someone
would confess to a crime they did not commit (Leo, 2009), let alone
a crime that never took place. Researchers, however, have long rec-
ognized the problematic nature of false confessions (e.g., Münster-
berg, 1908) and false confessions have been identified all around the
world (Gudjonsson, 2018), including Europe, Asia, Australia, and New
Zealand.
False confessions pose several challenges for the criminal legal
system (e.g., Kassin, 2014; Scherr et al., 2020). First, it may be diffi-
cult to distinguish between true and false confessions. For example,
prisoners were recorded providing one confession to a crime they
had committed and to which they had been convicted, and a second
confession to a fabricated crime that they had not committed. These
records were then presented to college students and police officers
who were asked to identify whether the confession they viewed was
true or false. While the police officers claimed to be more confident
in their judgments about the veracity of the confessions, they were
no more accurate than the college students, and both groups per-
formed poorly overall (Kassin et al., 2005).
Second, confessions—whether true or false—tend to persuade
juries and judges more than many other pieces of evidence. Mock jury
studies have found that confessions are seen as more inculpatory than
any other form of evidence (e.g., Kassin & Neumann, 1997)andthat
even when individuals are informed that a confession was coerced and
should legally be disregarded, the confession still influences guilt
decisions (Kassin & Sukel, 1997). In fact, though jurors seem able to
distinguish between coerced and non‐coerced confessions, they still
assign meaningful attributions of guilt when an interrogation produces
a confession as both types of confessions are significantly more likely
to result in a conviction (Mindthoff et al., 2024).
Finally, research examining forensic confirmation biases (see
Kassin et al., 2013) has found that knowing a confession has been
obtained can influence forensic experts charged with determining
whether fingerprints match (Kukucka & Kassin, 2014) and the con-
clusions of DNA tests (Dror & Hampikian, 2011). Specifically, infor-
mation about a confession increases the chances that forensic ex-
aminers will determine that a suspect cannot be ruled out or that the
suspect is in fact a match to the sample in question. The undue
influence of confessions on other evidence creates two problems: (1)
legal actors may be unaware that they have been influenced by a
(false) confession, and (2) the course of an investigation could be
altered by this unintentional strengthening of confirmation biases.
Once a suspect confesses and other available evidence is perceived
as pointing to the same suspect, it is less likely that an investigator
will use their limited resources to continue pursuing other potential
leads.
This snowball effect of false confessions is described by the
Cumulative Disadvantage Framework (CDF; Scherr et al., 2020). The
CDF states that once a false confession occurs, several cognitive
processes activate in legal actors and decision‐makers that can cul-
minate in a wrongful conviction. First, when an innocent suspect
TABLE 1 Summary of findings.
Summary of findings
Finding Number of studies Odds ratio Certainty Comments
An accusatorial approach results in more false confessions
compared to an information gathering approach
7 4.41 (1.77 to 10.97) Moderate Direct, indirect and network effect are highly similar.
There is a risk of publication selection bias.
An accusatorial approach results in more false confessions
compared to direct questioning
20 3.03 (1.83 to 5.02) Moderate Direct, indirect and network effect are highly similar.
There is a risk of publication selection bias.
An information gathering approach results in more true
confessions compared to direct questioning
5 2.43 (1.29 to 4.59) Moderate Number of studies is low. Direct and network effect
are statistically significant.
An information gathering approach results in more true
confessions compared to accusatory
5 0.55 (0.29 to 1.05) Low Effect size is meaningful in size but not statistically
significant.
4of37
|
CATLIN ET AL.
confesses, law enforcement is less motivated to track down other
leads and may turn their attention almost exclusively to the suspect
who confessed. Second, that confession can then taint other forensic
evidence making the innocent suspect appear even more guilty.
Third, if the innocent suspect does not plead guilty, at trial it is
unlikely that jurors (and even judges) will be able to ignore the con-
fession evidence once it is introduced, even if instructed to do so.
Regardless, in both plea and trial settings, confessions tend to serve
as a powerful piece of evidence (see Mindthoff et al., 2024). It comes
as no surprise then that individuals end up wrongfully convicted
because of a false confession (Scherr et al., 2020). Beyond the harm
to the wrongfully convicted individual, wrongful convictions could
also mean that the true perpetrator has not been identified and could
continue committing crimes. Estimates are that an additional 41,000
crimes per year are committed in the United States by true perpe-
trators who continued to commit crimes after an innocent individual
was targeted for their wrongdoing (Norris et al., 2020). Thus,
wrongful conviction creates two wrongs: an innocent individual is
punished for a crime they did not commit, and a guilty offender
remains unpunished and at large in the community to potentially
commit new crimes (see Norris et al., 2020).
The National Registry of Exonerations (NRE) estimates that 12%
of the more than 3500 known wrongful convictions in the United
States involved a false confession (National Registry of Exonera-
tions, 2020). Of the 123 exonerations recorded across 17 European
countries by Eurex, 35% involved a false confession (European
Registry of Exonerations). A key driver of wrongful convictions are
false confessions, which largely occur during interrogations, and have
been shown to be influenced by the interrogation method used.
Evaluating the relative performance of the two dominant interroga-
tion methods (i.e., accusatorial and information gathering) can inform
police department policies, training regarding interrogation methods,
and the broader public debate surrounding police reforms.
3.2 |Description of the intervention
The method by which an interrogation is conducted likely affects
both the rate of truthful and false confessions. An optimal interro-
gation method will maximize the former and minimize the latter.
While there are a variety of methods that could be used to conduct
an interrogation, two general styles are an accusatorial approach and
an information‐gathering approach (Kelly et al., 2013; Meissner
et al., 2015). The current meta‐analysis updated and extended a prior
meta‐analysis that focused on accusatorial and information‐gathering
approaches to interrogations (Meissner et al., 2012,2014). Specific
techniques associated with each of these approaches are briefly
described in the following paragraphs, as they are the most common
approaches addressed in the literature (Denault & Talwar, 2023).
The Reid Technique is often referred to as the exemplar of the
accusatorial approach. As the most common interrogation manual
taught to law enforcement in the United States (Kostelnik &
Reppucci, 2009), the Reid technique starts with an assumption of
guilt and relies on law enforcement's ability to detect deception—
typically through nonverbal behavior such as avoiding eye contact.
