He's Guilty!: Investigator Bias in Judgments of Truth and Deception

Abstract

Detecting deception is an inherently difficult task, but one that plays a critical role for law enforcement investigators in the interrogation room. In general, research has failed to indicate that performance in this domain is improved by training or prior experience. A signal detection framework is applied to the paradigm to better conceptualize the influence of these two factors. We found that although neither factor influenced discrimination accuracy, there was an effect on response bias such that training and prior experience appeared to increase the likelihood of responding "deceit" as opposed to "truth." This "investigator bias" was observed both in a review of the literature and in this study of North American law enforcement investigators who took part in a forensically based deception-detection task. Possible theoretical mechanisms and practical implications of these findings are discussed.

Full text

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
Law and Human Behavior, Vol. 26, No. 5, October 2002 (
C°
2002)
“He’s guilty!”: Investigator Bias in Judgments
of Truth and Deception
1
Christian A. Meissner
2,4
and Saul M. Kassin
3,4
Detecting deception is an inherently difficult task, but one that plays a critical role for
law enforcement investigators in the interrogation room. In general, research has failed
to indicate that performance in this domain is improved by training or prior experience.
A signal detection framework is applied to the paradigm to better conceptualize the
influence of these two factors. We found that although neither factor influenced dis-
crimination accuracy, there was an effect on response bias such that training and prior
experience appeared to increase the likelihood of responding “deceit” as opposed to
“truth.” This “investigator bias” was observed both in a review of the literature and in
this study of North American law enforcement investigators who took part in a foren-
sically based deception-detection task. Possible theoretical mechanisms and practical
implications of these findings are discussed.
KEY WORDS: deception; interrogation; response bias.
Police-induced confessions are a potent prosecutorial weapon that can have far-
reaching and rippling effects on the disposition of cases and on the criminal jus-
tice system as a whole. Indeed, modern police interrogations are so powerful that
they have, at times, elicited coerced-compliant and -internalized false confessions
from innocent people (Gudjonsson, 1992; Kassin, 1997; Leo & Ofshe, 1998; Radelet,
Bedau, & Putnam, 1992; Scheck, Neufeld, & Dwyer, 2000). Research suggests that
the process of interrogation is persuasive, if not too persuasive, in part because
it is explicitly based upon a presumption of guilt—an assumption that itself can
set in motion a number of cognitive and behavioral confirmation biases (Kassin,
Goldstein, & Savitsky, 2001). As described by Inbau, Reid, Buckley, and Jayne (2001)
1
Portions of this paper were presented at the 2002 American Psychology–Law Society Conference in
Austin, TX.
2
Department of Psychology, Florida International University, Miami, Florida.
3
Department of Psychology, Williams College, Williamstown, Massachusetts.
4
To whom correspondence should be addressed at Department of Psychology, Florida International Uni-
versity, University Park, Miami, Florida 33199; e-mail: meissner@fiu.edu or Department of Psychology,
Williams College, Williamstown, Massachusetts 01267; e-mail: skassin@williams.edu.
469
0147-7307/02/1000-0469/1
C
°
2002 American Psychology-Law Society/Division 41 of the American Psychology Association

