WWW\i\i’\i\\i\'\"\i\1§\e\i\\| \/

LEERARI’

Michigan state
University

 

This is to certify that the

dissertation entitled

Accuracy In Deception Detection:
A Quantitative Review

presented by

Pamela J. Kalbfleisch

has been accepted towards fulfillment
of the requirements for

Ph.D- degree in immunisation

 

Wflflmw

Major professor

Date August 1, 1985

 

MSUl‘nu WAWWWIM 0-12771

 

MSU

RETURNING MATERIALS:
Piace in book drop to

 

 

 

 

 

Augegg}

 

LIBRARJES remove this checkout from
w your record. FINES Win
,be charged if Book is
returned after the date
stamped below.
.,QY?T7713
177 A135
f2" .
"”1332 :,- 85' ”f2 2%811
D “C H 510 fi
Rh?
Us! 57‘? n L 30.02
4W" .5» (,5; 2.... MAngq L 200g 2
.gtdfiba
DECO a 1993
W

ACCURACY IN DECEPTION DETECTION:

A QUANTITATIVE REVIEW

By

Pamela Joy Kalbfleisch

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Department of Communication

1985

ABSTRACT

ACCURACY IN DECEPTION DETECTION:
A QUANTITATIVE REVIEW

33'

Pamela Joy Kalbfleisch

This dissertation reports a meta-analysis of the research on
the human ability to detect deception. A review of literature
and critique of previous deception detection meta-analyses are
considered prior to the deve10pment of a rationale for the meta-
analytic technique designed for this analysis. The predominant
use of repeated measures and mixed measures research designs in
the deception research negated the use of traditional meta-
analytic techniques, because these techniques do not feature a
method to extract effect sizes or to cumulate the results of
these designs. The meta-analytic technique developed for this
meta-analysis allows the researcher to cumulate the results from
dichotomous measurements in repeated measures, mixed measures and
between-subjects designs, but does not estimate sampling error,
measurement error nor restriction in range.

In general, this meta-analysis found humans to be poor lie
detectors. Differences in this ability across experimental
conditions are small. However, several patterns of human success

in deception detection are apparent. Specifically, the presence

Pamela Joy Kalbfleisch

of message content allows observers to obtain higher accuracy
scores than in observation conditions without content. Observers
viewing shots of communicators' bodies are more accurate than
observers viewing combination shots of heads and bodies, or shots
of heads only. Evidence also suggests observers are more likely
to judge a communicator to be telling the truth than to be lying.
Other suggested patterns were 1) observers familiar with the
truthful behavior of the person they are judging are more
accurate than those not exposed to a truthful baseline, 2)
females appear to be slightly more accurate than males in
detection deception, and 3) female deceivers appear to be

slightly easier to detect than male deceivers.

Dedicated to my parents

Paul and Marian Kalbfleisch

ACKNOWLEDGMENTS

It has been a privilege to observe and work with my advisor
Dr. Gerald R. Miller during my years at Michigan State. With his
guidance, I have learned much about scholarship and
professionalism. A hearty thanks G.R.!

Dr. John E. Hunter contributed to my training by sharing his
creations of research methodologies for the future. It has been
exciting to be aware of his mathematical innovations years before
they will be studied in most college classrooms.

Dr. William A. Donahue always provided helpful advice. I
have often been grateful for his savvy and perceptive
observations. Whether such understanding comes naturally or is
the result of spending years studying talk, is not clear to me.

Dr. Raymond Frankmann is responsible for much of my
statistical knowledge. To my surprise during the defense of this
dissertation, he emerged as a mechanical lie detection expert,
having spent numerous hours administering polygraph examinations.

Dr. Charles Atkin helped me complete my graduate work with
his input on this project. I appreciate having the Opportunity to

include this communication scholar on my guidance committee.

iii

Dr. Edward Fink, although he was not part of my guidance
committee, contributed significantly to my graduate training.
Since leaving Michigan State, he has continued to provide
unwavering support of my research and professional development.

James Stiff and Paul Mongeau were trusted research
colleagues. I will always appreciate their friendship.

Finally, those closest to me deserve the biggest thanks of
all. My parents Paul and Marian Kalbfleisch have provided years
of support for me regardless of the venture. From the rough and
tumble rodeo days to my scholarly pursuits. They have never lost
faith in me, nor let me lose faith in myself. To them I owe
much.

My dear friend Jan David Gierman provided hours, months and
years of understanding and help. He has been a positive force in
my life. Having never known me other than as a graduate
student, I know he is eagerly anticipating "life without graduate
school".

My Sister Karen Lind and her husband Eugene, were also
supportive trough this endeavor. I am pleased to have been

blessed with such a sibling and brother-in-law.

iv

TABLE OF CONTENTS
Page

LIST OF TABLES O O O O 0 O O O O O O O O O O O O O O O O O Vii

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . 1
Overview . . . . . . . . . . . . . . . . . . . . . . l
Lie Detection in the Past: The Ordeals . . . . . . . 2
Modern Physiological Detection of Deception . . . . . 3

Social Science Examination of Unaided Human
Human Ability to Detect Deception . . . . . . . . . . 7

The Significance Test Tallies . . . . . . . . . . . . 11

The Quantitative Reviews . . . . . . . . . . . . . . l3
Zuckerman, DePaulo and Rosenthal . . . . . . . . . . l4
Kra ut O O O O O O O O O O O O 0 O O O O O O O O O O 18

The Proposed Meta-Analysis . . . . . . . . . . . . . . 24
MOD 0 O O O O C C O 0 O O O O O O O O O O O O O 25

Extraction of Effect Sizes for
Repeated Measures: The Special Case . . . . . . . . 27

Sample Selection . . . . . . . . . . . . . . . . . . 31
Procedures for Cumulation . . . . . . . . . . . . . . 39
RESULTS . . . . . . . . . . . . . . . . . . . . . . 41
Observational Conditions . . . . . . . . . . . . . . 41
Familiarity . . . . . . . . . . . . . . . . . . . . . 49
Sex Differences in Deception Detection . . . . . . . . 51
Judgments of Truth vs Lie . . . . . . . . . . . . . . 56

Narrative Review . . . . . . . . . . . . . . . . . . 57

vi

DISCUSSION

Comparison With Previous Meta-Analyses

Sampling Error

Conclusion
REFERENCES

Footnotes

vi

67

67

71

72

74

91

10.

11.

12.

LIST OF TABLES

Studies Selected for Meta-Analysis . . . .

Deception Detection Accuracy: Conditions
Under Which Truthful and Deceptive
Messages are Observed . . . . . . . . . .

Deception Detection Accuracy: Conditions-
Under Which Truthful and Deceptive
Messages are Observed . . . . . . . . . .

Accuracy in Deception Detection: Conditions
Under Which Truthful and Deceitful
Messages are Observed . . . . . . . . . . .

Baseline Familiarity With Communicator
and Judges Accuracy in Deception Detection

Judges Sex and Accuracy in Deception
Detection O O I C O O O I O C O O O O O C

Deception Detection Accuracy
for Male Judges . . . . . . . . . . . . .

Deception Detection Accuracy
for Female Judges . . . . . . . . . . . . .

Male Judges: Deception Detection

Accuracy Observing Male vs Female
comMicators O O O O O O O O O O O O I 0
Female Judges: Deception Detection
Accuracy Observing Male vs Female

Communicators . . . . . . . . . . . . . .

Sex of Communicators and Accuracy
in Deception Detection . . . . . . . . . .

Observer Judgments of Truth vs Lie . . .

vii

Page

33

43

45

48

50

52

53

53

55

56

57

INTRODUCTION

Overview

There has long been concern over the accuracy of decisions
indicting suspected criminals and untrustworthy social contacts.
History is marked with attempts to ward off uncertainty
associated with assessments of veracity. The last eighty years
have seen American law enforcement and businesses turn to devices
such as the polygraph in order to avoid the uncertainty involved
in lie detection.

Social scientists have tried to assess human abilities
without the aid of such devices. Results of these investigations
have been both contradictory and complex. Through the use of
quantitative reviews, Zuckerman, DePaulo and Rosenthal (1981) and
Kraut (1980) have attempted to make this body of research easier
to comprehend and to provide overall assessments of human lie
detection ability. However, these attempts have shortcomings.

This dissertation first briefly reviews the lie detection
literature beginning with the techniques which have attempted to
substitute a tool or device for human judgment, followed by a
look at the social science research concerning human lie
detectors. Then each quantitative review of the deception

detection research is examined in depth followed by a

Deception Detection - 2

reexamination of accuracy in deception detection using an
alternate analytical technique.
Lie Detection In The Past: The Ordeals

Uncertainty over possible lies has long been a troubling
problem for members of the human race. As early as 900 B.C.
writings on papyrus outlined steps to detect liars (Trovillo,
1939a, 1939b).

Attempts to avoid relying on human judgment began with the
use of physical ordeals to ferret out liars (Trovillo, 1939a,
1939b). The ordeals assigned the decision making to forces the
society perceived as being greater than themselves (i.e. the
gods). Trovillo (1939a, 1939b) reviewed techniques such as the
red-hot iron ordeal where accused liars were forced to lick red-
hot irons, being exonerated if they emerged unburned, or the
balance ordeal, where the suspected were weighed before and after
a mystical exhortation with proof of veracity shown by a change
in the weight of the person under suspicion.

Trials by ordeal were still used in the third world
countries at the time Trovillo published his review. For
example, in some villages in India, those suspected of deception
were made to chew rice, guilt being determined if rice could not
be swallowed. In Africa deceit was determined by plunging one's
arm in boiling water and then looking for tell-tale "deception

blisters" the next day.

Deception Detection - 3

Some of these procedures may seem primitive. However, the
African medicine man sniffing potential deceivers in order to
ferret out deception may not be much different from the
physiological measures of palmer sweat used by modern mechanical
lie detectors. The dry mouth unable to swallow rice and the dry
tongue surface seared by the hot iron may also be related to the
physiological phenomena of "dry mouth" or decreased salivia flow
during times of high stress. Kleinmuntz and Szucko (1984a) in
their examination of primitive lie detection techniques noted
that if the mouth of the accused remained moist it could have
been possible to avoid being burned by the iron and possible to
swallow the dry rice.

The accuracy of these primitive lie detection techniques
remains a mystery. The decreased flow of saliva or the increased
sweating accompanying fear of discovery may well have
accompanied fear of the ordeal itself. Also the physiological
processes underlying ordeals such as plunging one's hand into
boiling water and weight change on a balancing scale still remain
unexplained and their relationship to actual lie detection is

vague.

Mbdern Physiological Detection of Deception
Modern attempts to detect deception have also focused on
physiological signs: breathing change (Benusi, 1914), variation

in systolic blood pressure (Marston, 1917; Larson, 1932) and

Deception Detection - 4

galvanic skin responses (Summers, 1937, 1939). Modern polygraphs
now measure changes in all three of these physical processes
(Yohman, 1978).

The mechanical lie detector has had a substantial impact on
the American public. Lykken (1981) estimated that over one
million Americans were subjected to polygraph tests in 1980
either in the legal system or employment screening. Kleinmuntz
and Szucko (1984a) noted that the American Polygraph Association
projected 2.3 million Americans would be taking polygraph tests
in 1984. However, despite increasing use of the technique, the
"ordeal" of the polygraph remains a controversial issue. The
controversy centers on mixed empirical evidence regarding the
accuracy of the polygraph and problems in validity of measures of
the machine's accuracy.

Mixed empirical evidence. In 1977 Podlesny and Raskin pub-

 

lished results indicating that the polygraph was accurate from 88
to 96 percent of the time. In 1978 Raskin and Bare reported
findings of 96 percent accuracy and 99 percent accuracy in a
subsequent study (Raskin, Barland & Podlesny as cited in Raskin,
1978). Raskin and Podlesny (1979) broke down the results of
their lie detector tests yielding findings of 90 percent accuracy
levels when subjects were lying and 89 percent accuracy when they
were telling the truth.