Given the assumption of guilt, the goal of the interrogation is to
obtain a confession from the suspect (Inbau et al., 2013), through the
use of minimization (e.g., suggestions of leniency, justifications, and
rationalizations) and maximization (e.g., refusing to accept denials,
exaggerating the severity of the situation, exaggerating and/or fab-
ricating evidence) tactics. In general, minimization and maximization
envelope a range of tactics that interrogators can use in an attempt to
lessen the suspect's perception of culpability or punishment and
increase their perception of the severity of the offense and evidence
of guilt, respectively. Both the “soft sell”and “hard sell”approaches of
minimization and maximization are intended to elicit a confession
from a suspect (Kassin & McNall, 1991). Research has suggested that
minimization and maximization can have unique effects on confes-
sion rates (e.g., Guyll et al., 2019); therefore we examined the influ-
ence of these tactics both together as part of an accusatorial
approach and independently. For a more complete discussion of ac-
cusatorial tactics, see Kelly et al. (2013).
The PEACE model is an example of an information‐gathering
approach that has been adopted by the United Kingdom, Australia,
New Zealand, and Norway. The PEACE framework for investigative
interviews—regardless of the interviewee's status in the investigation
(e.g., witness, suspect, cooperative, noncooperative)—includes Plan-
ning and Preparation of the interview, Engaging the interviewee by
developing rapport and Explaining the ground rules of the interview
process, obtaining a complete Account of the incident under inves-
tigation, Closure of the interview, and Evaluation of the interview
process (see Bull & Rachlew, 2020; Milne et al., 2008). In the United
States, the High‐Value Detainee Interrogation Group (HIG), led by
the Federal Bureau of Investigation, has supported the development
of science‐based approaches to interrogation that incorporate an
information‐gathering framework (see Brandon & Meissner, 2023;
Meissner et al., 2017,2023). The purpose of an information‐
gathering interview is to elicit credible information from an inter-
viewee. To that end, PEACE and other information‐gathering models
focus on the establishment of rapport and trust between interviewer
and interviewee to elicit information, the use of cognitive inter-
viewing tactics (e.g., the Cognitive Interview) to enhance interview-
ees' memory search, and the strategic presentation of evidence
(Strategic Use of Evidence) to challenge the interviewees' account of
the crime and evaluate the credibility of the elicited information (see
Meissner et al., 2015; Swanner et al., 2016).
We also consider studies that use a control condition, often
referred to as direct questioning. Direct questioning, as it is meant to
serve as a control comparison in experimental research, typically
avoids the use of any particular interrogation tactic(s). Instead, direct
questioning involves short, declarative sentences/questions that
are goal oriented. For example, asking a participant directly if they
committed the act in question or simply asking them to sign a con-
fession statement. In this way, direct questioning is an analog to par-
ticipants' baseline behaviors. In other words, without the introduction
of tactics meant to induce a confession (i.e., accusatorial approaches)
CATLIN ET AL.
|
5of37
or elicit information disclosure (i.e., information‐gathering approaches),
direct questioning approaches allow researchers to estimate how often
people accused of wrongdoing truthfully or falsely admit guilt.
3.3 |How the intervention might work
Most confessions are elicited during an interrogation, with researchers
suggesting that the interrogation approach could be directly respon-
sible for false confessions (Ofshe & Leo, 1997). Specifically, two fac-
tors are believed to influence whether an interrogation will result in a
false confession from an innocent person: (1) if interrogators enter an
interrogation with an a priori assumption of the suspect's guilt and (2)
how rapport‐based versus coercive tactics are used during the inter-
rogation. First, assumptions of guilt can be problematic because
interrogators are more likely to use coercive interrogation tactics on
those they believe are guilty potentially leading to false confessions
(Narchet et al., 2011). Theoretically, then, interrogation approaches
that encourage assumptions of guilt will be more likely to induce false
confessions, while interrogation approaches that avoid assumptions of
guilt should minimize false confessions. Thus, false confessions are a
natural consequence of accusatorial interrogations (Mortimer &
Shepherd, 1999). In turn, a false confession can influence the inter-
pretation of forensic evidence by investigators, suggesting even more
“inculpatory”evidence than existed before the false confession (e.g.,
Kassin et al., 2013). From there, the chances of innocents accepting a
plea deal or being convicted at trial increase dramatically (e.g.,
Leo, 2009;Appleby&Kassin,2016). Even if exonerated, the false
confession holds severe consequences for exonerees attempting to
reenter society (e.g., Kukucka & Evelo, 2019).
In contrast, information‐gathering approaches are argued to
reduce false confessions by avoiding presumptions of guilt. Theoreti-
cally, by changing the goal of an interview from obtaining a confession
to eliciting investigative‐relevant information, these techniques avoid
the harmful actions associated with confirmation biases in interroga-
tion settings (e.g., Scherr et al., 2020).
Second, assumptions—or lack thereof—of guilt can influence
the use of rapport‐based vs. coercive tactics during interrogations.
Rapport, broadly defined, is the interpersonal connection established
between the interviewee and the interviewer. Rapport tactics are
typically intentional behaviors employed by interviewers to encourage
information disclosure from interviewees (see Gabbert et al., 2021).
When interviewers use rapport tactics, especially those aimed at
aligning the interviewee with the interviewer, interviewees are more
likely to perceive rapport. Such increased perceptions of rapport
increase the likelihood that interview subjects will decide to cooperate,
which in turn increases the amount of information disclosed by the
interviewee (Brimbal et al., 2019,2021; Dianiska et al., 2021). Rapport
approaches can be verbal, para‐verbal, or non‐verbal, though a
common tactic identified in the literature is to simply engage in active
listening. Importantly, information‐gathering approaches recognize
that rapport is not a static characteristic and can change over the
course of an interaction (Alison & Alison, 2017; Kelly et al., 2016).
In contrast, the Reid Technique only considers rapport as
important at the start of the interview phase (Inbau et al., 2013).
However, studies show that the use of rapport can, and likely will,
wane as more accusatorial approaches are introduced in an effort to
obtain a confession (Kelly et al., 2016). Furthermore, by itself, rapport
could be used as a form of minimization if the interviewer were to
falsely lead interviewees to believe that they were working in their
best interest (David et al., 2017; Vallano et al., 2022). Research has
shown that by addressing rapport throughout the interview,
information‐gathering approaches do, in fact, enhance interviewee
cooperation and produce more reliable information (Brimbal
et al., 2021; Russano et al., 2024; Vanderhallen & Vervaeke, 2014;
Walsh & Bull, 2012). Thus, because information‐gathering ap-
proaches both avoid assumptions of guilt and actively build rapport
throughout an interrogation, this approach should result in fewer
false confessions than accusatorial approaches.