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
470 Meissner and Kassin
in their manual, Criminal Interrogation and Confessions, “An interrogation is con-
ducted only when the investigator is reasonably certain of the suspect’s guilt” (p. 8).
Justifying the use of heavy-handed tactics, many law enforcement professionals
believe that they can determine whether a suspect is being truthful or deceptive on
the basis of a preinterrogation interview during which a suspect is questioned in a
nonconfrontational manner about his or her knowledge, whereabouts, and possible
involvement in the crime. Furthermore, many believe that they can learn to make
these initial judgments at high levels of accuracy by observing various aspects of the
suspect’s verbal and nonverbal behavior (Inbau et al., 2001; Vessel, 1998). According
to John E. Reid and Associates, for example, investigators trained in their Behavior
Analysis Interview are able to distinguish truth and deception at an 85% level of
accuracy (http://www.reid.com/service-bai-interview.html).
Unfortunately, psychological research has generally failed to support the claim
that individuals can attain high levels of performance in making judgments of truth
and deception. Over the years, numerous studies have demonstrated that individuals
perform at no better than chance level in detecting deception (DePaulo, Stone, &
Lassiter, 1985), that training programs produce only inconsistent improvements in
performance compared with control conditions (Bull, 1989; Kassin & Fong, 1999;
Porter, Woodworth, & Birt, 2000; Vrij, 1994; Zuckerman, Koestner, & Alton, 1984;
Zukerman, Koestner, & Colella, 1985), and that police investigators and others
with relevant on-the-job experience perform only slightly better, if at all (Bull,
1989; DePaulo, 1994; DePaulo & Pfeifer, 1986; Ekman & O’Sullivan, 1991; Ekman,
O’Sullivan, & Frank, 1999; Koehnken, 1987; Porter et al., 2000). Thus, although many
in the law enforcement community assume, often with great confidence, that inves-
tigators can use verbal and nonverbal behavioral cues to make accurate judgments
of truth and deception, there is little hard evidence to support this assumption.
Unfortunately, the implications of falsely inferring deceit in the interrogation
room can be quite costly to an innocent suspect. Consider, for example, the mili-
tary trial of U.S. v. PFC Timothy Bickel (1999), in which one of us testified as an
expert witness (the second author). In this case, a confession to rape was extracted
by a combination of five agents who used persistent and highly aggressive techniques
(e.g., explicit and implied promises and threats, negative feedback on a polygraph
test that was described as infallible, and the use of minimization) despite the ab-
sence of independent evidence against the defendant. When asked why they chose
to interrogate Bickel and not others, one investigator noted that he showed “signs
of deception based on the training we have received.” More specifically, this inves-
tigator stated, “His body language and the way he reacted to our questions told us
that he was not telling the whole truth. Some examples of body language is that he
tried to remain calm but you could tell he was nervous and every time we asked
him a question his eyes would roam and he would not make direct contact, and at
times he would act pretty sporadic and he started to cry at one time.” Correctly, we
believe, this defendant was acquitted at trial. There are many other instances too,
such as the case of Tom Sawyer, in Florida, in which investigators accused the de-
fendant of sexual assault and murder, interrogated him for 16 hr, issued threats, and
extracted a confession likely to have been false (cited in Ofshe & Leo, 1997). The
confession was suppressed by the trial judge, and, in the absence of other evidence,

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
Investigator Bias 471
the charges were dropped. Most intriguing to us was the manner in which Sawyer
became a prime suspect—his face flushed and he appeared embarrassed during an
initial interview, a behavioral reaction interpreted as a sign of deception. However,
what the investigators did not know was that Sawyer was a recovering alcoholic with
a social anxiety disorder that caused him to sweat profusely and blush in evaluative
social situations.
Given the profound implications for error at this early stage of criminal investi-
gations, it is important to understand the influence of training and prior experience
on the deception-detection performance of police investigators. Regrettably, how-
ever, a quick survey of the literature may itself prove somewhat deceptive, as many
studies have reported only global accuracy rates (i.e., percentage correct) across
all targets. We believe that such a method of analysis may fall short of provid-
ing a complete account of the performance of individuals as a function of lev-
els of experience or training. Rather, considering the influence of such manipu-
lations across both truthful and deceitful targets would allow greater insight into
processes underlying deception-detection decisions, particularly when placed within
the context of signal detection theory (SDT; Green & Swets, 1966; MacMillan &
Creelman, 1991). Unfortunately, in only a few cases have researchers actually pro-
vided such estimates of truthful and deceitful performance (Ekman et al., 1999;
Kassin & Fong, 1999; Porter et al., 2000; Vrij, 1993; Vrij & Mann, in press), and
in no cases have researchers directly applied SDT to assessing deception-detection
performance.
Although it is possible that training or experience may increase an individual’s
ability to discriminate between truthful and deceptive target persons, it may alter-
natively bias decisions toward truth or deceit on the basis of a variety of factors. As
conceived by Green and Swets (1966), SDT provides a framework for separating
performance into two conceptually and computationally independent parameters,
namely discrimination accuracy—the ability of an individual to correctly detect a
signal (deception) versus correctly reject its absence (truth), and response bias—the
degree of evidence necessary for the individual to respond that a signal (decep-
tion) has been presented. Although SDT has been widely used throughout psychol-
ogy, as in the study of decision making among air-traffic controllers, doctors, and
clinical psychologists, it has not been applied to the study of deception-detection
judgments (but see Thompson, 1978, regarding the detection of social cues). The
implication is that various targets, expectations, experiences, and other manipula-
tions thus far presented in the deception literature may differentially influence the
two SDT parameters. Knowing how these parameters are influenced should also
provide insight into the nature of the psychological effects underlying the various
manipulations.
With regard to the effects of training or prior experience, proponents seem to
suggest that certain manipulations would only influence discrimination accuracy by
enhancing one’s ability to correctly differentiate truth from deceit. This certainly
represents the aim of most training methods that seek to provide knowledge of
verbal and nonverbal cues inherent in deception. On the basis of the relative in-
dependence of the two SDT parameters (cf. Snodgrass & Corwin, 1988), it may
also be assumed that training and experience would not influence the response