However, Lykken (1979) failed to find the high accuracy

ratings reported by these researchers. Instead he found the

Deception Detection - 5

polygraph to be 64 to 71 percent accurate with a strong bias
against truthful communicators: 36 to 39 percent of the truthful
comunicators were incorrectly classified as deceptive.
Kleinmuntz and Szucko (1984b) and Szucko & Kleinmuntz (1981)
found the false positive rate (number of truthtellers classified
as deceptive) to be as high as 55 percent.

Measures of polygraph accuracy. Lykken (1978,1979) and

 

Kleinmuntz and Szucko (1984a) have questioned the use of the
polygraph in lie detection claiming the accuracy of the machine
has not sufficiently been established and the tests of the
technique are fraught with problems. These researchers note that
although a technique for interpretation of the results does exist
in which fluctuations in the polygraphs records are submitted to
a set of measurements and scoring (Backster, 1963), many
interpreters eschew specific measurements and simply make global
assessments of communicators' veracity based on their overall
performance. After analyzing polygraph records, Kleinmuntz and
Szucko (1981) concluded that the interpreters' judgments often
bore little resemblance to the physiological data.

While laboratory studies may carefully control the inter-
preter's knowledge of the experimental conditions, the
interpreter's impressions of the communicator remain unchecked.
Field studies also share this problem, Ben-Shakhar, Lieblich and

Bar-Hillel (1982) contend that very few interpreters use any tec-

Deception Detection - 6

hnique other than global evaluations in assessing veracity. Also,
there may be contamination of lie detector results because
additional information about the suspect may be at the
interpreter's disposal, e.g. intelligence gathered about the sus-
pect, impressions of former interrogators, and records or
previous convictions (Ben-Shakhar et al., 1982).

The polygraph accuracy studies done in field settings also
lack a verifiable criterion of correctness. Accuracy in criminal
or civil legal cases is assessed either by comparing the polyg-
raph test results with the decisions of the judge, jury or panel
of legal experts, or by a confession of the accused (Barland &
Raskin, 1976; Bersh, 1969; Horvath, 1977; Horvath & Reid, 1971;
Hunter & Ash, 1973; Slowik & Buckley, 1975; Wicklander & Hunter,
1975). Thus, whether the polygraph is actually accurate in these
settings is verified only through human judgment.

Modern lie detection appears to depend upon the human
ability to detect deception. Members of the Reid and Arther
schools of lie detection (Reid & Inbau, 1977) apparently are
proud of this fact. Lykken (1979) noted numerous endorsements in
the polygraph trade journals for global assessments of suspected
deceivers, stressing that it was the technician who administers
and interprets the test and the technician who, in fact,
functions as the lie detector and not the polygraph.

In a response to Lykken's (1978) critique of Raskin and

Hare's (1978) high polygraph accuracy ratings, Raskin (1978)

Deception Detection — 7

disputed Lykken's criticism that one of the reasons for the high
ratings could be that the interpreters were very well informed
and free to make clinical interpretations. Raskin (1978)
defended his findings by asserting the human factor in lie
detection decisions is negligible. His defense of the
polygraph's high accuracy scores was based on the assumption that
humans are very poor lie detectors, an assumption that appears to
underlie use of alternative lie detection methods. However, the
human impact of the truth/lie classifications of the polygraph is
hard to assess. In field studies high polygraph scores represent
concurrent agreement between human and machine, and not a measure
of accuracy. Additionally, in both lab and field studies the
polygraph accuracy is the result of the combination of human and

machine classification.

Social Science Examination of Unaided Human Ability to Detect
Deception

Social scientists have studied the lie detection
abilities of human judgment without the polygraph. Studies as
early as 1941 (Fay & Middleton) have probed the question of how
accurate humans might be in in their attempts to tell lies from
truths. Review of these findings is important to determine the
degree of accuracy that could be expected in legal setting by
judges, jurors or polygraph interpreters. It is also important

to see how well humans may cope with the everyday social milieu.

Deception Detection - 8

The soundness of decisions made regarding whether to believe
a tardy lover's excuse, a repairman's cost estimate, or an
acquaintance's statement that a wallet left at a dinner table was
never noticed or found relies on an individual's own lie
detection ability. Friendships may end because a friend's
truthful statements were doubted; individuals may be exploited by
false claims, and frustration may build as one wonders just whom
to believe when conflicts arise. While these decisions may not
be life and death matters, everyday veracity assessments may
affect the course of human relationships and the ability to
manage one's environment.

Contradictory finding . Studies of human lie detection have

 

explored the effects of a variety of observational conditions on
human accuracy, investigated an assortment of observers and
their differential abilities and examined the deceptive perfor-
mance of differing types of communicators to see if the
falsehoods of some are more difficult for observers to detect
than others. The results of these studies appear complex and
often contradictory.1 For example, while human accuracy at
detecting lies and truths has sometimes been found to be signifi-
cantly better than chance (Hemsley & Doob, 1979; Lavrakas &
Maier, 1979; Potamkin, 1982); others have found these accuracy
levels do not exceed chance expectations (Bauchner, Brandt &

Miller, 1977; Matarazzo, Wiens, Jackson & Manaugh, 1970; Motley,

Deception Detection - 9

1974). The mediating effects of observational conditions also
remain unclear. Ekman and Friesen (1974) theorized that
deceptive communication could be detected better by watching the
bodies of potential deceivers rather than their heads and faces.
Ekman and Friesen posited that people are more aware of their
faces and hence exert more control over them both when deceiving
and truthtelling. This awareness allows them to use their faces
to more accurately convey messages and to more carefully mask
deception clues. Conversely, Ekman and Friesen reasoned that
people are less aware of their bodies and thus exert less control
over body cues that might mark their deceptive attempts. These
researchers back up their "leakage hypothesis" with empirical
results indicating observers who viewed only the bodies of
communicators made more accurate truth/lie decisions than those
viewing the heads.

Littlepage and Pineault's (1981) findings are consistent
with those .of Ekman and Friesen (1974). When communicators
express emotional information, such as descriptions of moods and
feelings, Hocking, Bauchner, Kaminski and Miller's (1979)
research is also consistent with the Ekman and Friesen theory.
However, when communicators lied and told the truth regarding
factual information, Hocking et a1. (1979) found observers
viewing communicators' heads were more accurate than those

viewing shots of their bodies. Finally, Wilson (1975) found

Deception Detection - 10

neither the head nor the body, but rather the combination of
both, was the best for accurate judgments of veracity.

Meier and Thurber (1968) and Sakai (1981) removed the
viewing condition altogether. Their results indicate that those
listening to communicators were better able to accurately
discriminate lies from truths than those viewing them. These
results were not found in other studies. For example, Wilson
(1975) noted that those who listened to deceptive and truthful
communication were not as accurate as those who viewed the commu-
nication. Bauchner et a1. (1977), Hocking et a1. (1979) and
Harrison et a1. (1978) found no significant differences in
judgmental decisions between the audio and video conditions.

Deception studies have not established whether the sexes
differ in gullibility. Sakai (1981) and Meier and Thurber (1968)
contend that females are better lie detectors than males. However
Atmiyanandana (1976), Parker (1978) and Rovira (1982) conclude
there is no difference between the sexes. Parker (1978) noted
that while a sex difference in observers was not found, observers
did have more difficulty detecting the deceptive communication of
female than of male deceivers. Sakai (1981) also noted a
significant difference in success at detecting male versus female
deceivers; however, Rovira (1982) and Wilson (1975) found no
significant differences in detection of male versus female

deceivers.

Deception Detection - 11

These studies illustrate the contradictory results in the
deception literature. The findings do not yield clear statements
regarding how accurately hunans can detect lies from truths or
how observational conditions and individual difference variables

may mediate this accuracy.

The Significance Test Tallies

 

The divergent findings of the deception research might
prompt a reviewer to start a tally of significant and
nonsignificant results, with the goal of allowing a preponderance
of evidence, either significant or nonsignificant, to arbitrate
these mixed findings. In the literature reviews that Open most
deception studies this tallying has implicitly taken place. In
rationales for research questions and hypotheses, researchers
often depend on significance tests reported by other researchers
to build a line of research reasoning. This line of reasoning may
indicate what questions have yet to be answered, propose the
presence of possible moderating variables or simply report past
research that supports the current investigation.

After reviewing the diverse findings in deception research,
a number of social scientists have called for more research to
assess human lie detection abilities (Feldman & White, 1980;
Harrison et al., 1978; Hocking et al., 1979; Hemsley & Doob,
1979; Maier & Lavrakas, 1976; Parker, 1978; Rovira, 1982; Rotkin,

1980; Sereno, 1981). In light of the numerous completed

Deception Detection - 12

deception studies, this plea overlooks the possibility that
additional exploration will only add to the contradictions. Other
researchers (Comadena,1982; Potamkin,l982) have attempted to
explain the contradictory findings by pointing out methodological
limitations in those studies that contradict their line of
reasoning.

Problems in using significance tests for cumulation.
Unfortunately, the use of significance test tallies is not the
best approach for better understanding the findings dealing with
human lie detection. Attempting to cumulate results across
studies by counting significant findings versus nonsignificant
findings can yield misleading results.

One problem is the significance test does not provide an
estimation of the strength or importance of a relationship. It
only suggests whether the obtained F differs significantly from
chance (Glass, McGaw & Smith, 1981). When cumulating
significance tests, significant results are classified as
presence of a relationship and nonsignificant results are
classified as absence of a relationship (Hunter, Schmidt &
Jackson, 1982). Confidence intervals for obtained values are not
considered in these tallies. Hence, significant and non-
significant results are classified as absolute (Glass et al.,
1981). Further, since results from large sample sizes are more

likely to be found significant than those from small samples,

Deception Detection - 13

some strong effects from small samples may be overlooked (Class
at al., 1981).

The Quantitative Reviews

 

Beyond the significance test tallies, two recent attempts
have been made to quantitatively review the deception research
and provide an answer to some of the unanswered questions
concerning lie detection. Zuckerman, DePaulo and Rosenthal
(1981) and Kraut (1980) used meta-analysis to evaluate research
findings. The large number of subjects in these aggregated
analyses should allow researchers to better estimate the presence
of actual effects (Hunter, 1982). Glass et al. (1981), Hunter
et al. (1982), and Levine, Romashko and Fleishman (1973) have all
developed techniques that use summary statistics other than the
significance test to cumulate study findings and thus avoid the
test's dependence on sample size. Hunter et al. (1982) have also
developed methods to assess the degree of sampling error present
in the cumulated data along with other statistical artifacts such
as measurement error and restriction in range (see Hunter et a1.
1982 for a review of these meta-analytic techniques and others).

The quantitative reviews of the deception detection
literature by Zuckerman et al. (1981), and by Kraut (1980) use
cumulative methods that differ from those developed by Class
et a1. (1981), Hunter et a1. (1982) and Levine et al. (1973).

These deception meta-analyses will be reviewed in terms of their

Deception Detection - 14

usefulness, application, and contribution toward the under-
standing of human lie detection.
Zuckerman, DePaulo and Rosenthal

The most extensive of these quantitative reviews was
completed by Zuckerman et al. (1981). This project reviewed both
the literature concerning accuracy in deception detection and
that which correlated verbal and nonverbal behaviors with
deceptive communication. They sought to use Cohen's d statistic

as a measure of effect size. They computed this statistic for

.42."?
m—

obtained value for F and df equals the degrees of freedom for the

 

this review as where F equals the
variable of interest. However, this computation is inappropriate
for studies with repeated measures, i.e. the majority of
deception studies.

Conceptual issues. In Zuckerman et al. (1979), studies are

 

presented with their corresponding d estimate of effect size.
The problem is that d as defined is a statistic based on a
comparison (Cohen, 1977). However, only one set of comparisons
made in the meta-analysis is defined. In this set the d is the
effect size of the difference between truthful and deceptive
judgments. In the other sets of comparisons it is unclear-
exactly what is being compared. In interpreting the d values
provided, knowledge of what comparisons the effect estimates

represent is critical.

Deception Detection - 15

A second problem in interpreting the tables in Zuckerman
et al. (1979) is that the results are provided in standard
deviation units and no other summary information is made
available for interpreting these findings. The standard
deviations alone do not provide sufficient information from which
to determine how accurate humans are at lie detection. For
example, how accurate are people at detecting deception at 1.2
standard deviations or at 1.4? The standard deviations suggest
that the observers in one study are more accurate than those in
another, but the magnitude of this accuracy is not specified.