3.4 |Why it is important to do this review
Meissner and colleagues (2012) conducted a Campbell systematic
review and meta‐analysis investigating rates of true and false con-
fessions across information‐gathering, accusatorial, and direct ques-
tioning (i.e., control) approaches for both field and experimental
(laboratory) studies. The synthesis of the field‐based studies found
that both information‐gathering and accusatorial interrogation
approaches were more likely to elicit a confession when compared to
direct questioning. These findings are limited, however, because field
studies cannot establish ground truth (actual guilt or innocence).
As such, estimated rates of false versus true confessions are not
possible, with the overall rate of obtaining a confession (regardless of
its reliability) as the only available metric.
In contrast, the findings from Meissner and colleagues' system-
atic review of the experimental studies suggested that while both
information‐gathering and accusatorial approaches increased the
likelihood of true confessions as compared to direct questioning (i.e.,
control), accusatorial approaches also increased the likelihood of false
confessions when contrasted directly with information‐gathering
approaches. The authors cautioned against strong claims regarding
the comparison of the two approaches because the analysis was
based on a small number of studies and they urged future researchers
to more explicitly compare information‐gathering and accusatorial
approaches (Meissner et al., 2012).
The current review is a partial update of Meissner and
colleagues' (2012) review, focused solely on experimental studies
to determine the diagnosticity (i.e., the ability to maximize true
confessions while minimizing false confessions) of interrogation ap-
proaches. Generally speaking, researchers rely on two experimental
paradigms to study interrogations: the alt‐key (Kassin & Kiechel,
1996) and the “cheating”(Russano, Meissner, et al., 2005) paradigm.
The alt‐key paradigm asks participants to complete a computer task
with strict instructions not to touch a specific button on the keyboard
that will supposedly cause the computer to crash. At some point in
6of37
|
CATLIN ET AL.
the study, the computer will appear to crash and the innocent par-
ticipant is then interrogated regarding the allegation of pressing the
forbidden alt‐key and asked to sign a confession (Kassin &
Kiechel, 1996). The cheating paradigm, on the other hand, involves a
confederate working through the study with the research participant.
The confederate–participant pair is instructed to complete a series of
individual and team problem‐solving questions. The confederate,
acting as a member of the research team, will either induce the
participant to help them with the individual problem‐solving or not.
When induced to assist in the individual problem‐solving task, the
participant has broken a rule of the experiment and is effectively
guilty of cheating. Here, both innocent and guilty participants are
then separated from the confederate and interrogated for the alleged
act of cheating before being asked to sign a confession (Russano,
Meissner, et al., 2005).
Experimental studies cannot ethically mirror certain aspects of
interrogations that happen in the real world and therein have lower
external validity than field studies. The tactics and scripts used in
experimental studies, however, are based on real‐world interrogation
manuals (e.g., Inbau et al., 2013), and are therefore designed to
induce meaningful physiological and psychological changes in the
accused participants (e.g., Guyll et al., 2019; Normile & Scherr, 2018).
As such, experimental studies attempt to model the underlying psy-
chological processes that exist in real interrogations. In addition, to
further increase external validity, researchers typically involve some
level of deception to maintain psychological realism (see Meissner
et al., 2010). Thus, it is reasonable to consider experimental studies as
analogous to real‐world interrogations to assess diagnosticity and
evaluate the efficacy of interrogative approaches. Such research,
thereby, offers evidence‐based considerations for the development
of policy related to law enforcement interrogation practices.
To our knowledge, there have been four other meta‐analyses
examining interview and interrogation approaches since 2012 (though
see Meissner, 2021, for meta‐analyses involving other aspects of
investigative interviews). The most recent meta‐analysis examined
methodological concerns raised by the “cheating”paradigm (Russano,
Meissner, et al., 2005) regarding the treatment of participants who
violate random assignment (Redlich et al., 2023). Results suggested
that maintaining random assignment through intent‐to‐treat analyses
attenuated, but did not eliminate, the effect of innocence on legal
decision‐making (i.e., Miranda waiver, confession, guilty plea). The
methodological concern informed our assessment of risk‐of‐bias but
otherwise is limited in its applicability to the current review as Redlich
et al. (2023) did not limit their investigation to studies resulting in
confessions. Another recent effort used meta‐analytic approaches to
examine the effect of rapport‐building and support tactics on chil-
dren's disclosure in forensic interviews (Lavoie et al., 2021). The focus
on child witnesses, however, limits the applicability of these results to
the current update (which focuses solely on juvenile and adult sus-
pects). The third meta‐analysis is more pertinent to our efforts as it
examined the prevalence of false confessions across experimental
paradigms. Results indicated that the alt‐key paradigm was most likely
to result in false confessions regardless of typing speed. Furthermore,
compared to all other tactics, false evidence ploys (i.e., lying or bluffing
to suspects about evidence) were more likely to result in false con-
fessions (Stewart et al., 2018). The last meta‐analysis examined the
social, cognitive, and affective factors associated with true and false
confessions obtained using the cheating paradigm (Russano, Meissner,
et al., 2005). Results demonstrated that false confessions were asso-
ciated with perceptions of the consequences of confessing and per-
ceptions of the interrogation context (Houston et al., 2014). None of
these meta‐analyses, however, looked at the influence of interrogation
approach on suspect true and false confessions. Therefore, updating
Meissner and colleagues' (2012) review was important, with particular
attention paid to experimental paradigms and different types of ac-
cusatorial tactics. Additional lab‐based research conducted since 2010
and advances in the methods of meta‐analysis over the past decade,
such as the development of network meta‐analysis and associated
software implementations, allowed us to extend prior synthesis work
in this area.
4|OBJECTIVES
The current study is a partial update and extension of Meissner and
colleagues' (2012) Campbell systematic review titled Interview and
Interrogation Methods and their Effects on True and False Confessions.
Our focus in the present meta‐analysis was solely on experimental or
laboratory‐based studies. Our objective was to assess the effects of
interrogation approach on true and false confession outcomes for
juvenile and adult criminal (mock) suspects.
To address our objective, a series of network meta‐analyses were
conducted, contrasting accusatorial, information‐gathering, and
direct questioning interrogation approaches on their ability to elicit
true and false confessions. Like its predecessor, the current meta‐
analysis focused on (mock) suspects as the population of interest,
interview style as the intervention, and the diagnosticity (i.e., ability
to increase true confessions while minimizing the number of false
confessions) of the interview styles as the indicator of effectiveness.
To expand on past work, the current effort also examined specific
accusatorial tactics (e.g., evidence ploy, minimization) for a more
nuanced understanding of the influence of such tactics on true and
false confessions. To accomplish this goal, a network meta‐analysis
was conducted to compare not only macro‐level interrogation ap-
proaches, but also how different techniques within the accusatorial
approach (i.e., minimization and maximization) compare to one another
and to other interrogation techniques regarding true and false con-
fession rates.