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
472 Meissner and Kassin
bias parameter, that is, the individual’s tendency to overly respond “truth” versus
“deceit.”
EFFECT SIZE ANALYSIS OF PREVIOUS
DECEPTION-DETECTION RESEARCH
We first evaluated this hypothesis by conducting a brief survey and reanalysis
of existing research. In particular, we looked for studies that examined the influence
of either experience, such as law enforcement, or specific training in the detection of
deception. Our literature search was constrained by the requirements that studies
(a) report accuracy for both truthful and deceitful targets and (b) include a no-
training or no-experience control condition. For studies in which only gross accuracy
estimates were reported, authors were contacted for the additional relevant infor-
mation. Overall, we were able to locate four studies that examined the influence of
prior experience (Chahal & Cassidy, 1995; DePaulo & Pfeifer, 1986; Ekman et al.,
1999; Porter et al., 2000) and two that assessed the influence of training on deception-
detection performance (Kassin & Fong, 1999; Koehnken, 1987). Together, these six
studies represented the responses of 1,161 participants. Other studies were excluded
either because of insufficient reporting of data (deTurck , 1991; deTurck & Miller,
1990; Ekman & O’Sullivan, 1991; Kraut & Poe, 1980; Vrij, 1994; Zukerman et al.,
1984, 1985) or because they did not include a control condition (Vrij, 1993; Vrij
& Graham, 1997; Vrij & Mann, 2001). A studywise SDT analysis (MacMillan &
Creelman, 1990; MacMillan & Kaplan, 1985) was performed on the performance
means that were either reported in the paper or provided by the authors. Estimates
of hits (i.e., the proportion of deceptive persons correctly identified as such) and false
alarms (i.e., the proportion of truthful persons incorrectly identified as deceptive)
were used to compute studywise SDT estimates of both discrimination accuracy (A
0
)
and response bias (B
00
D
) for each condition (see Donaldson, 1992). Finally, the differ-
ence in performance between the experimental and control conditions was computed
by estimating the variance for each SDT parameter and computing Cohen’s d effect
size (see Rosenthal, 1994). Positive effect sizes for A
0
would indicate that participants
in the experimental condition (trained or experienced) outperformed the controls,
whereas positive effect sizes for B
00
D
would indicate that participants in the experi-
mental condition exhibited a more conservative response bias (i.e., higher likelihood
of responding “truthful” across all cases) than did controls.
Studywise estimates of discrimination accuracy (A
0
), response bias (B
00
D
), and ef-
fect size (Cohen’s d) are presented in Table 1. Across studies, the average weighted
effect size for the A
0
measure was Cohen’s d =−0.129, a non-significant effect size,
z = 1.92, with confidence intervals (−0.261, 0.003). In contrast, the B
00
D weighted
effect size was significant, Cohen’s d =−0.317, z = 4.75, p <.001, with confidence
intervals (−0.448, −0.186). Thus, across studies, both training and prior experience
engendered a more liberal response criterion (i.e., a bias toward responding “deceit-
ful”) when compared to participants in the no-training and no-experience control
conditions. In contrast to what one might have expected, however, training and ex-
perience had no significant effect on participants’ ability to accurately discriminate
truth from deceit.

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
Investigator Bias 473
Table 1. Estimates of Signal Detection Performance (A
0
and B
00
D
) and Effect Sizes (Cohen’s d ) for the
Influence of Experience or Training in a Deception-Detection Task
A
0
B
00
D
Study Manipulation NMCohen’s dMCohen’s d
Porter et al. (2000) Control vs. parole officers 32 vs. 32 0.31 −0.867 0.02 −1.050
0.10 −0.60
Ekman et al. (1999) Control vs. police officers 425 vs. 105 0.70 −0.005 −0.04 −0.342
0.69 −0.24
DePaulo & Pfeifer (1986) Control vs. police officers 161 vs. 258 0.59 −0.187 0.47 −0.208
0.54 0.35
Chahal & Cassidy (1995) Control vs. social workers 40 vs. 20 0.75 −0.041 0.16 −0.576
0.73 −0.18
Kassin & Fong (1999) Na¨ıve vs. trained 20 vs. 20 0.63 −0.451 0.14 −0.547
0.52 −0.18
Kohnken (1987) Na¨ıve vs. trained 20 vs. 60 0.41 −0.159 −0.60 −0.203
0.37 −0.48
EMPIRICAL VALIDATION OF THE “INVESTIGATOR BIAS” EFFECT
Our effect size analysis of the existing literature yielded an investigator bias ef-
fect, suggesting that training and prior experience lead to a perceptual bias toward
judgments of deceit. As this result was unexpected, and as previous studies had nei-
ther computed signal detection parameters nor predicted (a priori) such a finding,
we attempted to assess directly the validity of this investigator bias in a sample of
law enforcement officers. We also sought to extend the ecological validity of the
deception-detection task to more closely match the kinds of forensic judgments that
investigators routinely make. Of the studies that were included in our studywise anal-
ysis, only Kassin and Fong’s paradigm utilized stimuli that simulated an investigative
interview (Kassin & Fong, 1999). Other studies employed either (a) the classic de-
ceptive opinion/attitude paradigm (DePaulo & Pfeifer, 1986; Ekman et al., 1999),
(b) deceptive witnesses to an event (Chahal & Cassicy, 1995; Kohnken, 1987), or
(c) deceptive accounts of personal experiences (Porter et al., 2000). Although some
previous studies have developed more realistic, forensically relevant detection tasks,
these studies either did not include the necessary control condition (e.g., Vrij, 1993)
or failed to provide the statistics necessary to examine an investigator bias effect (e.g.,
Kraut & Poe, 1980). As noted earlier, our concern is that if investigators are truly
biased toward deception (or guilt) in their initial judgments, then preinterrogation
interviews could prompt the subsequent use of strong, pressure-filled methods of
interrogation that, in turn, could increase the risk of coerced false confessions.
To provide a true empirical test of the investigator bias hypothesis, we assessed
the deception-detection abilities of police investigators in the context of an inves-
tigative interview. In doing so, the present investigation extended recent work by
Kassin and Fong (1999), who experimentally trained some student participants, but
not others, in the detection of truth and deceit, before obtaining judgments of mock
suspects. Their study was unique in two ways. First, the authors administered the pop-
ular “Reid Technique,” which has been employed in the training of tens of thousands
of law enforcement personnel. This training purports to enhance deception-detection
ability through the use of various verbal and nonverbal behavioral cues (see Inbau