Operational issues. While these are both problems that

 

affect interpretation of the results of the Zuckerman et a1.
(1979) study, the inappropriate computation of Cohen's d
statistic in the estimation of effect sizes is far more serious.
Cohen's d statistic as computed by Zuckerman et al. and applied
to the deception research may have yielded estimates which are
grossly over-inflated, leaving the overall validity of the
Zuckerman et al meta-analysis in question.

The d statistic as defined in this meta-analysis is actually
a transformation of F to t with the sample size removed from the
t to obtain d. The df in the denominator is used for the
estimate of sample size. The trouble with this effect size
statistic is that it is not appropriate for the most common

designs used in deception research. The d statistic described by

Deception Detection - 16

Zuckerman et a1. assumes an independent groups analysis of
variance design with two levels.

For example, the use of degrees of freedom as an estimate of
sample size is inappropriate in the case where multiple factors
are used in a study. In the independent group designs where the
degrees of freedom equal N-l, use of the degrees of freedom to
estimate sample size of this value is equal to the sample size
minus one. However, in the multiple factor independent group
designs the degrees of freedom will be N-K, where K equals the
number of factors in the design. This error in Zuckerman
et al.'s application of the d statistic to multiple factor
designs will yield d's that are artificially large due to the
reduced size of the denominator with N-K replacing the N-l for
multi-factor designs.

While this sample size estimate is a small problem, a more
serious error is inherent in applying the d statistic to research
using the research design with repeated measures. These designs
are implemented in deception detection by having each observer
judge the veracity of a number of communicators who lie or tell
the truth. Typically each judge will view a number of
communicators who represent different manipulations of the
independent variables in the design. The judge may also be asked
to make the decisions in different viewing conditions. The
differences in judges are typically treated as between-subjects

factors and the differences in the communicators and the

Deception Detection - 17

conditions under which they are observed then are treated as
within-subjects factors (repeated measures).

The major problem with the Zuckerman et al. use of the d
statistic is that it fails to properly estimate the effect size
of these repeated measures. There are several reasons for this
failure. The first problem with estimating the effects of
within-subjects factors in this way is that the variance of
within-subjects factors is the variance of difference scores and
not an estimate of the within-cell variance. The d as used by
Zuckerman et. al. assumes an estimate of the within-cell
variance. Therefore it does not properly assess the variance of
difference scores.

The degrees of freedom estimate for the within subjects F
values are also not comparable to the degrees of freedom value
assumed by Zuckerman et al. to be N-l. Within-subjects F values
will differ depending on the error term associated with the
degrees of freedom. This difference in the size of the degrees
of freedom is another reason the Zuckerman d values estimated
from within-subjects designs will not be comparable to those from
between-subjects designs.

These problems with the d statistic in the within-subjects
case created by an inappropriate estimate of variance will
manifest themselves in over inflated d values. Also, Zuckerman

et. al.'s meta-analytic estimation of effect size based on the

Deception Detection - 18

value of F for within-subjects factors may not even be possible
without access to the original data. Current statistical
reporting process in many social science journals that publish
deception research does not include the publication of the
variance estimates necessary for determining the effect sizes for
these within-subjects designs.

A final criticism of the Zuckerman et al. meta-analysis is
that for singular studies, estimates of effect size were not
provided. Instead, Zuckerman et al. dealt with these studies by
narratively describing their results in terms of statistical
significance. Since the authors of these studies provided the
necessary summary statistics for the computation of Zuckerman's d
statistic, the results of the individual studies should have been
provided in a common metric with the studies cumulated in the
meta-analysis.

The Zuckerman et a1. meta-analysis is the most
comprehensive quantitative review of the deception literature to
date. However, it has severe problems in the operationalization
and application of the Cohen's d effect size statistic and hence
the results should be considered with caution.
£222.:

The Kraut (1980) quantitative literature summary of accuracy
in deception detection is less extensive. He presents a fairly
simple analysis of this research in the form of an average

compiled from the results of ten studies. His estimate of average

Deception Detection - 19

human lie detection ability is 57 percent where 50 percent
accuracy would be expected by chance.

The summary statistic used by Kraut, the mean, is easier
understood than the Zuckerman et a1. effect size measures. For
example, if observers are reported as being able to tell lies
from truths 50 percent of the time, it is easy to picture
observers being incorrect in half of their judgments. A 75
percent accuracy level can easily be interpreted as being
incorrect in one out of four judgments or being correct three
fourths of the time.

However, Kraut's results may be misleading. In Kraut's
review he summarizes a sizable body of complex and somewhat
contradictory studies with a single average (572) and its
standard deviation (7.82). Overlooked is the fact that these
studies have measured accuracy of judges placed in several
different types of observational conditions in experiments
testing the effects of a number of different independent vari-
ables in varying experimental paradigms.

For example, Kraut and Poe (1980) looked at the accuracy of
customs inspectors compared to laypersons, while Hemsley (1977)
tested male versus female ability to detect deceptive
communication of members of the same and opposite sex. Maier and
Thurber (1968) had their communicators role play the parts of

deceivers or truthtellers while observers watched, heard or read

Deception Detection - 20

the transcript of their performance, Ekman and Friesen (1974),
on the other hand, had their communicators respond deceitfully or
truthfully to an interviewer's questions. Observers in the Ekman
and Friesen study viewed silent video tapes of these responses,
but were allowed to view only the heads or bodies of the stimulus
persons. Kraut's single overall average does not indicate what
differences if any there may be under these different conditions
and others.

By consolidating the accuracy ratings for each study into
one score for averaging, Kraut may have misrepresented the
findings of this body of research. For example, Maier and
Thurber (1968) found that those who listened to and those who
read transcripts of people lying and telling the truth were more
accurate (77.02 and 77.32) than those who watched people lying
and telling the truth (58.32). However Kraut averaged these
conditions to yield 70.9 percent as the representative summary
statistic for the Maier and Thurber study in his overall
cumulation of the deception studies. By doing this Kraut not
only overlooked differences in experimental conditions when
constructing his overall average but his results may also be
misleading, with an overall average of a study's accuracy levels
across conditions not providing a very good picture of the
outcomes of the Maier and Thurber study or others.

Further problems in the utility of Kraut's overall estimate

lie in the sample of studies he selected for inclusion in the

Deception Detection - 21

average. These problems lie in the: 1) derivation of mean
accuracy ratings, 2) chance probability associated with included
estimates, and 3) the selection procedures used in obtaining the
mean accuracy ratings.

If one reviews the published reports of the studies included
in Kraut's overall estimate of accuracy several of these problems
become apparent. For example, Kraut utilized an accuracy
estimate of 46 percent from the Kraut and Poe (1980) experiment.
However, their published report does not contain any reference to
a mean accuracy estimate. Possibly Kraut's use of an unreported
mean estimate can be understood given he was senior author of the
study and had ready access to the unpublished raw data.

This was not the case in the Maier and Janzen (1967)
article. In this report, counts were provided, recording the
number of observers who correctly identified liars and
truthtellers. To obtain a mean accuracy rating in a 0-1 metric,
Kraut was required to convert this count to a percentage by
dividing the number of correct judgments by the total number of
judgments made. Dividing the Maier and Janzen count of 97
correct judgments by the 228 total judgments yields a mean
accuracy estimate of 43 percent. However, Kraut listed an
estimate of 61 percent from the Maier and Janzen study in

constructing his overall mean accuracy.

Deception Detection - 22

The origin of the mean accuracy estimates that Kraut used to
represent the Kraut and Poe (1980) and Maier and Janzen (1967)
studies is unclear. A brief explanation by Kraut regarding their
derivation would have added credibility to their usage as
estimates of accuracy in deception detection.

A second problem in the sample of mean accuracy ratings
employed by Kraut lies in the chance level of one of the studies.
While Kraut interprets his overall estimate as having a chance
level of 50 percent in all studies, the chance level in the
Geizer, Rarick & Soldow (1977) study appears to be 33.3
percent. Instead of asking observers to decide whether people
where lying or telling the truth (where chance would be 502),
they asked their observers to decide which of three peeple was
telling the truth in a "To Tell The Truth" game show excerpt. In
this case the possibility that the judges would make the correct
choice by chance was 33.3 percent. In Kraut's article he notes
this difference in probability, but he still includes the Geizer
et a1. estimate in his overall average of detection ability. In
doing this he has created an overall estimate based on estimates
not converted to the same metric, hence the Hemsley and Doob
estimate is not comparable to the others included in the same
overall estimate.

The specific sample of studies may also contribute to the
problems with Kraut's review. The total number of lie detection

judges is 855 people. Across 32 studies2 reporting mean accuracy

Deception Detection - 23

ratings in a metric ranging from 0 to 1 (with .5 chance
accuracy level), there are 3439 possible lie detection judges in
the deception literature or 3,577 if one wishes to include the
Kraut and Poe (1980) and Geizer et al. (1977) studies. Why Kraut
choose to use less than half the available studies is not clear,
nor are the selection procedures that he used to compile his
sample. For example, Kraut chose to use three of the four
deception detection studies reported by Maier and his associates
(Lavrakas & Maier, 1979; Maier & Janzen, 1967; Maier S Thurber,
1968), but did not indicate why the Maier and Lavrakas (1976)
study was not included. Other lines of research were totally
excluded from the quantitative review. This incomplete selection
of studies brings the representativeness of the Kraut estimate of
human lie detection ability into question.

In sunmmry, problems with the quantitative review by Kraut
include use of a small unrepresentative sample, failure to
differentiate the variables considered in studies included, and
utilization of misleading accuracy estimates that resulted from
study-wide means pulled into the middle range by averaging
overall experimental conditions. Use of estimates that were
possibly unsuitable for this review also added to the reduced

utility of this quantitative summary.

Deception Detection - 24

The Proposed Meta-Analysis

Critiques of the Zuckerman et al. (1979) and Kraut (1980)
meta-analyses indicate that the results yielded by these analyses
are misleading. An accurate assessment of human lie detection
ability and those observational conditions and individual
difference variables that may mediate it are still not available.

Given our dependence on deception detection ability in
legal, business and social settings, an assessment of this skill
is essential. A reexamination of the deception detection
literature in a different context should provide a more
enlightening understanding of our ability to detect deceit.
Because this body of research relies heavily on mixed and
repeated measures designs, this dissertation will present a meta-
analysis that uses a technique allowing for cumulation of effect
sizes across these designs. The meta-analytic techniques of
Hunter et a1. (1982), Glass et a1. (1981) and Levine et a1.
(1973) currently do not provide the methods necessary for this
type of cumulation.

In conjunction with the presentation of this meta-analysis
the cumulation technique will be explained as will the strategy
for study selection suitable for use with this method. This
explication will be followed by the results of the meta-analytic

procedure and a discussion of these findings.

METHOD

The deception detection research with its habitual choice of
the within-subjects and mixed research designs, does not easily
lend itself to traditional meta-analytic procedures. The most
widely used meta-analytic techniques were developed for designs
that utilize between-subjects factors (Class at al, 1981; Hunter
et al,1982; Levine et a1, 1978; Rosenthal, 1978).

The reason that the effect sizes have not been assessed for
within-subjects factors is that the variance of these factors is
difficult to estimate if it is not provided by the authors.
Variance estimation for each factor in a design is critical in
the determination of the size of effect present.

In factorial designs the between-subjects variance is the
within-cell variance, which can be easily calculated from the
sumary statistics that are provided in research articles.
However, the variance for a within-subjects design is the
variance of the difference scores. If this variance is not
provided in a research report, it can not be calculated
straightforwardly. This difficulty in determining the variance in
within-subjects designs stems from the differing composition of
the F-ratio depending on which mean square estimates provide the

appropriate error term for that particular ratio. The error

25

Deception Detection - 26

terms utilized in the computation of this ratio are often
difficult to determine from the published information. This
difficulty is in direct contrast to the construction of F-ratios
in between-subjects designs in which the error term is always the
same mean square for each F-ratio that is computed. Since the F-
ratio is constructed in the same manner for each between-subjects
factor, the estimation of variance from these designs can be
carried out using the same set of techniques for each depending
on how much information is provided. The estimates of variance
yielded from the decomposition of summary statistics will be on a
consistent scale across all variables in the between-subjects
designs and across all studies that utilize this type design.
This is not the case in the within-subjects designs where a
specific set of techniques for variance decomposition are
difficult to employ given that the construction of F-ratios is
not consistent and is often difficult to determine.