5|METHODS
Readers should be aware of a few differences between the protocol
and the current review. First, as anticipated, changes were made to
the coding protocol (see Supporting Information S6) during the pilot
coding protocol.
CATLIN ET AL.
|
7of37
Changes were made either for clarification or expansion of the
information captured. The only information removed from the coding
protocol was information about manipulations that were included in
the experimental studies beyond the interrogation approach and guilt
status of primary interest to this review. However, we include such
information in the description provided of each study in the Results
section. The coding took place in RedCap instead of LibreOffice as
originally anticipated. RedCap, like LibreOffice, allows for the one‐to‐
many hierarchy necessary for a meta‐analysis but also simplifies the
process of double‐coding.
Second, we originally intended to conduct traditional meta‐
analyses as the primary analyses and network‐analyses as sec-
ondary. Because network meta‐analysis provides direct effects,
indirect effects, and combined effects as output, the initially
planned approach seemed redundant. The direct effects duplicate
traditional univariate meta‐analytic estimates. Therefore, we focus
our analyses on the network meta‐analyses and offer the tradi-
tional meta‐analyses.
5.1 |Criteria for considering studies for this review
All criteria for study inclusion were based on the original review by
Meissner and colleagues (2012). However, we have noted where we
made departures from those original criteria.
5.1.1 |Types of studies
For this update, we included only experimental studies, regardless
of publication status, that randomly assigned mock subjects
(i.e.,notrealcriminaljusticesuspects in field studies) to two (or
more) interrogation (interview) conditions. The experimental
manipulation included the random assignment of an accusatorial
or information‐gathering interrogation approach. The two ap-
proaches could be compared with each other, compared to a
control interrogation technique (e.g., direct questioning). For
studies with only accusatorial techniques, we include some con-
trasts between minimization, maximization, and control tactics.
Participants (mock suspects—see below) could be entirely aware of
thenatureofthestudy(e.g.,somestudieschallengeparticipantsto
“get away”withanactofwrongdoing)orcouldbedeceivedto
various degrees (e.g., some studies lead participants to believe they
are facing academic consequences for the supposed wrongdoing).
Any experimental paradigm was eligible (e.g., cheating and alt‐key
paradigms) and studies could include more than one manipulated
factor; however, manipulated factors not pertinent to our review
were not analyzed.
The prior review by Meissner and colleagues (2012) also included
field‐based observational studies. These were excluded from this
review as our objective focused on the reliability of confessions,
which is not possible to assess with field studies (where ground truth
remains unknown).
5.1.2 |Types of participants
Participants were mock suspects who were accused of some
wrongdoing. Studies that included victims or witnesses of
wrongdoing exclusively were not eligible. Thus, only data relevant
to mock suspects from the included studies were considered for
this current analysis. However, the type of mock suspect was not
limited by race, age, ethnicity, gender, or any other demographic
characteristics.
5.1.3 |Types of interventions
For the purposes of this partial update, an interview or interrogation
method was defined as (1) an intentional use of one or more (2)
established interrogation tactics to (3) induce a confession. An
intentional use means that the intervention was part of the experi-
mental manipulations (see Section 5.1.1) where at least part of the
interrogation was scripted. In other words, studies where experi-
menters were allowed complete freedom in how they attempted to
get a confession were not eligible for the current review. By estab-
lished interrogation tactic, we mean those tactics that have been
associated with either an accusatorial or information‐gathering
approach. For example, false evidence ploys are associated with ac-
cusatorial methods and are included in accusatorial interrogation
manuals (Inbau et al., 2013). When not identified by name, an ac-
cusatorial technique was identified by its goal: to obtain a confession,
which is the same goal of the Reid technique (see Inbau et al., 2001).
There was only one instance in which the authors had to make this
assumption. In this case, the authors described the condition in
question as a “guilt bias condition,”which we labeled as accusatorial
given that accusatorial approaches, but not information‐gathering
approaches, are premised on an assumption of guilt. Accusatorial
techniques also include the use of minimization and maximization
tactics, which themselves encompass a wide range of more specific
tactics (see Kelly et al., 2015, for an overview). Conversely,
information‐gathering techniques were identified by their goal to
seek investigative‐relevant information, an example being the PEACE
model (see Milne & Bull, 1999). Information‐gathering techniques
often use cognitive interview and rapport‐building tactics. Any tactic
identified by Gabbert et al. (2021) was eligible. There was one
instance where a study was coded (by consensus) as containing an
information‐gathering condition though it was not explicitly called
that by the authors. In this case, the authors had a “rapport”and “no
rapport”condition. Given the emphasis of rapport‐building in
information‐gathering approaches, we coded the rapport condition as
information‐gathering. Direct questioning or control techniques
could involve elements of the other two techniques, but the primary
approach of control techniques is to use short, declarative sentences/
questions that are goal oriented. Finally, to determine that a tactics'
purpose was to induce a confession, the tactic must have been
introduced before the request for a confession. The temporal or-
dering of tactic and confession, which was largely described in the
8of37
|
CATLIN ET AL.
description of a study's paradigm or within Section 5, was evaluated
during the full‐text screening by two independent coders.
5.1.4 |Types of outcome measures
Primary outcomes
Eligible studies included either true or false (or both) confession rates
as the dependent variable. Note that we refer to our primary out-
come as confession rates to reflect the language used in the original
research, but technically these values are proportions as there is no
time element in the denominator. Depending on the study design
(e.g., some studies included exclusively innocent participants), the
study reported either the number of true confessions (i.e., confes-
sions provided by guilty participants), false confessions (i.e., confes-
sions provided by innocent participants), or both. The ground truth of
the confession (i.e., true vs. false confession) was the priority for the
coding stage. If a study contained both primary confession and sec-
ondary confession information, only the primary confession data was
considered. Primary confessions are those provided by the supposed
wrongdoer and were the outcome of interest. Secondary confessions
are typically defined as either an individual admitting they witnessed
an act of wrongdoing, which could make them complicit because
of their lack of action (e.g., Swanner et al., 2010), or an individual
conveying that the supposed wrongdoer confessed to them (e.g.,
Wetmore et al., 2014). In either case, the individual confessing was
not the supposed wrongdoer and thus was not considered in our
analyses.
Secondary outcomes
There were no secondary outcomes for this review.
5.1.5 |Duration of follow‐up
This was not relevant for the review. The dependent variable was
measured during a single laboratory session with the participant.
5.1.6 |Types of settings
There was no geographic limitation to study location. However, for
pragmatic reasons, only studies published in English were considered.