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
474 Meissner and Kassin
et al., 2001). Second, these authors created a set of forensically relevant stimulus tapes
depicting brief interviews and denials of individuals who were either truly guilty or
innocent of committing one of four mock crimes.
Kassin and Fong (1999) showed the videotapes to trained and na¨ıve student
participants who attempted to detect whether the mock suspects were lying or truth-
ful. Their results indicated that although training did not increase overall detection
accuracy, it did increase the confidence that trained students had in their judgments
as well as the number of reasons they cited as a basis for their judgments. As shown
in Table 1, our reanalysis of the data also indicated that the training procedure trig-
gered a response bias toward guilt. From a practical standpoint, of course, these
data are importantly limited by the fact that the observers were college students, not
police detectives, and that the training manipulation was brief and condensed, not
offered in the context of professional specialization. As such, this study was designed
to extend this paradigm by testing experienced police investigators, some of whom
had received training in interviewing, interrogation, and the detection of deception.
Our objectives were twofold: (1) to compare the judgments of police investigators to
Kassin and Fong’s trained and na¨ıve college students (Kassin & Fong, 1999) and (2) to
examine—within our samples of investigators—the correlations between experience,
training, and various indices of performance.
METHOD
Participants
Forty-four North American law enforcement investigators participated in the
study. Twenty-five investigators were affiliated with local police departments in
Florida, and 19 were from local departments in Ontario, Canada. No differences
were found in the experience or performance of the two samples. Investigators aver-
aged 13.7 years of law enforcement experience (SD = 6.5), and 68% had undergone
formal professional training in interviewing, interrogation, and the detection of the
deception.
Materials
Investigators were shown Kassin and Fong’s crime interview tapes, the details
of which are provided in their study (Kassin & Fong, 1999, pp. 501–504). Altogether,
eight guilty suspects committed one of four mock crimes (vandalism, shoplifting,
breaking and entering, a computer break-in), and eight innocent suspects were in-
structed merely to appear at the scenes of these crimes. All suspects were appre-
hended, taken to an interrogation room, and questioned by an adult male playing
the role of a detective. The interrogator, “Detective McCarthy,” was informed about
the mock crime that was committed but was blind to each suspect’s guilt or inno-
cence. The interviews, which were videotaped, ranged from 3.5 to 6.0 min in duration,
with a mean of 4 min 35 s. Self-reports of the participant suspects indicated that the
sessions were somewhat stress-provoking, and that the detective was intimidating in
his demeanor.