The within-subjects studies share an additional problem in
terms of cumulating effect sizes across studies, e.g. that of
scale. If the variance can be identified for a within-subjects
factor it will not be on the same scale as the variance of other
within-subjects factors from the same design or with the variance
of factors from any other study. Effect sizes based on variances
that differ in scale from variable to variable have limited

utility in a quantitative cumulation of research findings.

Deception Detection - 27

In the past, researchers encountering repeated measures in
construction of meta-analyses have removed them from
consideration (Hunter, 1984; Mongeau, 1984). However, deception
detection research is not an area of study where the issue of
within-subjects factors can be caped with by simply excluding
those studies which employ repeated measures from cumulation of
data. Questions of interest in deception detection studies are
often directed at factors affecting the accuracy in veracity
judgments. The designs of these studies typically match these
research questions by asking the same observers to make such
decisions across a number of different observation conditions,
over a number of different deceivers/truthtellers or with
differing amounts of information. This type of design results in
judgments being made by the same individuals across different
conditions, e.g. a design of repeated measures.

Extraction of Effect Sizes for Repeated Measures: The Special
9133

Lacking complete sunmmry statistics, estimation of effect
sizes for within subjects factors is not possible. While this
dissertation does not provide an answer to the problem of
assessing within-subject effect sizes for all cases, it does
present a solution to cumulation of within-subject effect sizes
in a specific case. The special case addressed in this study is
that of measurements of phenomena recorded in naturally occurring

dichotomous units.

Deception Detection - 28

The research on accuracy in deception detection freely lends
itself to natural units in its assessment of detection accuracy.
For example, researchers often address questions such as the
following: "How many times was the observer correct in her
judgments of the deceptiveness of communicators?" or " What was
the percentage or proportion of correct judgments made by the
observer?" These questions and those like them can be easily
measured with counts of correct judgments that can be converted
into percents and proportions.

Conceptually, it is easy to see why this form of measurement
is popular with deception researchers. Researchers who report
their results in terms of counts, proportions or percentages
derive them by asking observers to make the the dichotomous
decision of "truth", if they believe the communicator is telling
the truth and "lie", if they believe the communicator is lying to
them. The observers' judgments then are compared to the actual
behavior of the communicator: i.e. Did the communicator actually
lie or tell the truth? The correct judgments are then tallied
and the results presented as proportions or percentages of
correct judgments or as a simple count of the number of times the
observers were correct. Measurements of accuracy in
deception detection reported in these natural units easily lend
themselves to a straightforward method of meta-analysis that is

appropriate for cumulation of study results for both within-

Deception Detection - 29

subjects and between-subjects variables. The repeated measures
and mixed designs of deception detection research, along with the
occasional between-subjects designs, can all be assessed using
this method providing the accuracy scores are reported in
percentages, proportions or counts. Results reported in counts
can easily be converted by the reader to proportions by noting
how many times the observers are asked to make judgments and
comparing the number of judgments on which the observer responded
correctly.

These naturally occurring measures provide the meta-
analyst with a constant unit magnitude that is scaled both within
studies and across studies. This constant unit can be used for
cumulative comparison, a type of comparison currently not
possible with estimates from within-subjects designs that do not
use this form of measurement.

Use of proportions or percentages for cumulation focuses on
the mean values for comparisons. This avoids drawing upon the
variance of differences from within-subjects factors for
comparison with the variance of differences from other within-
subject factors or for comparison with the within-cell variances
of between-subjects factors. Comparison of means with the same
scale of magnitudes allows the researcher to avoid the problem of
differing variance composition that are inherent in the
comparison of effect sizes computed with non-comparable

variances.

Deception Detection - 30

Proportions or percentages of correct veracity judgments
measured on a dichotomous rating of truth or lie create a scale
with possible values ranging from 0 to l with chance accuracy
represented by .5 or 50 percent. One benefit of this type of
scale, which may have drawn primary researchers to its use, is
that accuracy in deception detection can be easily understood. A
cumulative answer of .5 to the question of how accurate humans
are at detecting deception can easily be translated as meaning
people are correct in their judgments about half of the time.
This percentage/proportion scale is equally as useful for
understanding a question posed in a primary study as it is for
understanding a question posed in a meta-analysis.

The use of mean percent accuracy by Kraut (1980) as an
appropriate cumulative statistic in his quantitative sumary was
a good decision; however, the meta-analysis presented will avoid
errors made by Kraut: poor study sample selection, inclusion of
studies containing measurement scales and means that are not
comparable, and failure to look at the differing impact of
experimental conditions on the ability to detect deceptive

communication.

Deception Detection - 31

Sample Selection

The studies included in this meta-analysis were extracted
from extensive computer and manual searches through library
indices, abstracts and through investigation of the supporting
references that accompanied deception detection articles. The
measurement scales of the located articles where then assessed
in terms of the apprOpriateness for inclusion in this meta-
analysis.

The studies chosen for this meta-analysis were selected
because they included measurements of deception detection rated
on dichotomous scales of truth or lie that were converted to
proportions or percentages, or they included sufficient
information to convert the counts into proportions or
percentages.

The final requirement placed on the studies selected was
that they contain a counterbalanced manipulation of truth and
lies so that each person observed an equal number of truthful
messages and deceitful messages. Specifically, the ability to
detect deception can be defined as the ability to discriminate
lies from the truth. This ability can be tested by having
observers try to discriminate lies from truths when both are
present. Asking an observer to discriminate lies from truths
when only lies are present does not measure deception detection
ability because the measurement is confounded with the variable

of suspicion. In such a test of lie detection ability, highly

Deception Detection - 32

suspicious people will make the most judgments of deception and
hence will have the highest accuracy rating. These people may
actually have very poor lie detection ability as defined in this
meta-analysis, as they may consistently make the error of
suspecting truthful messages are deceptive. Experiments that
only supply truthful messages also provide a confounded measure
of deception detection. In this case very trusting or gullible
peeple would have the highest accuracy scores. Again these
peOple may be very poor at discriminating lies from the truth,
trusting not only those who are telling the truth, but also those
who are deceiving.

Thirty-two studies were found that conformed to these
specifications (see Table 1) representing a total of 3,439
observers. This represents approximately two thirds of the
observers examined in deception detection research.3 Studies
that were not included in this meta-analysis were excluded for
methodological reasons.

Excluded from this meta-analysis is lie detection research
that relied on Likert-type scales ranging from 1 to 7 or 1 to 10.
These studies, such as those by DePaulo, Rosenthal, Green and
Rosenkrantz (in press), Geis and Moon (1981), Hemsley and Doob
(1978), and Rotkin (1980) measured observer veracity judgments in

terms of degree of certainty in their judgments. For example, a

Deception Detection

Table 1

Studies Selected for Meta-Analysis

- 33

 

 

 

No Year Authors

1 1976 Atmiyanandana

2 1977 Bauchner, Brandt & Miller

3 1980a Brandt, Miller & Hocking

4 1980b Brandt, Miller & Hocking

5 1982 Brandt, Miller & Hocking

6 1974 Ekman & Friesen

7 1941 Fay & Middleton

8 1978 Harrison

9 1977 Hemsley

10 1979 Hemsley & Doob

11 1979 Hocking, Bauchner, Kaminski & Miller
12 1977 Lavrakas

13 1979 Littlepage & Pineault

14 1981 Littlepage & Pineault

15 1982 Littlepage & Pineault

16 1983 Littlepage, McKinnie & Pineault
17 1976 Maier & Lavrakas

18 1967 Maier & Janzen

19 1968 Maier & Thruber

20 1970 Matarazzo, Wiens, Jackson & Manaugh
21 1983 Miller, deTurck & Kalbfleisch
22 1974 Motley

23 1978 Parker
24 1982 Potamkin

25 1982 Rovira
26 1981 Sakai
27 1981 Sereno
28 1984 Stiff & Miller

29 1975 Wilson

30 1984 Zuckerman, Koestner & Alton

31 1984 Zuckerman, Kernis, Driver, & Koestner
32 Press Zuckerman, Koestner, & Colella

 

Deception Detection - 34

rating of "1" represented extreme certainty that the
communicator was lying, "2" indicated some certainty the
communicator was lying, "3" a guess the communicator was lying,
"4" uncertain whether the communicator was lying or telling the
truth, "5" a guess the communicator was telling the truth, "6"
somewhat certain the communicator was telling the truth, and "7"
extremely certain the communicator was telling the truth.
Measurements of lie detection accuracy rated on scales from (1)
to (10) were similar with inclusion of more degrees of certainty
in veracity judgments.

The basic assumption of researchers that measure lie
detection accuracy thus is that lie detection is a matter of
degree. Therefore, high accuracy scores can be achieved by
extremely certain judgments that are correct. People who are
extremely certain in their correct decisions receive higher
accuracy scores than pe0ple who are somewhat certain and
correspondingly people who are extremely certain in their
incorrect decisions are less accurate than people who are less
certain of their incorrect decisions. The weakness of using this
measurement scale to assess accuracy in deception detection is
that the observers accuracy judgments are confounded with an
assessment of personal assuredness in their decision.

Research by Miller and Kalbfleisch (1982), Hocking (1976),

and Littlepage and Pineault (1979) indicates that the

Deception Detection - 35

relationship between accuracy in veracity judgments and
confidence in these judgments is weak. Hocking (1976) reported
correlations between accuracy and confidence ranging from .061
to .063. Miller and Kalbfleisch (1982) noted that high confidence
scores accompanied low accuracy ratings. Miller and Kalbfleisch
also found that observers were more confident in their judgments
of women than in their judgments of men.

Inclusion of a measure of confidence into an accuracy
measure such as in the DePaulo et al. (in press), Geis and Moon
(1981), Hemsley and Doob (1978) and Rotkin (1980) studies,
results in an accuracy measure that is confounded by an unrelated
variable. For example, this scale will measure cautious
observers that make mostly accurate veracity decisions as less
accurate than observers who are extremely certain on a few
correct judgments and unable to make a judgement of truth or
deceit on the remainder of the messages. 0n the other hand, some
observers may express extreme confidence in their judgments of
truth, but be cautious in their attributions of deceit.
Researchers finding significant differences in observers' ability
to detect deception in male as Opposed to female communicators
through use of likert type scales may actually be finding sex
related differences in judgmental confidence and not differences
in discerning truth from lies.

Studies that utilized a second technique to assess accuracy

in deception detection were also excluded from this meta-

Deception Detection - 36

analysis. These studies (DePaulo, Davis S Lanier, 1980; DePaulo
Lanier S Davis, 1983; DePaulo, Lassiter S Stone, 1982; DePaulo S
Rosenthal,l979; DePaulo, Stone S Lassiter, submitted; Olson,
1978; and Streeter, Krauss, Olson S Apple, 1977) had observers
rate their veracity decisions on scales from 1 to 6 or from -3 to
+3 with the lower rating indicative of extreme. certainty the
communicator was lying and the highest rating indicative of
extreme certainty the communicator was telling the truth. The
ratings between these extremes represented lesser degrees of
certainty in the observers' decisions. Researchers using this
method then subtract observers' ratings of truthfulness when the
communicators were telling the truth from observers' ratings of
truthfulness when the communicators were lying. The resulting
value was used to represent the accuracy score of each observer.
These accuracy scores are confounded by confidence in judgments
as were the untransformed Likert measures of accuracy. Extreme
confidence in a correct judgment is also assumed to be a more
accurate assessment of veracity than a less. confident correct
judgement.

Some studies have tried to measure deception indirectly by
measuring differences in pleasantness (Feldman, 1979; Feldman,
Jenkins S Popoola, 1979), and differences in genuineness feedback
(Feldman, 1979). For example, Feldman (1979) and Feldman et a1.