5.2 |Search methods for identification of studies
5.2.1 |Electronic searches
The listed databases and keywords were heavily influenced by the
meta‐analysis being replicated (Meissner et al., 2012). The keyword
combinations and filters, however, were generated by the first author
of the current effort. When possible, we searched across titles,
abstracts, author‐supplied keywords, and indexing terms. All searches
were limited to those results available in English. We did not restrict
our search or eligibility criteria by publication date, as all relevant
studies, even those captured in the original effort, were included
in our analyses. The following databases, organized by publisher
platforms, were searched:
ProQuest
1. Australia & New Zealand Database
2. Criminal Justice Database
3. ERIC
4. ProQuest Dissertations & Theses Global
5. Psychology Database
6. Social Science Database
7. Sociological Abstracts
8. Sociology Database
9. UK & Ireland Database
EBSCO
1. APA PsycExtra
2. APA PsycInfo
3. Criminal Justice Abstracts
4. National Criminal Justice Reference Services Abstracts
5. Psychology and Behavioral Sciences Collection
Web of Science
1. Conference Proceedings Index: Social Sciences & Humanities
2. Social Science Citation Index
Other
1. CINCH: Australian Criminology Database (Informit)
2. Google (first 10 pages of results)
3. Google Scholar (first 10 pages of results)
The following keywords were used in the systematic search (*
denotes that the word could have various endings; e.g., interrogat*
could be interrogation or interrogator):
1. Interrogat*
2. Information (gathering)
3. Inquisitorial
4. Interview*
5. Suspect*
6. Confess*
7. “Cognitive interview*”
8. “Conversation management”
9. “Ethical interview*”
10. Disclos*
11. “Strategic evidence”
12. Accusat*
CATLIN ET AL.
|
9of37
13. “Deception detection”
14. PEACE
15. PACE
16. Adversar*
17. Miranda
18. Coerc*
19. Entrap*
20. “Due process of law”
21. Reid
22. Minimi?*
23. Maximiz*
24. Random*
25. Control
26. Comparison
27. Experiment*
28. RCT
29. Manipulat*
30. Lab*
31. Factorial
32. Guilt*
33. Innocen*
34. Responsib*
35. Commit*
36. Effect
The keywords were condensed to the following search logic for
databases that allow for advanced Boolean logic: (interrogat* OR
information OR inquisitorial OR interview* OR accusat* OR “decep-
tion detection”OR PEACE OR PACE OR adversar* OR REID OR
minimi?* OR maximiz* OR “cognitive interview*”OR “conversation
management”OR “ethical interview*”OR “strategic evidence”OR
miranda OR coerc* OR entrap* OR responsib* OR commit*) AND
(random* OR control OR comparison OR experiment* OR RCT OR
manipulat* OR lab* OR factorial OR effect) AND (confess* OR dis-
clos*) AND (suspect* OR guilt* OR innocen*).
1
See Supporting
Information for the full searches. Databases were searched starting
October 20, 2022, with the final search conducted on May 23, 2023.
5.2.2 |Searching other resources
The reference lists for all studies deemed eligible for inclusion were
harvested for additional potentially eligible studies. Each eligible study
was also used for forward citation searching in Google Scholar. Further-
more, we used several resources to complement the search, including:
1. Meissner, C. A., Redlich, A. D., Bhatt, S., & Brandon, S. (2012).
Interview and interrogation methods and their effects on true and
false confessions. Campbell Systematic Reviews.https://doi.org/
10.4073/csr.2012.13 [Lists approximately 80 references]
2. Madon, S., More, C., & Ditchfield, R. (2019). Interrogations and
confessions. In N. Brewer & B. Douglass (Eds.) Psychological sci-
ence and the law. New York: Guilford Press. [Lists approximately
75 references]
3. Kassin, S. M., Drizin, S., Grisso, T., Gudjonsson, G., Leo, R. A., &
Redlich, A. D. (2010). Police‐induced confessions: Risk factors and
recommendations. Law and Human Behavior, 34(1), 3‐38. https://
doi.org/10.1007/s10979-009-9188-6. [Lists approximately 300
references]
4. Stewart, J. M., Woody, W. D., Pulos, S. (2018). The prevalence of
false confessions in experimental laboratory simulations: A meta‐
analysis. Behavioral Sciences & the Law,36(1), 12‐31. https://dor.
org/10.1002/bsl.2327. [Lists approximately 100 references]
5. High‐Value Detainee Interrogation Group. (2016). Interrogation: A
review of the Science. [Lists approximately 780 references]
Finally, the reviewers have many well‐established contacts with
researchers studying interviewing and interrogation in the United
States and abroad. In Supporting Information S5, we include a list of
researchers, both known and unknown, we contacted for
unpublished or “in press”studies to possibly include.
5.3 |Data collection and analysis
5.3.1 |Description of methods used in primary
research
The typical study randomly assigned mock suspects (volunteers, typi-
cally college/university students) to one of two or more experimental
conditions representing different interrogation methods. As mentioned
previously, the two most common experimental paradigms are the
cheating paradigm (Russano, Meissner, et al., 2005)andthealt‐key
paradigm (Kassin & Kiechel, 1996). In the cheating paradigm, interro-
gation methods were typically crossed with guilt status. Guilt status
was manipulated by a confederate who either did or did not induce
participants to help them solve an independent logic problem. In the
alt‐key paradigm, all participants were typically innocent as the com-
puter is programmed to crash regardless of participant action. Thus,
participants (randomly) assigned to be innocent can only produce false
confessions and participants (randomly) assigned to be guilty can only
produce true confessions, allowing true and false confessions rates to
be estimated from the same study depending on whether participants
were innocent, guilty, or randomly assigned to both guilt conditions.
Regardless of guilt, all participants were accused of some wrongdoing.
It is possible that the mock suspects could have been entirely aware of
the nature of the study (e.g., some studies challenged participants to
“get away”with an act of wrongdoing), but mock suspects were typ-
ically deceived to various degrees (e.g., some studies led participants to
believe they were facing academic consequences for the supposed
wrongdoing). At the point of accusation, studies typically manipulated
1
Note that we used the US spelling of maximization (maximiz*) in our search. This spelling
could have limited our search. However, given the expanse of search terms used, the fact
that most research on interrogations is conducted in the US, and maximization is a tactic
used in US interrogation approaches, we believe any impact likely negligible.
10 of 37
|
CATLIN ET AL.
the interrogation method through experimenter scripts before asking
for a signed confession from participants (see Stewart et al., 2018,for
an overview of typical experimental confession studies). We did not
discover any studies with non‐standard designs (e.g., longitudinal).