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
Investigator Bias 475
Table 2. Deception-Detection Performance of Students and Police Investigators
Kassin & Fong (1999)
Na¨ıve students Trained students Police investigators
(N = 20) (N = 20) (N = 44)
Judgment accuracy 56% (15%) 46% (17%) 50% (19%)
Confidence 5.91 (0.78) 6.55 (1.02) 7.05 (0.96)
No. of reasons citd 2.98 (0.82) 3.96 (0.85) 2.14 (0.71)
Hits 56% (17%) 45% (17%) 62% (21%)
False alarms 42% (34%) 60% (38%) 67% (26%)
A
0
0.63 (0.23) 0.52 (0.24) 0.48 (0.28)
B
00
D
0.14 (0.72) −0.18 (0.80) −0.46 (0.55)
Note. Standard deviations are provided in parentheses.
Procedure
In both samples, investigators were randomly assigned to view one of two video-
tapes, each depicting eight male suspects (four guilty and four innocent) repeat-
edly denying their involvement in the crime. After completing a brief background
questionnaire, investigators were instructed that they would view a tape of eight
male suspects, and that they were to decide whether each individual was lying or
truthful regarding his involvement. As in Kassin and Fong’s study, investigators were
told that between one fourth and three fourths of the suspects were lying to protect
themselves.
5
In addition to making a truth/deception judgment regarding each sus-
pect, investigators rated their confidence on a scale from 1 (not at all confident)to
10 (extremely confident), and wrote in their own words the basis for their judgment.
Given that Kassin and Fong’s student participants completed the same task under
identical circumstances, their performance will be compared with that of our police
investigators.
RESULTS
Table 2 presents the results for our police investigators as well as those of
Kassin and Fong’s na¨ıve and trained student participants who responded to the same
stimuli (Kassin & Fong, 1999). A one-way analysis of variance (ANOVA) was per-
formed on each performance variable for the three-way contrast of na¨ıve students
versus trained students versus police investigators. Planned comparisons then as-
sessed statistical differences in the performance of police investigators to that of the
student samples. The p <.05 convention for significance was employed across all
analyses.
5
This range of deception was provided to capture variability in the response bias parameter, and is
a generally accepted detection task instruction. Any systematic fluctuations in response bias across
conditions resulting from such an instruction would demonstrate the influence of differential strategies
or preconceived base rates. Alternatively, overt specification of the 50% deception used in this study
would have suppressed the response bias parameter and hindered any such variation across conditions.

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
476 Meissner and Kassin
Overall Judgment Accuracy, Confidence, and Reasons Cited
So that our results would be comparable with previous studies, we computed
judgment accuracy scores across all targets. As shown in previous studies, overall ac-
curacy (including both truthful and deceitful targets) for the investigators and student
participants did not significantly differ, F(2, 81) = 1.54, ns, η
2
= .03. However, a
significant effect on confidence was observed, F(2, 81) = 11.50, p <.001, η
2
= .22.
Planned comparisons revealed that both na¨ıve and trained students were significantly
less confident than police investigators, ts(62) = 4.87 and 2.16, ps <.001 and .05, re-
spectively. Interestingly, judgment accuracy and confidence were not significantly
correlated across investigators, r(43) = .20, ns, indicating poor calibration in detec-
tion performance. Participants were also asked to indicate their reasons for judging
the target as truthful versus deceitful. Results revealed a significant effect across
participant samples, F(2, 81) = 39.25, p <.001, η
2
= .49. Planned comparisons in-
dicated that both na¨ıve and trained students produced significantly more reasons
than did police investigators, ts(62) = 4.15 and 8.95, respectively, all ps <.001. Re-
sults of a comparative analysis of the combined sample of students versus police
investigators also replicated the above pattern.
Signal Detection Analysis
As described earlier, we separated performance into estimates of “hits” (the
proportion of deceitful suspects correctly identified as deceitful) and “false alarms”
(the proportion of truthful suspects incorrectly identified as deceitful). Analysis of hit
responses yielded a significant effect across participant samples, F(2, 81) = 5.38, p <
.01, η
2
= .12. Planned comparisons revealed that trained students produced signifi-
cantly fewer hits relative to the police investigators, t(62) = 3.18, p <.01; na¨ıve stu-
dents, however, did not significantly differ from the investigators, t(62) = 1.13, ns.A
significant main effect of participant sample was also observed with false alarm re-
sponses, F(2, 81) = 4.58, p <.01, η
2
= .10. Planned comparisons revealed that na¨ıve
students generated significantly fewer false alarms relative to police investigators,
t(62) = 3.25, p <.01, although the difference between trained students and investi-
gators was not significant, t(62) = 0.88, ns. When student samples were combined, re-
sults indicated that investigators generated significantly more hits, t(82) = 2.71, p <
.01, and more false alarms, t(82) = 2.00, p <.05.
When these estimates were used to compute measures of A
0
and B
00
D
, results
indicated a significant investigator bias effect, consistent with our studywise analy-
sis. Analysis of the discrimination accuracy parameter revealed no significant differ-
ences across participant samples, F(2, 81) = 2.11, ns, η
2
= .05. Yet a significant effect
of response bias was observed, F(2, 81) = 5.92, p <.01, η
2
= .13. Planned compar-
isons revealed that na¨ıve students significantly differed from police investigators,
t(62) = 3.68, p <.001. Although trained students demonstrated a more conserva-
tive response bias relative to police investigators, this difference failed to reach
conventional levels of significance, t(62) = 1.68, ns. A comparison of the combined
sample of students versus investigators also indicated no significant differences on
discrimination accuracy, t(82) = 1.61, ns, but an investigator bias effect in estimates
of response criterion, t(82) = 3.05, p <.01.