(1979) asked judges to rate the pleasantness expressed by

Deception Detection - 37

children as they drank Kool-Aid that was either sweetened
(pleasant) or unsweetened (unpleasant). Children were asked to
either express their actual satisfaction with a drink or to
express their opposite reaction to it. Pleasantness was then
rated on a six point scale by the judges and the experimenters
converted these scores into difference values by subtracting
ratings of pleasantness when the children where lying from
ratings of pleasantness when the children where being honest in
their expression of emotion. The inference in these studies is
that ratings of pleasantnesslunpleasantness can be converted into
ratings of truth/lie so that accuracy in deception detection can
be inferred.

There are several problems with this inference. First,
children may differ in their abilities to communicate
pleasantness and unpleasantness. Charlesworth and Krevtzer
(1973) have noted that children are at lower levels of cognitive
development and have less control over their facial muscles than
adults. These children may not have been able to express the
emotions the researchers expected of them. Second, the deception
detection judges may differ in their definitions of pleasant and
unpleasant behavior. This may have resulted in unreliable
measurement. Third, the children may not have perceived the
sweet drinks to be pleasant and the unsweetened drinks to be
unpleasant. Hence, the researchers may have been unaware whether

children were actually lying or telling the truth. Finally,

Deception Detection - 38

"unpleasantness" is one of the cues Mehrabian (1977) cites as
being indicative of deceptive communication. Therefore, children
being truthful about the unpleasantness of the drink may have
been more likely to be judged as deceptive than those being
deceptive about the unpleasantness of the drink. Conversely,
children lying about the pleasantness of a drink may have been
judged as more deceptive than children being truthful about the
pleasant taste.

Studies using indirect measures of deception and truth such
as those used in Feldman (1979) and Feldman et al. (1979), have
been excluded from this meta-analysis.

Other studies not included in this meta-analysis have
idiosyncratic components that make them unsuitable for
cumulation. The deception detection studies by Geizer, Rarick
and Soldow (1977) and Zuckerman, Amidon, Bishop and Pomrantz
(1982) were excluded because they used a trichotomous measure of
accuracy instead of a dichotomous measure. For these studies
chance accuracy would be .333 rather than .50. Percentage
accuracy means from these studies can not be accurately cumulated
with the percentage accuracy means generated from dichotomous
judgmental choices and ultimately tested against an overall
chance accuracy of .50.

Kraut and Poe (1980) and Fugita, Hogrebe and Wexley (1980)

were also be excluded from this meta-analysis because neither

Deception Detection - 39

study reported the experimental means. Finally, Hildreth (1953)
and Littlepage and Pineault (1978) are excluded. While these
researchers provided observers with a dichotomous truth/lie
measure, observers were also given the option to avoid making a
decision if they were not sure Of their judgment. The difficulty
in including these studies is that the mean percentage accuracy
is based on only confident judgments, unlike the other studies in
the meta-analysis which did not exclude the judgments made in
uncertainty. It is possible that by excluding uncertain
judgments Hildreth (1953) and Littlepage and Pineault (1978)
measured only those observers with confidence in their deception
detection ability. Conversely it is also possible that the
accuracy ratings from these studies is yielded from judgments
based on only those communicators who presented themselves in an
obviously deceitful or honest umnner.

The studies of deception detection included in this meta-
analysis are those that assessed accuracy by dichotomous
truth/lie measures without a no judgment Option. These studies
all reported mean accuracy ratings in proportions/percentages or
their results were convertible to proportions/percentages.

Procedures for Cumulation

 

The first step of this meta-analysis consisted of searching
for deception detection studies and selecting studies for
cummlation that met the specifications outlined above. Second,

the nean accuracy score for each level of research design was

Deception Detection - 40

extracted. These mean accuracy ratings were then placed into
tables with other means taken from studies that measured these
same variables. These tables are arranged according to the
combinations of variables and levels that conceptually‘ form
general designs for the deception detection studies. Entries in
these tables are weighted and averaged to provide cumulative
estimates of the effect of each variable. Information is provided
on the number of Observers that each mean effect size represents.
Means from deception detection studies that investigate variables
not considered in other experiments are presented narratively
after the cumulative estimates have been discussed.

In evaluating the effect sizes yielded from this meta-
analysis, the size of each cumulative mean will be considered
relative to the chance criterion of .5. Since variances can not
be determined for the means represented in these cumulative
estimates, significance tests of the difference between the means
and .5 are not possible. Instead visual comparisons of relative
size are made. In the analysis of cumulative means, these
comparisons are made in light of the sizes of accuracy ratings
common to this area of research. Since variance estimates are
unavailable, sampling error can not be assessed in this meta-

analysis.

RESULTS

The results of this meta-analysis indicate that humans are
not very skilled at detecting deception. The cumulative accuracy
ratings cluster from .45 to .70 with only a few cumulative
accuracy ratings surpassing or falling behind these scores. These
results also suggest that lie detection ability varies only
slightly across experimental conditions. Keeping these modest
differences in mind, the design tables from this meta-analysis
will be evaluated for those variations present in the human

ability to detect deception.

Observational Conditions

The first three tables of this meta-analysis examine the
support for two major theories of deception detection: Ekman and
Friesen (1969, 1974) and Maier and Thurber (1968).

Ekman and Friesen (1969, 1974) suggest that peOple are less
aware of their bodies and extremities than they are their faces.
They contend that since the face has a larger message sending
capacity, most people learn to control their facial movements in
order to accurately communicate with others. Consequently, when
people are confronted with the task of concealing or distorting
messages they will exercise control over their facial regions.

Accordingly since the body and extremities have less sending

41

Deception Detection - 42

capacity for messages (Ekman S Friesen,l969), peOple will have
used them less to convey meaning and therefore be less aware Of
non-facial movement when they are concealing or distorting
messages. Ekman and Friesen reason that given this tendency of
communicators to concentrate on control of the facial regions and
to remain unaware Of other body movement, Observers attempting to
detect deception should focus on changes in body movement and to
place less emphasis on facial displays.

Table 2 displays the cumulative results of studies that
contained experimental conditions that test this theory. The
observational condition containing full shots of both heads and
bodies was added to the design to examine the impact of full
visual information in comparison to the body only and head only
conditions.

While the differences between these conditions appear small,
the experimental condition of body only, where persons were
allowed to Observe only the bodies of communicators, yields the
highest mean accuracy ratings. Conversely, observers were the
least accurate in their Observations of the head only condition.
This cumulation of mean accuracy ratings supplies some support
for Ekman and Friesen's theory. Specifically, Observers in the
viewing condition that Ekman and Friesen posit displays the most
deception clues, the body, were more accurate than those
Observers who viewed the area Ekman and Friesen maintain is the

most readily controlled, the face.

Deception Detection - 43

Table 2

Deception Detection Accuracy: Conditions Under Which Truthful
and Deceptive Messages are Observed

 

 

 

 

Head
Head Body S
No Study ‘ Only Only Body
2 Bauchner, Brandt S Miller .47
6 Ekman S Friesen .46 .52
11 Hocking, Bauchner, Brandt S Miller .51 .51 .52
13 Littlepage S Pineault .53 .76
14 Littlepage S Pineault .49
29 Wilson .49 .61 .60
32 Zuckerman, Koestner S Colella .59
Weighted Mean .51 .54 .53
No. of Observers Represented by Estimate 1395 1169 984

 

Those observers who viewed full head and body shots were
more accurate than those Observers that viewed heads only, but
were less accurate than those Observing bodies only. Deception
clues supplied by the body, in the combined head and body
condition, may enhance accuracy ratings when compared to ratings
yielded from head only Observations. On the other hand, full
body and head information may inhibit accuracy ratings achievable
when viewing bodies only. In this instance, the face, which is
readily controlled by the communicator, may present contradictory

information to the body's messages that confuses the observers

Deception Detection - 44

or facial cues may be attended to more readily by observers than
cues from the body.

Maier and Thurber (1968) theorize that Observers are
overwhelmed by verbal, nonverbal, and content information when
they decode messages. This overabundance of information
distracts the observers from noticing clues that indicate they
are being deceived. Maier and Thurber reason that deception clues
will be more apparent if nonverbal information is reduced.

In Maier and Thurber's test Of this theory, three
experimental conditions were created: 1) watchers (who watched
and listened to deceptive/truthful messages), 2) listeners ( who
listened to deceptive /truthful messages), and 3) readers (who
read transcripts of deceptive/truthful messages). Their study
found that those who listened and read were the most successful
at detecting deception .77 in both cases. Observers who both
watched and listened were less successful in deception detection
with a mean accuracy rating of .58. Based on these results,
Maier and Thurber reasoned that to accurately detect deception,
communication should not be visually observed.

Table 3 presents the cumulative results of studies which
test the experimental conditions of transcripts, audio only, and
audio/visual. The results presented in Table 3 expand the

conditions originally tested by Maier and Thurber by adding a

Deception Detection - 45

Table 3

Deception Detection Accuracy: Conditions Under Which Truthful
and Deceptive Messages are Observed

 

 

Visual Audio Audio/ Tran-

 

 

NO Study Only Only Visual Script
2 Bauchner, Brandt S Miller .47 .32 .47
6 Ekman S Friesen .49
7 Fay S Middleton .56

ll Hocking, Bauchner, Kaminski S

Miller .48 .54 .55 .57

13 Littlepage S Pineault .64

14 Littlepage S Pineault .49 .63

19 Maier S Thurber .77 .58 .77

22 Motley .67

26 Sakai .51 .58 .58

29 Wilson .62 .53 .57

31 Zuckerman, Kernis, Driver S

Koestner .51
32 Zuckerman, Koestner S Colella .56 .62 .62
Weighted Mean .51 .58 .57 ‘61

No.of Observers Represented by Est. 1623 1676 1536 1018

 

condition of visual only, where the observers can see but not
hear deceptive and truthful messages. This addition, adopted by
some primary researchers, balances the audio only condition in
this design.

The cumulative estimates reported in Table 3 indicate
Observers were the most accurate at detecting deception when they
read transcripts of deceptive/truthful messages. These results

support Maier and Thurber's contention that reduction of

Deception Detection - 46

nonverbal information will increase accuracy in detecting
deception. In this condition both visual and paralinguistic cues
were unavailable to the Observes. The cumulative findings
reported in Table 3, also suggests that persons with access to
only audio cues and persons with access to both audio and visual
cues have the same accuracy rates. This observation does not
support Maier and Thurber's position that due to absence of
distracting nonverbal information, Observers in the audio only
condition will be able to perform better than those in the full
information condition. The new condition added to Maier and
Thurber's original three, diSplayed the lowest accuracy ratings.
When compared to the full audio visual condition, the low rating
in the visual only condition suggests that the audio channel
provides helpful information for the detection of deception.
Table 4 incorporates both the head and body conditions on
which Ekman and Friesen concentrated their theory, with the
audio, audio/visual and transcript conditions focused on by Maier
and Thurber. The full head and body condition and the visual
only condition added to Table 2 and Table 3 are also included.
This combined design suggests that persons are the most
accurate in detecting deception when Observing 1) the transcript
only condition, 2) the audio only condition, 3) the body only

condition with full audio information, 4) the combined head and

Deception Detection - 47

body viewing condition with full audio information, and 5) the
head only condition with full audio information. The three
conditions with the lowest accuracy ratings are: 1) the body only
viewing condition without sound (6th), 2) the combined head and
body viewing condition without sound (7th) and finally, 3) the
head only viewing condition without sound (8th).

In general this table indicates that Observers were more
accurate in Observational conditions that contained the content
of the message than observational conditions that did not. These
higher accuracy scores may have resulted from additional
information available in message content. This information may
have allowed Observers to check messages for logic, consistency
and length.

Observers viewing body only shots were more accurate than
those viewing heads only. However, when Observers viewed
combined head and body shots, they were less accurate than when
viewing bodies only and more accurate than when viewing heads
only. While the body may be a source Of helpful clues to
deception as Ekman and Friesen contend (1969, 1974), this may
only be the case when it is viewed alone without a view of the

face and head.