5.3.2 |Selection of studies
Two independent coders screened the titles and abstracts of all studies
identified in the digital search for potential eligibility using Abstrackr
(Abstrackr.cebm.brown.edu). The inter‐rater agreement (96.92%) was
acceptable. The first author trained the coders by going over the pre‐
registered protocol, practicing screening a subset together, and then
going over a pilot round of abstracts screened independently. Fur-
thermore, the coders met on a weekly basis to discuss any confusion
or disagreements as they arose. All studies deemed potentially eligible
(i.e., both raters answered “yes”or “maybe”when asked if the study
meets all eligibility criteria) were then assessed in their full form against
the eligibility criteria. This second round of screening was also con-
ducted by two independent coders who determined final eligibility.
The lead author further scanned the reference lists of eligible studies;
secondary search sources, such as prior reviews; Google; Google
Scholar; and the forward citation searches of eligible studies. Any
study identified by the lead author as potentially eligible during this
phase was evaluated for eligibility by a second independent coder.
5.3.3 |Data extraction and management
All studies deemed eligible for inclusion were coded for key variables
(e.g., effect size information) and study characteristics (e.g., publica-
tion type) by two independent coders. Discrepancies were resolved
through discussion, and when consensus could not be reached, one
of the lead reviewers made the final decision.
Coders were trained by the lead reviewers in steps: (1) coders
were verbally walked through the code sheet for discussion and
clarification, (2) the lead reviewers demonstrated how to code an
article in its entirety, and (3) the coders practiced on a small subset of
articles for review and feedback from the lead reviewers. This small
subset of articles was also used as a testing phase of the initial coding
protocol. Any changes made to the coding protocol at this point were
incorporated into the previously coded studies. This iterative process
continued throughout the coding process with regular meetings to
discuss coding issues.
Coders were the first and fourth authors, both of whom had prior
experience with meta‐analyses. Generally, coding included four hierar-
chical data levels: a study level, an experimental condition level, an
outcome level, and an effect size level. Using RedCap, we created a
database that allowed for the one‐to‐many hierarchical nature of our
coding protocol (e.g., one study could include several experimental
conditions, measure more than one outcome, and have several effect
sizes). Below is a more detailed description of the hierarchical data levels:
•Study level variables included static information (e.g., publication
type, publication year, geographic location). As such, there was
one record per study at this level of coding.
•Experimental condition‐level coding was conducted for each rel-
evant group of the research design. Thus, there was one record for
each eligible experimental condition within a study. For example, if
a study included a factor with three levels of interrogation tech-
niques (i.e., accusatorial vs. information‐gathering vs. direct
questioning), three condition coding sheets were completed to
capture each group. Information specific to each condition or full
sample (before or after attrition) was coded at this level, such as
sample size and interrogation method, with a variable to indicate
at what level of precision the information was reported.
•The outcome level included information specific to each eligible
outcome measure. Thus, there was one record per outcome.
•The effect size level included all necessary statistical information
to calculate a logged odds ratio (L
OR
) and its variance for each
outcome. As such, there was one record per coded effect size.
Coders were instructed to identify the most detailed numerical
data when coding for effect size information (see Supporting
Information S6 for the full coding sheet). When eligible studies did
not report all necessary data, we made a good‐faith effort to
contact authors to obtain the necessary information.
5.3.4 |Assessment of risk of bias in included studies
We assessed the risk of bias in the included studies through a combi-
nation of unique coding items we developed specifically for this research
literature and items adapted from the Cochrane risk‐of‐bias tool for
randomized trials (Higgins, 2019). The tool's language was modified to
better fit the characteristics of studies eligible for this review. We ex-
cluded items that were not relevant to this literature. The specific items
are in the coding protocol (see Supporting Information S6).
More specifically, the risk of bias items addressed the following
methodological issues: random assignment to both the interrogation
technique and guilt conditions, treatment of violations to the ran-
domization process (e.g., participants assigned to the guilty condition
who refused to cheat), attrition, level of deception employed in the
study, treatment of participants suspicious of the true purpose of the
study, and whether mock interrogators were blind to the guilt status
of mock suspects. To address the confession outcome, coders
documented any missingness of confession outcomes, including
selective reporting of confession outcomes.
5.3.5 |Measures of experimental effect
We used the odds ratio as the effect size index. The outcomes were
binary, and we were interested in comparing pairs of experimental
conditions. Thus, the data could be represented as a 2 × 2 contin-
gency table.
CATLIN ET AL.
|
11 of 37
5.3.6 |Unit of analysis issues
The unit of analysis was the individual study participant. We en-
countered no complex issues around the unit‐of‐analysis or unit‐of‐
assignment (also the participant) among the eligible designs.
5.3.7 |Criteria for determination of independent
findings
There were only two eligible outcomes for this review: the rate of
true confessions and false confessions. However, because these
studies may have any number of experimental conditions, numerous
effect sizes may be possible for each outcome. For example, if a study
had three conditions (accusatorial, information‐gathering, and direct
questioning), there were three possible pairings of these conditions
and, as such, three possible effect sizes for each outcome. We coded
all possible effect‐size combinations but maintained independence at
the analysis stage by performing a network meta‐analysis that takes
advantage of the network of comparisons provided by these studies.
Cell sizes were collapsed when necessary to create a single effect size
to maintain independence (i.e., collapsing across two or more accu-
satorial conditions). We did not discover any studies that measured
outcomes at multiple time points.
5.3.8 |Dealing with missing data
We contacted authors to request missing effect‐size data. Any study
that met all eligibility criteria but for which we were unable to
compute an effect size and were unable to get the needed data from
the authors was coded for descriptive information and reported.
5.3.9 |Assessment of heterogeneity
We assessed heterogeneity using the Q‐test and the I
2
statistic.
5.3.10 |Assessment of reporting biases
As part of our risk‐of‐bias tool, we noted if there was any indication in
a coded manuscript that one of the two outcomes of interest was
measured but not reported. The greater risk in this literature is pub-
lication bias, which we attempted to mitigate by purposefully seeking
unpublished work as well as published manuscripts. Publication bias
was assessed when there were at least ten effect sizes for a given
analysis. We did so in two ways: (1) a visual inspection of the funnel
plot and (2) an Egger's regression test. We assessed publication bias
using the three‐node networks. Note that these funnel plots center
the logged odds ratio around the mean for its contrast pair, enabling an
assessment of publication bias for the entire network. That is, both the
funnel plot and Egger's regression test are comparison‐adjusted.