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
Investigator Bias 477
Correlational Analyses
Comparison of law enforcement professionals to college students is one way
to gauge the role of experience and training on deception-detection performance.
Within our sample of investigators, we also examined the correlations between the
amount of training and experience the investigatorshad received and key measures of
task performance. First, prior experience (years) in law enforcement was not related
to either overall judgment accuracy, r(44) =−.19, decision confidence,r (44) =−.07,
or the number of reasonsinvestigators cited for their decisions, r(44) =−.03. In terms
of signal detection parameters of performance, however, greater prior experience
was significantly correlated with both an increase in false alarm responses, r(44) =
.33, p <.05, and an increased bias in responding “deceit,” r(44) =−.34, p <.05. A
similar pattern emerged for the effect of deception-detection training with regard
to its lack of influence on both judgment accuracy and decision confidence, rs(44) =
−.04 and −.01, respectively. As might have been expected, however, prior training
did significantly increase the number of reasons that investigators cited for their
decisions, r(44) = .41, p <.01, and marginally prompted an increase in false alarm
responses, r(44) = .25, p <.10, and a bias to see deceit, r(44) =−.28, p <.10.
DISCUSSION
While detecting deception seems an inherently difficult task, the judgments that
police investigators make of suspects on the basis of preinterrogation interviews
represent a critical choice point in law enforcement, as they determine whether
suspects are sent home or subjected to guilt-presumptive interrogation. Given the
importance of this decision-making process, deception-detection techniques that rely
on an analysis of verbal and nonverbal behavioral cues have been used as a basis
for training. Despite the assumption that such programs enhance accuracy, however,
published psychological research has generally failed to demonstrate performance
increments as a function of special training or prior law enforcement experience
(Bull, 1989; DePaulo & Pfeifer, 1986).
Our goal was to better understand the influence of training and experience on
deception-detection judgments. Unfortunately, previous studies had focused upon
global accuracy rates rather than separating performance for both truthful and deceit-
ful targets. By separating these two aspects of performance, the application of SDT
affords the opportunity to distinguish between two conceptually and computation-
ally independent parameters of detection skill: discrimination accuracy—correctly
detecting a signal (deception) versus correctly rejecting its absence (truth), and
response bias—the threshold of evidence necessary for an individual to respond
that a signal (deception) has been presented. On the basis of the assumptions under-
lying manipulations of training or experience, one would expect that discrimination
accuracy would be enhanced; however, given the relative independence of the two
parameters, response bias should not be influenced by such manipulations.
To test this hypothesis, we first conducted a studywise analysis of the previous
literature in which estimates of discrimination accuracy and response bias were cal-
culated. Effect sizes were computed for each study, after which they were weighted

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
478 Meissner and Kassin
by sample size and aggregated across studies for each SDT parameter. The results of
this analysis indicated a reliable effect on estimates of response bias. In particular,
training and experience appeared to loosen participants’ response criterion, thereby
increasing the likelihood that they would judge targets as deceitful rather than truth-
ful relative to a no-training or experience control group. Across studies, however,
training and experience produced no reliable effects on the ability to discriminate
between truth and deceit.
To more directly assess the validity of this “investigator bias effect” in a foren-
sically relevant setting, we presented a sample of law enforcement investigators
with videotapes of brief interviews in which guilty or innocent male suspects de-
nied their involvement in a mock crime. The task was to determine whether each
suspect was truthful in his denial or lying to cover his guilty actions. We compared
the performance of our police investigators with that of na¨ıve and trained student
participants from the study of Kassin and Fong (1999), who completed the task
with the same stimuli and under identical conditions. Overall, the investigators did
not outperform Kassin and Fong’s trained or na¨ıve student participants (Kassin
& Fong, 1999), in terms of either global accuracy or discrimination performance
(A
0
). However, they were significantly more confident in their judgments and sig-
nificantly more likely to respond “deceitful” rather than “truthful”—a liberal re-
sponse bias that is consistent with our reanalysis of the literature. In short, the piv-
otal decision investigators must make regarding whether to further interrogate a
suspect may be based on prejudgments of guilt, confidently made, but frequently in
error.
One might argue that our investigators were limited in their performance by
the relative brevity and low-stakes nature of the taped interviews. As noted earlier,
however, their accuracy rate is consistent with past studies of law enforcement pro-
fessionals. Moreover, this possible limitation with regard to the targets of judgment
used in this study cannot explain the response bias or high confidence levels exhib-
ited by investigators relative to students (indeed, from a metacognitive perspective,
one would hope that professionals would adjust their confidence levels according
to such perceived limitations). One might also argue that investigators were limited
by their ability to merely observe, not actively conduct, the interviews. Importantly,
however, research in other contexts has shown that judgments of truth and decep-
tion are more accurate when made by observers than by conversational interactants
(Buller, Strzyzewski, & Hunsaker, 1991).
Given the potential and serious consequences of an investigator bias effect, it
will be important in future studies to consider both the theoretical mechanisms and
practical implications of the observed bias. From a theoretical perspective, social
cognitive mechanisms may be responsible for the investigator bias due to experi-
ence or training. For example, investigators in a deception-detection task may be
influenced by subjective probabilities, or base rates, for the incidence of truth versus
deceit (Fiedler, 2000), and these probabilities may be influenced by factors, such as
the race, age, ethnicity, criminality, or status of the suspect (Ruby & Brigham, 1996),
or by the contextual basis of the interview itself (e.g., whether there is indepen-
dent, corroborative evidence). These initial attributions may then set the stage for
other mechanisms, including the manner in which investigators encode and process