 

 

 

mace same «as ncua ~m~_ Ana men an- masseuse an subcommuaoa mumsammno no .oz
~o. an. an. an. an. an. as. on. as»: embewaoz
No. on. No. wgflufloo e uncommox .cmauuxoam mm
mm. «a. an. "a. as. comma: aw
on. an. an. «exam ea
3 . .3: no: «N
nu. us. an. uncanny a Howe: mg
no. as. so. basemaam a mamamauuaa «A
oh. mm. Adamoeam a mwmaoauuan me
an. em. as. on. as. mm. as. an. cause: a Assamese .nocaUamm .mcua°ox an
on. counseeaz a awe a
«m. as. commune a cease e
no. mm. he. mm. uofiam: a Deccan .uocnosmm N
ummuum mace Omva<loz Omo=< Owo=<loz Omo=< Omos<loz vaa< hmoum oz
lemma Omo=< moon zoom hHso mace afico mane
a a boom boom 28: use:
one: one:

 

 

ou>ummno was someone: Houufluuon one anemones sownz moms: seemuwmcoo “sewquuon newueouon cw zomuaoom

o

a manna

Deception Detection - 49

Familiarity

 

Some researchers have examined whether individuals are more
accurate in deception detection when they have some knowledge
regarding the typical communication style Of the person they are
judging (Brandt, Miller S Hocking, 1980a; Brandt, Miller S
Hocking, 1980b; Brandt, Miller S Hocking, 1982; Ekman S Friesen,
1974). They reason that familiarity with this style should
provide Observers with a frame of reference from which deviations
in communication patterns can be spotted.

In these experiments familiarity was operationalized by
showing observers a short segment of communicators telling the
truth. Observers were told that they were viewing truthful
samples of behavior. These truthful segments were shown prior to
the message segments where observers were asked to detect lies
and truths.

Table 5 presents the cumulative results from studies that
provided observers with truthful baseline messages. These
truthful baselines were either not shown tO Observers, or they
were shown one, two, three or six times prior asking subjects to
make veracity assessments. This table reveals that in all

conditions observer given a frame of reference were more accurate

 

on and mm Nam «we oumawumm an ouucommuemz muu>ummno mo .02

 

 

.3. S. on. S. «a. use: 323...:
~m. so. comumum a emaxm c
on. an. an. maeauom a amaze: .ueesum n
so. we. manage: a tween: .ueemum s
as. am. An. mm. manage: a amaze: .uecmaa m
nonsmooxm o monomooxm m nouanoexm N mounmomxm u homemadwamm aozum oz
aumummawamm humwmammamm humummawamm humummafiamm oz

 

 

cemuOOuon nodumouon cw humusoo< «swash ecu neumowcsasoo nun: homemadfiamm new~Ommm

m magma

Deception Detection - 51

in detecting deception than Observers without this base for
comparison. Observers who viewed the truthful baseline messages
three times were the most accurate, followed by those who viewed
the baseline two times, and those who viewed the baseline once.
This pattern seems to indicate that the more familiar observers
are with a communicator's idiosyncratic truthful behavior, the
better able they are to detect deviations from this behavior, and
in turn, the better they are able to detect deceptive
communication.

While this pattern of increased accuracy ratings is evident
for baseline observations of up to three message repetitions,
Observers who viewed truthful baseline messages six times had
lower accuracy ratings than any of the other conditions where the
familiarity segment was provided. An explanation for this
decreased accuracy may be that observers in this condition became
fatigued or bored watching the same segment six times. This
boredom or fatigue may have reduced observers efficiency in

information processing.

Sex Differences in Deception Detection

Male versus female ability to detect deception has also been
explored by prior researchers. Table 6 displays the results of a
cumulation of studies that examined sex differences in detecting

deceit. The cumulative mean accuracy ratings suggest that

Deception Detection - 52

Table 6

Judges Sex and Accuracy in Deception Detection

 

 

 

 

No Study Male Female

1 Atmiyanandana .55 .52

7 Fay S Middleton .55 .56

9 Hemsley .51 .54
19 Maier S Thurber .69 .74
23 Parker .51 .50
25 Rovira .52 .54
26 Sakai .54 .57
Weighted Mean .59 .61
No. of Observers Represented by Estimate 331 327

 

females are better at detecting deception than males. The
results are consistent with nonverbal research on sex differences
and decoding ability. Hall (1978,1980) has noted that females
are more skilled in decoding nonverbal messages than men. This
heightened sensitivity to nonverbal cues that allows females to
more accurately interpret their meaning may also be of use in
detecting discrepant nonverbal cues or other such clues that
deception may be occurring. Tables 7 and 8 break down this
detecting ability across Observation conditions for males and

females respectively.

Deception Detection - 53

Table 7

Deception Detection Accuracy for Male Judges

 

 

Audio/ Tran-

 

 

No Study Visual Audio Visual Scripts
l9 Maier S Thurber .57 .76 .73
26 Sakai. .49 .57 .57
Weighted Mean .54 .68 .57 .73
No.of Observers Represented by Est. 205 205 90 115

 

Table 8

Deception Detection Accuracy for Female Judges

 

 

Audio/ Tran-

 

NO Study Visual Audio Visual Scripts
19 Maier S Thurber .63 .78 .80
26 Sakai .53 .60 .59

 

Weighted Mean .58 .70 .59 .80
No.of Observers Represented by Est. 194 194 90 104

 

Deception Detection - 54

The Observation conditions examined in these tables appear
to have similar moderating effects on the accuracy of both male
and female lie-detectors. However, in each observation condition
female Observers maintained higher rates of success in detecting
deception.

Sex differences in deception detection ability have also
been studied in terms of the sex of the communicators for whom
male and female observers make judgments. The rationale for
these comparisons in primary research centered on whether
observers were better able to identify lies and truths
perpetrated by same or opposite sex communicators (i.e. Parker,
1978; Rovira, 1982). Results cumulated from these studies are

presented in Table 9 and Table 10.

Table 9

Male Judges: Deception Detection Accuracy Observing
Male vs Female Communicators

 

 

 

 

No Study Male Female
7 Fay S Middleton .57 .53
9 Hemsley .51 .51

23 Parker .46 .49

25 Rovira .47 .56

Weighted Mean .50 .52

NO. of Observers Represented by Estimate 90 90

 

Deception Detection - 55

Table 10

Female Judges: Deception Detection Accuracy Observing
Male vs Female Communicators

 

 

 

 

No Study Male Female
7 Fay S Middleton .58 .54
9 Hemsley .51 .56
23 Parker .46 .52
25 Rovira .50 .58
Weighted Mean .50 .54
NO. of Observers Represented by Estimate 97 97

 

The cumulative means show that both male and female judges
are more accurate when they are judging females than when they
are judging males. While female observers were better able to
judge members of their own sex and males were better able to
judge members of the opposite sex this finding may be more
related to the deceptive and truthful communication patterns of
females and males, and less related to specific Observer skills

in detecting members of their own or Opposite sex.

Deception Detection - 56

Table 11

Sex of Communicators and Accuracy in Deception Detection

 

 

 

 

No Study Male Female

7 Fay S Middleton .58 .54

9 Hemsley .51 .53
12 Lavrakas .52 .58
23 Parker .47 .50
25 Rovira .49 .57
Weighted Mean .51 .54
No. of Observers Represented by Estimate 287 287

 

Table 11 presents the combined ratings of male and female
observers broken down by the sex of the communicators who were
being judged. These results show that observers may be able to

more accurately detect deception by females than by males.

Judgments of Truth vs. Lie

 

The studies included in this meta-analysis were all
counterbalanced so that Observers were presented with equal
numbers Of truths and lies. Only counterbalanced studies were
included in order to reduce the effect of guessing bias on the
accuracy measures. Table 12 shows this preference for judgments
of truth or deceit. From this cumulative finding it appears that
observers are more likely to judge communicators to be telling

the truth than to judge them to be lying.

Deception Detection - 57

Table 12

Observer Judgments of Truth vs Lie

 

 

 

 

No Study Truth Lie
6 Ekman S Friesen .45 .53

7 Fay S Middleton .50 .61

8 Harrison .75 .48
12 Lavrakas .50 .60
13 Littlepage S Pineault .68 .60
15 Littlepage S Pineault .81 .30
16 Littlepage, McKinnie S Pineault .57 .60
18 Maier S Janzen .34 .50
19 Maier S Thurber - .71 .71
24 Potamkin .68 .58
28 Stiff S Miller .67 .41
31 Zuckerman, Kernis, Driver S Koestner .60 .41
32 Zuckerman, Koestner, S Colella .60 .58
Weighted Mean .59 .55
No. of Observers Represented by Estimate 1233 1233

 

Narrative Review

The studies included in this narrative review met the
sampling requirements of the meta-analysis for measurement and
counterbalancing. Even so these studies can not be cumulated
because few other studies examining the same variables are
available, or meet the sampling requirements. The low accuracy

ratings in the cumulative results are also prevalent in these

Deception Detection - 58

findings. The differences between experimental conditions are
also small. These estimates are based on smaller samples than
most of the cumulative findings and the variations suggested in

this narrative review should be interpreted with caution.

Deceiver differences: rehearsal, self-monitoring age and
.2£2&_° Researchers have explored the detectability of
communicators who have rehearsed their truthful and deceptive
messages (Littlepage S Pineault, 1982; Miller, deTurck S
Kalbfleisch, 1984). In their 1982 study, Littlepage and Pineault
found that if observers were given time to plan their messages,
observers had more difficulty detecting deception (.52) than when
they were not given rehearsal time (.59).

Miller et al. (1983) further explored the impact of planning
on observer success in deception detection by adding a self-
monitoring measure (Snyder, 1974) to this design. In this study,
message rehearsal time differentially affected detectability
depending on the degree to which communicators self-monitored.
High self-monitors were more difficult to detect when given time
to plan their messages (.45) than when not given time to rehearse
(.50). Conversely, low self-monitors were easier to detect after
given time to rehearse (.56) and more difficult to detect when
their lies were spontaneous (.53). These researchers reasoned
that high self monitors had greater confidence in their

performance abilities, and therefore used the planning time to

Deception Detection - 59

rehearse and better prepare their responses to an interviewer's
questions. On the other hand, low self-monitors were not
confident in their performance abilities, and became nervous and
apprehensive when given time to think about the messages~ they
would soon have to communicate. Consequently, low self-monitors
were not as successful as they would have been without the
rehearsal time. In both the rehearsal and no rehearsal
conditions high self-monitors were more difficult to detect than
low self-monitors.

Age may also affect the detectability of deceivers. Parker
(1978) found the combination of sex and age differentially
affected Observer success in deception detection. Specifically,
thirteen to fourteen year-old females were easier to,detect (.52)
than adult females over eighteen (.50) and seven-year-Old female
children (.48). Conversely, thirteen to fourteen year old males
were easier to detect (.48) than seven-year-old males (.47) and
adult males over eighteen (.45).

Potamkin (1982) studied deception by heroin addicts. Given
the experience addicts have had in hiding their habit from
others, she reasoned that addicts should be harder to detect than
nonaddicts. However, the study found that addicts were easier
to detect (.64) than were nonaddicts (.62). It appears from this
study that despite their experience in deceiving others about

their habits, addicts are no more skilled than others.

Deception Detection - 60

Messgge characteristics. The type of message also appears

 

to affect how accurate Observers will be in deception detection.
Hocking, Bauchner, Kaminski and Miller (1979) found that
observers more accurately detected factual lies (.54) than lies
about emotional states (.50). Rovira (1982) found that observer
accuracy was affected by whether or not they agreed with the
content of a communicator's message. The results of Rovira's
study suggest that observers are more accurate in detecting
deception when they agree with the message (.53), than when they
disagree (.52). Finally, Maier and Lavrakas (1976) found that
persons observing messages that a polygraph had correctly
identified, were better able to detect deception (.68), than when
they observed messages which the polygraph was unable to
correctly identify (.51).