5.3.11 |Data synthesis
Data were synthesized via random‐effects network meta‐analysis
based on the logged odds ratio. The models were estimated using the
restricted maximum likelihood (REML) estimator of τ
2
. We assumed a
common τ
2
for the network. A network meta‐analysis extends a tra-
ditional meta‐analysis by examining all available comparisons (both
direct and indirect) in a network. A network meta‐analysis is particu-
larly suited to our goal as it allowed us to compare the relative
effectiveness of interrogation techniques directly. Within this network,
we differentiated variations on the interview method, such as maxi-
mization and minimization accusatorial tactics. This separation is
important to investigate as some studies have found that minimization
tactics increase confessions (Guyll et al., 2019;Normile&
Scherr, 2018). Others find no differences between direct questioning
and minimization tactics (Woestehoff, 2016). Following suggestions
from the Campbell methods brief on network meta‐analysis, we
present a network diagram, information on inconsistency factors (i.e.,
tests of heterogeneity both within and between designs), and ranko-
grams of the interrogation approaches (see Wilson et al., 2016). The
network meta‐analyses were conducted in R using the netmeta pack-
age (Rücker et al., 2021).
5.3.12 |Subgroup analysis and investigation
of heterogeneity
Conducting a moderator analysis within the frequentist statistical
framework implemented in the netmeta package is currently not pos-
sible. It is possible, however, to do so within a Bayesian framework.
However, we opted to conduct the moderator analyses at the level of
contrast pairs rather than across the network as a whole. We did so
because our network is sparse, and our moderator variables do not
vary within each node. We conducted moderator analyses on the
paradigm type and on the risk of bias items to determine how potential
bias could influence the interpretation of our results. Moderator
analyses were conducted only when the total number of effect sizes
was at least 10 and each cell contained at least 2 effect sizes.
5.3.13 |Treatment of qualitative research
Qualitative research was not considered as part of this review.
5.3.14 |Summary of findings and assessment of the
certainty of the evidence
We provide a Summary of findings Table 1with the results of the
meta‐analyses. However, we did not use GRADE or a GRADE‐like
system as we did not believe it appropriate for this review. The focus
here was not assessing whether a treatment is effective or
ineffective. Rather, we were trying to establish the relative
12 of 37
|
CATLIN ET AL.
performance of different interview methods in eliciting reliable ver-
sus unreliable confessions. The confidence intervals around the mean
effect sizes were the first line of information on the certainty of the
evidence. The second line of information on the certainty of the
evidence was methodological weaknesses identified via our risk of
bias assessment. The overall results were interpreted within the
context of any weaknesses identified, particularly if they were
prevalent across studies.
6|RESULTS
6.1 |Description of studies
Table 2provides general information on the included studies. Most
studies were conducted in the United States (69%) with college
students (79%) and used the Russano, Meissner, et al. (2005) cheating
paradigm (59%). It is worth noting the variation in the number of
attempts made to elicit a confession: some participants were asked
once and others were asked more than three times. There also was a
lack of information regarding the identity of the individuals con-
ducting the interrogations.
Confession varied in how it was measured across these studies.
Specifically, our search captured studies that measured confession as
binary (yes/no); resistance (number of attempts to confession); some
combination of compliant (binary), internalized (indication that con-
fessor believed in their own guilt), and confabulated (confessor
fabricated details of their wrongdoing); and in terms of information
yield. All studies measured confession in a binary fashion, although
one study did not report these data. The binary confessed/did not
confess is the data used for all effect sizes. We did not analyze the
other versions of confession for two reasons: (1) once broken down
by interrogation approach, there were too few contrasts to conduct
meaningful analyses, and (2) for information yield, it was unclear at
what point a suspect could be considered as having confessed. In
other words, there is no agreement in the literature regarding
whether there is a meaningful difference in guilt judgment between
an individual who provides one unit of information and one who
provides two.
6.1.1 |Results of the search
Twenty‐seven eligible research articles were identified and coded.
Perillo and Kassin (2011) report on the results of three independent
studies. As such, 29 independent studies were available for analysis
(see Figure 1). The number of experimental conditions varied across
these 29 studies, with 16 having 2 conditions, 7 having 3 conditions,
2 having 4 conditions, and 2 having 5 conditions. All studies provided
adequate information for analysis or provided the information upon
request. Therefore, no studies were excluded for inadequate data.
Brief descriptions of the eligible studies and key ineligible studies are
provided below. In total, 81 effect sizes were calculated.
6.1.2 |Eligible studies
Blair (2007)
This study used the alt‐key paradigm described by Kassin and
Kiechel (1996) (see below) with some variations. In the Blair (2007)
study, undergraduate student participants were tasked with re-
sponding to pictures on a computer and instructed specifically to
avoid simultaneously touching the Control, ALT, and Delete keys.
Participants were randomly assigned to one of four conditions: pre-
sentation or no presentation of false evidence and minimization/
maximization or no minimization‐maximization interrogation tactics.
In the false evidence condition, the experimenter told the participants
that their activity on the computer was monitored by a remote
server, which showed that they had pressed the wrong keys. In
the minimization/maximization condition, the experimenter stated,
“Look, there is no doubt that you pressed the Control, Alt, and Delete
keys. That is the only way that this could happen. It has happened a
few times during this study. There are usually only two reasons for
someone to do something like this. Either they were just goofing
around to see what would happen or they were trying to ruin the
experiment. I want to believe that you were just goofing around, but
the only way I can know it is if you tell the truth and sign this paper.
Otherwise, I have to assume that you did it to ruin the experiment.”
(p. 178). No differences were found by condition.
Cole et al. (2013)
The purpose of this study was to replicate Kassin and Kiechel
(1996) using a different rule violation. Here, participants were
accused of breaking a lamp, an act that the authors argued is less
ambiguous than hitting an ALT computer key. Fifty‐five under-
graduate students were accused of breaking a lamp and randomly
assigned to either an incriminating false evidence condition (where a
confederate “eyewitness”falsely claimed to have seen the subject
break the lamp) or to a no false evidence condition (control). No
participants in either condition falsely confessed.
Eastwood et al. (2022)
This study used the cheating paradigm (Russano, Meissner,
et al., 2005) described below. Here, 80 participants were either
exposed to a dialog‐based or persuasion‐based interrogation before
being asked to confess. For our purposes, a dialog‐based approach
was considered an information‐gathering interrogation, and a
persuasion‐based approach, which used minimization (“Now, I am
sure you didn't mean any harm when you shared the answer with
the other participant, in fact, I bet you were just trying to be a good
person and help the other student out,”p. 3) and justification (“And
if asked for help, most people probably would have done the same
thing,”p.3) was considered accusatorial. Results indicated no dif-
ferences in the rate of true confessions; however, the persuasion‐
based approach elicited a significantly higher number of false
confessions. No innocent participants in the information‐gathering
interrogation condition confessed.