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
Investigator Bias 479
seemingly relevant information (Kassin et al., 2001; Nickerson, 1998). For example,
police investigators may encode and weight information that has little discriminative
validity (see Simon, 1991). Although such mechanisms were not tested in this study,
future research should consider their influence in the investigator bias effect and
seek to dissociate the effects of training and prior experience.
From a practical standpoint, the investigator bias uncovered in this study sug-
gests that many innocent suspects are in a precarious situation before undergoing
formal interrogation. Routinely, investigators seek to assess a suspect’s veracity in
order to determine whether to launch into an interrogation, a process that, at times,
elicits coerced false confessions from innocent people. Another possible detrimen-
tal effect on the innocent suspect may arise from the influence of trial testimony
provided by investigative officers on the perceived truthfulness of the defendant.
Although this proposition remains to be empirically tested, it is possible that such
confident and descriptive testimony—accurate or not, and biased or not—may have
a strong effect on jurors and jury decision-making. Thus, in light of the profound risks
to the accused, both in the interrogation room and subsequently in the courtroom,
we believe that further research is warranted regarding the theoretical mechanisms
responsible for the effect and the practical ramifications on the disposition of a given
case.
ACKNOWLEDGMENTS
The authors thank John Turtle for assisting with data collection.
REFERENCES
Bull, R. (1989). Can training enhance the detection of deception? In J. C. Yuille (Ed.), Credibility assess-
ment (pp. 83–99). London: Kluwer Academic Publishers.
Buller, D. B., Strzyzewski, K. D., & Hunsaker, F. G. (1991). Interpersonal deception: II. The in-
feriority of conversational participants as deception detectors. Communication Monographs, 58,
25–40.
Chahal, K., & Cassidy, T. (1995). Deception and its detection in children: A study of adult accuracy.
Psychology, Crime and Law, 1, 237–245.
DePaulo, B. M. (1994). Spotting lies: Can humans learn to do better? Current Directions in Psychological
Science, 3, 83–86.
DePaulo, B. M., & Pfeifer, R. L. (1986). On-the-job experience and skill at detecting deception. Journal
of Applied Social Psychology, 16, 249–267.
DePaulo, B. M., Stone, J. I., & Lassiter, G. D. (1985). Deceiving and detecting deceit. In B. R. Schlenker
(Ed.), The self and social life (pp. 323–370). New York: McGraw-Hill.
deTurck, M. A. (1991). Training observers to detect spontaneous deception: Effects of gender. Commu-
nication Reports, 4, 81–89.
deTurck, M. A., & Miller, G. R. (1990). Training observers to detect deception: Effects of self-monitoring
and rehearsal. Human Communication Research, 16, 603–620.
Donaldson, W. (1992). Measuring recognition memory. Journal of Experimental Psychology: General, 121,
275–277.
Ekman, P., & O’Sullivan, M. (1991). Who can catch a liar? American Psychologist, 46, 913–920.
Ekman, P., O’Sullivan, M., & Frank, M. G. (1999). A few can catch a liar. Psychological Science, 10,
263–266.
Fiedler, K. (2000). Beware of samples! A cognitive–ecological sampling approach to judgment biases.
Psychological Review, 107(4), 659–676.
Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. New York: Wiley.