Social-cultural observer differences. While this meta-
analysis has suggested that females are slightly better at
deception detection than males, the primary research by Parker
(1978) suggests that this ability may differ according to age. In
this study adult females over eighteen were better at lie
detection (.51) than seven-year-old females (.50), and thirteen
to fourteen year old females (.49). Thirteen to fourteen year
old males, on the other hand, were more accurate (.52) than were
seven-year-old males (.50) and adult males over eighteen (.50).
According to the results yielded from this study, females are

somewhat better lie detectors than males when they are adult,

Deception Detection - 61

equal in lie detection ability as children and worse than males
as teenagers.

Instead of examining observer age, Sereno (1981) looked at
past experience in his study of success in detecting the truths
and lies of children. This study found that elementary school
teachers were the most accurate in judging children (.77).
Sereno had posited that those observers with the most experience
with children would be the best at determining when they were
lying and telling the truth. While teachers had the highest
accuracy rates, adults who were parents of children were the
least accurate in detecting the lies of children (.63). Adults
with no children were slightly more accurate than the adults with
children (.64). In interpreting these results, it could be
reasoned that teachers were better able to detect children's
deception because of the wide range of children they must work
with on a daily basis. This experience may have made them aware
of children's typical styles of communicating, which in turn
could have allowed them to better discern when the children were
deviating from these patterns. Conversely, the parents of
children may only have been exposed to the idiosyncratic
communication behaviors of their own children, hence they could
not generalize this knowledge as well as the teachers. The
childless adults may also have been generalizing from limited

knowledge.

Deception Detection - 62

In her study of heroin addicts discussed in the proceeding
section, Potamkin (1982) examined the differential deception
detection abilities of addicts and nonaddicts. Potamkin found
that nonaddicts were slightly more accurate at detecting
deception (.64) than were addicts (.62).

Finally, Atmiyanandana (1976) considered differences in
deception detection accuracy by observers from different
nationalities. In his study Asian, Latin American, and North
American observers were assessed. Atmiyanandana found that Asian
judges were the least successful with an accuracy rate of .50,
while both North and Latin Americans had a slightly higher
accuracy rate of .54.

Experimentally induced observer differences. Several

 

researchers have studied the impact of various experimental
manipulations on ability to detect deception. For example, Motley
(1974), explored whether telling Observers to attend to a
specific nonverbal cue would increase their accuracy in deception
detection. In his study Motley told half his observers to pay
particular interest to speech latency while listening to audio
tapes. Results indicated that the observers told to attend to
speech latency were less accurate (.63) than those who were given
no listening clues (.67). This study could be interpreted as
indicating that directing observers' attention to specific cues
associated with deception does not increase the ability to Spot

deceivers and truthtellers. However, it may also be the case

Deception Detection - 63

that speech latency by itself is not that helpful an indicator Of
deception, and other cues more strongly associated with deception
might produce greater Observer accuracy.

Zuckerman, Koestner and Alton (1984) attempted to teach
observers how to spot lies and truth by showing them audio/visual
tapes of communicators and telling them when the. communicators
were lying and when they were telling the truth. In this study,
the accuracy of observers given no instructions was .62, while
the accuracy ratings for the teaching conditions ranged from .61
to .70 depending on the instruction method used.

The highest accuracy rating in Zuckerman et al. (1984) was
shared by observers in two experimental conditions. In one of
these conditions experimenters showed observers tapes of eight
communicators. After each tape the researchers had observers
make veracity judgments followed by instructions regarding which
communicators were lying or telling the truth. These observers
reached an accuracy level of .70, with accuracy computed on all
eight judgments. The other condition that achieved a .70
accuracy rating was one in which observers were told before the
first four tapes which communicators were telling the truth or
lying, and told after the last four tapes which communicators
were lying or telling the truth. Accuracy scores were computed
on judgments of the last four tapes. Observers told which

communicators were lying or telling the truth before the first

Deception Detection - 64

four tapes but given no information for the last four tapes were
accurate at the .66 level, with accuracy computed on the last
four judgments. The results of these conditions suggest that
Observers may have been learning how to spot deceptive and
truthful communication.

Observers in two conditions in the Zuckerman et al. (1984)
study had accuracy levels below the control group. In one of
these conditions observers were informed after the first four
tapes which communicators were deceptive or truthful, but not
after the last four tapes. Observers in this group were accurate
at the .61 level, with accuracy computed on all eight judgments.
In the experimental condition with the lowest accuracy, Observers
were given half truthful and half false feedback concerning which
communicators were lying or telling the truth. This feedback was
provided after each judgment was made on the eight communicators.
Accuracy in this condition was .53, computed on all judgments.
The Zuckerman et a1. (1984) rationale for inclusion of this
condition was to examine whether observers were actually learning
how to Spot deceivers, or whether they were simply making their
judgments based on anticipated proportions of truth and
deception. The same order sequence of truth and lie feedback was
provided in this condition as was provided in the eight-after
condition (with truthful feedback), which yielded the high
accuracy score of .70. Zuckerman et a1. reasoned that if

observers were using feedback to calculate probabilities of

Deception Detection - 65

correct responses in both the veridical eight after group and in
the false-eight after group their accuracy scores should be
similar. But if Observers were using the feedback to learn to
spot truth and lies the observers with the veridical information
should have superior accuracy scores to those in the false
feedback condition. From the comparison of the false condition
to the veridical one it appears that some learning may have been
taking place among the deception judges.

Stiff and Miller (1984) did not provide observer with
feedback, but they did provide observers with additional
information not typically available in traditional deception
studies. These researchers showed observers tapes of
communicators lying and telling the truth in response to an
interviewer's question. After the communicators had responded
to this question the interviewer then further probed their
reaponse. These probes either suggested the interviewer believed
the communicators or disbelieved them. The video tapes shown to
the observers included the communicators' responses to these
probes. Findings indicate that observers viewing communicators
experiencing a positive probe (i.e. indicating the interviewer
believed communicator) were able to detect detection at the same
accuracy rate (.54) as observers who viewed communicators
experiencing a negative probe (i.e. indicating the interview

disbelieved subject).

Deception Detection - 66

Finally, most deception research has observers make
judgments of the veracity of communicators from a vantage point
in which they do not have to interact with the communicators.
Instead of this pOpular Observation position, Hemsley and Doob
(1979), had 27 persons interview communicators and make veracity
judgments. The interviewers were told to make their veracity
decisions based on observation of behavioral cues, and not by
attempting to trap the respondent with questions. These
interviewers detected deception at a .58 level of accuracy.

Hemsley and Doob (1977) did not have a comparison condition
for subjects Observing communicators only. However, Matarazzo
et a1. (1970) examined this issue by having one person interview
communicators and another person watch the interviews. In this
study, the interviewer had an accuracy rate of .58 and the
observer had an accuracy rate of .60.

In summary, the research concerning the experimental
conditions discussed in this section is mixed. Directing
observers' attention to deception clues was found to decrease
accuracy. While providing observers with some training may
increase accuracy. In general, positive and negative probes by
interviewers did not differentially affect deception detection
accuracy. Those interviewing deceivers also appeared to be less

accurate than those who only Observed.

DISCUSSION

Comparison with Previous Meta-Analyses

The findings of this meta-analysis are both similar and
different from the findings generated by Zuckerman et al. (1981)
and Kraut (1980). One of the major findings of this quantitative
review was that humans were the most accurate in detecting
deception when they had access to message content. Transcripts
were found to yield higher accuracy than audio information, and
the transcript only condition facilitated the highest accuracy
ratings.

Zuckerman et al. also found that people were more accurate
when they had access to message content. However these
researchers found that observers exposed to audio information
were more accurate in detecting deception than those who were
eXposed to transcripts only. Expressed in 'standard deviation
units, accuracy scores yielded by the audio only condition had a
mean of 1.09, and accuracy scores for the transcript only
condition had a mean of .70. Zuckerman et al. also found that
all visual Observational conditions when combined with audio
information yielded higher accuracy scores than the transcript
only condition. Observations of the body only with audio had a

mean of 1.49. Observations Of the body and face with audio had a

67

Deception Detection - 68

mean of 1.00. And, observations of the face only with audio
yielded a mean of .99. These accuracy scores, expressed in
standard deviation units, suggest that the audio channel may be
more useful for detecting cues to deception than the transcripts.
Zuckerman et al. found all five of these observational conditions
yielded accuracy scores significantly greater than chance.

A second major finding of this meta-analysis was that in the
context of visual Observations, shots of the body yielded the
highest accuracy ratings. This condition was followed by accuracy
ratings from the combined body and face condition, and the head
only condition. This pattern was the same for visual conditions
with audio and the visual conditions without audio; with visual
conditions with audio having higher accuracy ratings than visual
conditions without audio.

Zuckerman et al. also found this pattern. With audio
information, body shots yielded the highest accuracy (1.49),
followed by combined body and head shots (1.00) and shots of the
head only (.99). Without audio information, body shots yielded
the highest accuracy (.43), followed by combined body and head
shots (.35), and shots of the head only (.05). These accuracy
ratings, expressed in standard deviation units, were all found to
be significantly greater than chance, except for the shots of the

head only without audio.

Deception Detection - 69

Zuckerman et al. concluded their meta-analysis by indicating
that observational conditions substantially moderate deception
detection. These researchers further indicate that humans are
more accurate than chance in detecting deception in all of the
Observation conditions with the exception of the head only
condition without audio.

These statements seem to conflict with the results of this
meta-analysis, which found human accuracy to be low and to not
differ greatly from the chance rate of .50. The "substantial"
differences in observational conditions are also in conflict with
the slight variations found in this meta-analysis.

One explanation for these differences may be that the d
statistic as operationalized by Zuckerman et al., is inflated and
has provided estimates of accuracy much larger than the actual
population values. This inflation might account for substantial
differences and for the significant accuracy rates yielded in the
Zuckerman et a1. meta-analysis. Furthermore, these inflated
values might also explain the differences in accuracy patterns
for the transcript condition and audio conditions.

Other results from this meta-analysis and the Zuckerman
et al meta-analysis are more similar. Both meta-analyses found
that: 1) females were better able to detect deception than males,
2) observers could more easily detect female deceivers than male
deceivers, and 3) high self-monitors were more difficult to

detect than low self-monitors. This meta-analysis found these

Deception Detection - 70

differences to be slight, and Zuckerman et al. also found that
the differences not to be significantly more than chance.
Perhaps these differences were actually so small that even
inflated d values could not make them appear substantial.

This meta-analysis found that familiarity with s truthful
baseline Of the person to be judged facilitates increased
accuracy scores. The exception to this pattern was found when
the baseline was repeated six times. In their narrative review,
Zuckerman et al. also noted the same pattern. However, these
researchers did not provide cumulative estimates of this
relationship.

The meta-analysis presented here found that humans were not
very accurate in detecting deception. While this finding is in
contrast to the evaluation of deception detection by Zuckerman
et al., it is similar to that of Kraut (1980). This meta-
analysis shows that in general, Kraut's sole estimate of accuracy
is correct in its low rating of the human ability to detect
deception. However, this meta-analysis also suggests this rating
may differ across experimental conditions and that important
patterns in deception detection may be overlooked through the use
of a singular estimate of human ability. For example, the
facilitating impact of message content in accurately detecting

deception can not be retrieved from the Kraut estimate. Neither

Deception Detection - 71

can the importance of the body, as a source of cues to deception

be discovered in a single estimate.

Samplinngrror

 

The results yielded by this meta-analysis have some
uncertainty associated with them. This uncertainty is
attributable to the sampling error present in the estimates.
Actual accuracy rates of humans in detecting deception may differ
somewhat from these estimates. Variation in deception detection
ability across the experimental conditions may also be less than
the present findings suggest.

Without the availability of full summary statistics for the
studies included in this meta-analysis, the amount of sampling
error present in the meta-analytic results can not be assessed.
Meta-analysts have yet to determine how to assess sampling error
with limited summary statistics for the within-subjects research
designs.

While the amount of sampling error present can not be
assessed, there are some indications that it may be less
extensive in some of the estimates than in others, although the
amount of this difference can not be ascertained. For example,
those estimates with large cumulative sample sizes will have less
sampling error than those with smaller samples. In this meta-
analysis, the accuracy ratings for the observational conditions

and for the judgments of truth and deceit will have the least

Deception Detection - 72

sampling error. The accuracy ratings with the most sampling
error will be those yielded by the singular studies in the
narrative review, and by cumulative ratings based on small sample
sizes.