Evans et al. (2013)
This study aimed to develop a paradigm based on the cheating
paradigm that would allow for gathering empirical data in
CATLIN ET AL.
|
13 of 37
TABLE 2 Description of eligible studies.
Reference Publication type Paradigm Subject Attempts Interrogator Experimental conditions
Blair (2007) Thesis/Dissertation Alt‐Key College Students 2 Unclear Accusatory Evidence‐Ploy, Accusatory Min/Max, Accusatory
Other, Direct Questioning
Cole et al. (2013) Journal Other College Students 3 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Eastwood et al. (2022) Journal Cheating College Students 3 Unclear Accusatory Minimization, Information Gathering
Evans et al. (2013) Journal Other College Students 1 Unclear Accusatory Min/Max, Information Gathering
Guyll et al. (2019) Journal Cheating College Students 3 Unclear Accusatory Maximization, Accusatory Minimization
Hill et al. (2008) Journal Cheating College Students Unclear Unclear Accusatory Other, Direct Questioning
Huang and Teoh (2019) Journal Cheating College Students 1 Student Direct Questioning, Information Gathering
Kassin and Kiechel (1996) Journal Alt‐Key College Students 2 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Klaver et al. (2008) Journal Alt‐Key College Students 1 Unclear Accusatory Maximization, Accusatory Minimization
Meissner et al. (2011) Other Cheating Unclear 3 Unclear Accusatory Min/Max, Direct Questioning, Information Gathering
Narchet et al. (2011) Journal Cheating College Students 1 Unclear Accusatory Other, Direct Questioning
Noc et al. (2023) Journal Alt‐Key College Students Unclear Student Accusatory Min/Max, Information Gathering
Normile and Scherr (2018) Journal Cheating College Students 3 Unclear Accusatory Evidence‐Ploy, Accusatory Minimization
Normile et al. (2017) Other Cheating College Students 3 Unclear Accusatory Other, Direct Questioning
Paton et al. (2018) Journal Mock Crime Combination 3 Student Accusatory Minimization, Information Gathering
Perillo and Kassin (Study
1) (2011)
Journal Alt‐Key College Students 2 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Perillo and Kassin (Study
2) (2011)
Journal Alt‐Key College Students 2 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Perillo and Kassin (Study
3) (2011)
Journal Cheating College Students 2 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Redlich and Goodman (2003) Journal Alt‐Key Combination 2 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Rigoni (2007) Thesis/Dissertation Cheating College Students 3 Student Accusatory Other, Direct Questioning, Information Gathering
Russano, Meissner, et al. (2005) Journal Cheating College Students >3 Unclear Accusatory Minimization, Direct Questioning
Russano, Narchet, et al. (2005) Conference
Presentation
Cheating Unclear Unclear Unclear Accusatory Evidence‐Ploy, Direct Questioning
Smalarz et al. (2011) Conference
Presentation
Cheating Unclear 1 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Swanner et al. (2010) Journal Alt‐Key College Students Unclear Unclear Accusatory Evidence‐Ploy, Direct Questioning
14 of 37
|
CATLIN ET AL.
intelligence‐gathering contexts. Participants were exposed to a rule
violation when the confederate brought a cell phone into the room.
Innocent participants witnessed the confederate make a phone call
during the study, therefore breaking a minor rule, but no further
violation occurred. Guilty participants, however, (1) listened to the
confederate as they attempted to use an incomplete cheat sheet
provided by their roommate; (2) watched as the confederate realized
the sheet was incomplete and called their roommate to obtain the
remaining answers; and (3) copied all the answers to share with a
third friend. Thus, the transgression was complex and provided an
opportunity for an assessment of the amount of information yielded
from both innocent and guilty participants. In this study, 103 un-
dergraduates were exposed to either an information‐gathering or
TABLE 2 (Continued)
Reference Publication type Paradigm Subject Attempts Interrogator Experimental conditions
Villalba (2014) Thesis/Dissertation Cheating College Students 3 Student Direct Questioning, Information Gathering
Wachi et al. (2018) Journal Cheating Community Members 1 Practitioner Accusatory Evidence‐Ploy, Direct Questioning, Information
Gathering
Wilford and Wells (2018) Journal Cheating College Students 3 Unclear Accusatory Evidence‐Ploy, Direct Questioning
Woestehoff (2016) Thesis/Dissertation Cheating College Students 3 Unclear Accusatory Maximization, Accusatory Minimization, Direct
Questioning
Wright (2013) Thesis/Dissertation Other College Students 1 Unclear Accusatory Evidence‐Ploy, Direct Questioning
FIGURE 1 Prisma flow diagram.
CATLIN ET AL.
|
15 of 37
accusatorial interrogation approach. Participants were more likely to
confess when exposed to an information‐gathering interrogation, but
there was no effect of culpability (guilt or innocence).
Guyll et al. (2019)
This study measured 296 undergraduates' physiological reactions
to being accused of something they either did (i.e., guilty) or did not
(i.e., innocent) do. Furthermore, during the interrogation, participants
were either exposed to a low‐threat accusation where the violation
was described as “pretty minor”and “not a big deal”(p. 311) or a
high‐threat accusation where the rule violation was described as
“pretty serious”and a “major problem”(p. 311). For the purposes of
our study, the “low threat”was considered minimization, and “high
threat”was considered maximization. Participants were more likely to
confess when they were guilty and exposed to a low‐threat accu-
sation, although the interaction was not significant. The physiological
data suggested that the mobilization of resources (i.e., increased
physical reactions) aided participants in their resistance to confess,
which the researchers suggested explains the difference in confes-
sion rates between interrogation approaches.
Hill et al. (2008)
This publication consisted of three separate studies. Only Study
2 was eligible to be included here. Sixty‐four undergraduates self‐
selected as either innocent or guilty of cheating (accepting answers
from a confederate) during a laboratory task. Half the guilty and half
the innocent participants were questioned with guilt‐presumptive
questions (accusatorial style), whereas the other half were ques-
tioned with neutral questions (control), and confession outcomes
were measured. A main effect of interview style did not emerge for
either true or false confessions.
Huang and Teoh (2019)
This study aimed to distinguish between the effects of a
relationship‐based (where the focus is on a positive interrogator–
participant bond) and procedure‐based (where the focus is on
interrogation procedure and roles) rapport on true confessions. Using
a modified version of the cheating paradigm, all 87 undergraduate