P1: ZBU
Law and Human Behavior [lahu] pp614-lahu-450951 September 4, 2002 20:33 Style file version June 4th, 2002
480 Meissner and Kassin
Gudjonsson, G. H. (1992). Interrogative suggestibility: Factor analysis of the Gudjonsson Suggestibility
Scale (GSS 2). Personality and Individual Differences, 13(4), 479–481.
Inbau, F. E., Reid, J. E., Buckley, J. P., & Jayne, B. C. (2001). Criminal interrogation and confessions
(4th ed.). Gaithersburg, MD: Aspen Publishers.
Kassin, S. M. (1997). The psychology of confession evidence. American Psychologist, 52(3), 221–233.
Kassin, S. M., & Fong, C. T. (1999). “I’m innocent!”: Effects of training on judgments of truth and deception
in the interrogation room. Law and Human Behavior, 23, 499–516.
Kassin, S. M., Goldstein, C. J., & Savitsky, K. (2001). Behavioral confirmation in the interrogation room:
On the dangers of presuming guilt. Unpublished manuscript.
Koehnken, G. (1987). Training police officers to detect deceptive eyewitness statements: Does it work?
Social Behavior, 2, 1–17.
Kraut, R. E., & Poe, D. (1980). Behavioral roots of person perception: The deception judgments of customs
inspectors and laymen. Journal of Personality and Social Psychology, 39, 784–798.
Leo, R. A., & Ofshe, R. J. (1998). The consequences of false confessions: Deprivations of liberty and
miscarriages of justice in the age of psychological interrogation. Journal of Criminal Law and Crim-
inology, 88(2), 429–496.
Macmillan, N. A., & Creelman, C. D. (1990). Response bias: Characteristics of detection theory, threshold
theory, and “nonparametric” indexes. Psychological Bulletin, 107, 401–413.
MacMillan, N. A., & Creelman, C. D. (1991). Detection theory: A user’s guide. New York: Cambridge
University Press.
MacMillan, N. A., & Kaplan, H. L. (1985). Detection theory analysis of group data: Estimating sensitivity
from average hit and false alarm rates. Psychological Bulletin, 98, 185–199.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General
Psychology, 2, 175–220.
Ofshe, R. J., & Leo, R. A. (1997). The social psychology of police interrogation: The theory and classifi-
cation of true and false confessions. Studies in Law, Politics, and Society, 16, 189–251.
Porter, S., Woodworth, M., & Birt, A. R. (2000). Truth, lies, and videotape: An investigation of the ability
of federal parole officers to detect deception. Law and Human Behavior, 24, 643–658.
Radelet, M. L., Bedau, H. A., & Putnam, C. E. (1992). In spite of innocence: Erroneous convictions in
capital cases. Boston: Northeastern University Press.
Rosenthal, R. (1994). Parametric measures of effect size. In H. Cooper & L. V. Hedges (Eds.), The
handbook of research synthesis (pp. 111–124). New York: Russell Sage Foundation.
Ruby, C. L., & Brigham, J. C. (1996). A criminal schema: The role of chronicity, race, and socioeconomic
status in law enforcement officials’ perceptions of others. Journal of Applied Social Psychology, 26,
95–112.
Scheck, B. C., Neufeld, P., & Dwyer, J. (2000). Actual innocence: Five days to execution and other dispatches
from the wrongly convicted. New York: Doubleday.
Simon, D. (1991). Homicide: A year on the killing streets. New York: Ivy Books.
Snodgrass, J. G., & Corwin, J. (1988). Pragmatics of measuring recognition memory: Applications to
dementia and amnesia. Journal of Experimental Psychology: General, 117, 34–50.
Thompson, S. C. (1978). Detection of social cues: A signal detection theory analysis. Personality and Social
Psychology Bulletin, 4, 452–455.
Vessel, D. (1998, October). Conducting successful interrogations. FBI Law Enforcement Bulletin, 1–6.
Vrij, A. (1993). Credibility judgments of detectives: The impact of nonverbal behavior, social skills, and
physical characteristics on impression formation. Journal of Social Psychology, 133, 601–610.
Vrij, A. (1994). The impact of information and setting on detection of deception by police detectives.
Journal of Nonverbal Behavior, 18, 117–132.
Vrij, A., & Graham, S. (1997). Individual differences between liars and the ability to detect lies. Expert
Evidence, 5, 144–148.
Vrij, A., & Mann, S. (2001). Who killed my relative? Police officers’ ability to detect real-life high-stake
lies. Psychology, Crime and Law, 7, 119–132.
Vrij, A., & Mann, S. (2001). Telling and detecting lies in a high-stake situation: The case of a convicted
murderer. Applied Cognitive Psychology, 15, 187–203.
Zuckerman, M., Koestner, R., & Alton, A. O. (1984). Learning to detect deception. Journal of Personality
and Social Psychology, 46, 519–528.
Zuckerman, M., Koestner, R., & Colella, M. J. (1985). Learning to detect deception from three commu-
nication channels. Journal of Nonverbal Behavior, 9, 188–194.