The fact that most of the studies comprising this meta-
analysis have come from within-subjects designs suggests that
the amount of sampling error present in these accuracy ratings is
less than what it would have been had these studies come from
factorial designs (Hunter, 1984). Sampling error is reduced in
the repeated measures design, because individual differences in
observer responses are not a source of error as they are in
factorial designs. Instead these individual differences become a
measure of the effect of the independent variables in the within-

subjects design.

Conclusion

This meta-analysis has implications for both researchers
engaged in the study of human deception detection, and in the
development of improved research methodology.

For the deception researcher, this meta-analysis may provide
some guidelines for future deception research. The most
powerful guideline should come from the observational
conditions. These conditions yielded the most distinct pattern

in accuracy ratings and, with their large sample sizes, they

Deception Detection - 73

should contain the least amount of sampling error. Perhaps
future deception research can further explore the viability of
message content in facilitating increased deception detection
accuracy. Second, the guessing bias found in judgments of truth
and deceit should caution researchers to counterbalance truthful
and deceptive messages. Finally, the low accuracy ratings
evident in this meta-analysis may direct investigation into
understanding how humans might best cOpe with poor lie detection
skills.

For the research methodologist, this meta-analysis
illustrates the need for methods of extracting effect sizes from
within-subjects designs when complete summary statistics are
unavailable. While an initial step in extracting effect sizes
from within-subjects designs, this study provides only a partial
solution to the problem. Techniques need to be developed for
extracting effect sizes from studies that use a variety of
measurement schemes, and which may provide differing amounts of
summary information. Development of apprOpriate estimates for

sampling error should also be considered.

References

Atmiyanandana, V. (1976). An experimental study of the detection
in cross-cultural communication. Doctoral dissertation,
Florida State University. (University Microfilms No.76-29,416)

Backster, C. (1963). Total chart minutes concept. Law and Order,

 

11, 77-79.

Barland, G., S Raskin, D.C. (1976). Validity and reliability of
polygraph examinations of criminal suspects (Report NO. 76-1,
Contract 75-NI-99-0001). Washington, DC: U.S. Department of
Justice.

Bauchner, J.E., Brandt, D.R., S Miller, G.R. (1977). The truth/
deception attribution: Effects of varying levels of

information availability. In B.D. Rubin (Ed) Communication

 

Yearbook I, New Brunswick, New Jersy:Transaction Books.
Ben-Shakhar, c., Lieblich, 1., & Bar-Hillel, M. (1982). An
evaluation of polygrapher's judgments: A review from a

decision theoretic perspective. Journal of Applied

 

Psychology, §1(6), 701-713.

Bersh, P.J. (1969). A validation study of examiner judgment.
Journal of Applied Psychology, 53, 399-403.

Benusi, V. (1975). The response systems of lying. Polygraph, 4,

52-76. (Original work published 1914).

74

Deception Detection - 75

Brandt, D.R., Miller, G.R., S Hocking, J.E. (1980a). The truth-
deception attribution: Effects of familiarity on the ability

of observers to detect deception. Human Communication

 

Research, 6(2), 99-110.
Brandt, D.R., Miller, G.R., S Hocking, J.E. (1980b). Effects of
self-monitoring and familiarity on deception detection.

Communication Quarterl , 38(3), 3-10.

 

Brandt, D.R., Miller, G.R., S Hocking, J.E. (1982). Familiarity

and lie detection: A replication and extension. The Western

 

Journal of Speech Communication, 4g, 276-290.

 

Buck, R., Miller, R.E., S Caul, W.F. (1974). Sex, personality,
and physiological variables in the communication of affect via
facial expression. Journal of Personality and Social
Psychology, 39(4), 587-596.

Bugental, D.E., Kaswan, J.W., S Love, L.R. (1970). Perception of
contradictory meanings conveyed by verbal and nonverbal

channels. Journal of Personality and Social Psycholggy,

 

1_6(4), 647-655.

Burgoon, J. S Saine, T. (1978). The unspoken dialog: An intro-

 

duction to nonverbal communication. Boston: Houghton Miffin.

 

Charlesworth, W.W., S Krevtzer, M.A. (1973). Facial expressions

of infants and children. In P. Ekman (Ed), Darwin and Facial

 

Expression: A Century of Research in Review. New York:

 

Academic Press.

Deception Detection - 76

Cohen, J. (1977). Statistical power analysis for the behavioral
sciences (Rev. ed.) New York: Academic Press.

Comadena, M.E. (1981). Examinations of the deception attribution
process of friends and intimates. Unpublished Doctoral Dis-
sertation, Purdue University, Indiana.

Comadena, M.T. (1982). Accuracy in detecting deception: Intimate
and friendship relationships. In M. Burgoon (Ed.),

Communication Yearbook 6. Beverly Hills: Sage.

 

DePaulo, B.M. (1981). Success at detecting deception: Liability

or skill? Annals New York Academy of Sciences, 245-255.

 

DePaulo, B.M., Davis, T., S Lanier, K. (1980, April). Planning
lies: The effects of spontaneity and arousal on success at
deception. Paper presented at the Eastern Psychological
Association, Hartford, Conn.

DePaulo, B.M., S Jordan, A. (1982). Age changes in deceiving and
detecting deceit. In R.S. Feldman (Ed.), DevelOpment of
Nonverbal Behavior in Children, New York: Springer-Verlag.

DePaulo, B.M., Jordan, A., Irvine, A., S Laser P.S. (1982). Age
changes in the detection of deception. Child Development, 53,
701-709.

DePaulo, B.M., Lanier, K., S Davis, T. (1983). Detecting the
deceit of the motivated liar. Journal of Personality and

Social Psychology, 42(5), 1096-1103.

 

Deception Detection - 77

DePaulo, B.M., Lassiter, G.D., & Stone, J.I. (1982). Attentional

determinants of success at detecting deception and truth.

Personalityyand Social ngchology Bulletin, 8(2), 273-279.

 

DePaulo, B.M., Lassiter, G.D. S Stone, J.I. (1982). Attentional
determinants of success at detecting deception and truth.
Personality and Social Psychology Bulletin, 8,273-279.

DePaulo, B.M., S Rosenthal, R. Telling lies. (1979). Journal of

Personalityyand Social Psychology, 11(10), 1713-1722.

 

DePaulo, B.M., S Rosenthal, R. (1979). Ambivalence, discrepancy,
and deception in nonverbal communication. In R. Rosenthal
(Ed.) Skill in Nonverbal Communication, Cambridge, Mass:
Oelgeschlager.

DePaulo, B.M., Rosenthal, R., Green, C.R., S Rosenkrantz, J. (in
press). Diagnosing deceptive and mixed messages from verbal
and nonverbal cues. Journal of Experimental Social Psychology.

DePaulo, B.M., Rosenthal, R., Rosenkrantz, J., S Green C.R.
Actual and perceived cues to deception: A closer look at
speech. (Unpublished Manuscript)

DePaulo, B.M., Stone, J.I., S Lassiter, G.D. (1984). Telling
ipggatiating_lies: Effects of target sex and targsp
attractiveness on verbal and nonverbal deception success.

Manuscript submitted for publication.

Deception Detection - 78

DePaulo, B. M., Stone, J. I. S Lassiter, G. D. (in press).
Deceiving and detecting deceit. In B. R. Schlenker (Ed.)

The Self and Social Life. New York: McGraw-Hall.

 

Dollinger, S.J., Reader, M.J., Marnett, J.P., S Tylenda, B.
(1983). Psychological-mindedness, psychological-construing,
and the judgment of deception. The Journal of General
Psychology, 198, 183-191.

Edelman, R.I. (1970). Some variables affecting suspicion.

Journal of Personalityyand Social Psychology, 12(4), 333-337.

 

Ekman, P., S Friesen, W.V. (1969). Non-verbal leakage and clues

to deception, Psychiatry, 32, 88-106.

 

Ekman, P., S Friesen, W.V. (1974). Detecting deception from the
body or face. Journal of Personality and Social Psychology,
32(3), 288-298.

Exline, R.V., Thibaut, J., Hickey, 0.8., S Gumpert, P. (1970).
Visual interaction in relation to machiavellianism and an

unethical act. In R. Christie and F.L. Geis (Eds). Studies in

 

Machiavellianism, New York: Academic Press.

 

Fay, P.J., S Middleton, W.C. (1941). The ability to judge truth-
telling, or lying, from the voice as transmitted over a public

address system. The Journal of General Psychology, 2&, 211-

 

215.

Deception Detection - 79

Feldman, M., S Thayer, S. (1980). A comparison of three measures

of nonverbal decoding ability. The Journal of Social

 

Psychology, 111, 91-97.

Feldman, R.S. (1976). Nonverbal disclosure of teacher deception

and interpersonal affect. Journal of Educational Psychology,

 

§§(6), 807-816.
Feldman, R.S. (1979). Nonverbal disclosure of deception in urban

Koreans. Journal of Cross-Cultural Psychology, 19(1), 73-83.

 

Feldman, R.S., Jenkins, L., S Popoola, O. (1979). Detection of
deception in adults and children via facial expressions.

Child Development, 59, 350-355.

 

Feldman, R.S., S White, J.B. (1980). Detecting deception in
children. Journal of Communication, 99(2), 121-128.

Fugita, S.S., Hogrebe, M.C., S Wexley, K.N. (1980). Perceptions
of deception: Perceived expertise in detecting deception,

successfulness of deception and nonverbal cues. Personality

 

and Social Psychology Bulletin, 9(4), 637-643.

 

Geis, F.L., S Moon, T.H. (1981). Machiavellianism and deception.

Journal of Personality and Social Psychology, 91(4), 766-775.

 

Geizer, R.S., Rarick, D.L., S Soldow, G.F. (1977). Deception and

judgment accuracy: A study in person perception. Personality

 

and Social Psychology Bulletin, 9, 446-449.

 

Glass, G.V., McGaw, B., S Smith, M.L. (1981). Meta-Analysis in

 

Social Research, Beverly Hills: Sage.

 

Deception Detection - 80

Hall, J. A. (1978). Gender effects in decoding nonverbal cues.

Psychological Bulletin, 92, 845-857.

 

Hall, J. A. (1980). Gender differences in nonverbal communication
skills. In R. Rosenthal (Ed.) Quantitative Assessment of

Research Domains. San Fransciso, California: Jossey-Bass.

 

Harrison, A.A., Hwalek, M., Raney, D.F., S Fritz, J.G. (1978).
Cues to deception in an interview situation. 925151
Psychology, 91(2), 156-161.

Hemsley, G.D. (1977). Experimental studies in the behavioral

indicants of deception. Unpublished Doctoral Dissertation,

 

University of Toronto.
Hemsley, G.D., S Doob, A.N. (1979). The detection of deception

from nonverbal behaviors. Paper presented at the meeting of

 

the Canadian Psychological Association, Quebec City, Quebec,
Canada.

Hemsley, G.D., S Doob, A.N. (1978). The effect of looking
behavior on perceptions of a communicator's credibility.
Journal Of Applied Social Psychology, 9(2), 136-144.

Hildreth, R.A. (1953). An experimental study of audiences'
ability to distinguish between sincere and insincere speeches.
Unpublished doctoral dissertation, University of Southern

California.

Deception Detection - 81

Hocking, J.E. (1976). Detecting deceptive communication from

verbal, visual and paralinguistic cues: An exploratory

 

experiment. Unpublished Doctoral Dissertation, Michigan State

 

University.
Hocking, J.E., Bauchner, J., Kaminski, E.P., S Miller, C.R.
(1979). Detecting deceptive communication from verbal,

visual, and paralinguistic cues. Human Communication

 

Research, 9(1), 33-46.

Horvath, F.S. (1977). The effects of selected variables on the
interpretation of polygraph records. Journal of Applied
Psychology, 91, 127-136.

Horvath, F.S., S Reid, J.E. (1971). The reliability of polygraph

examiner diagnosis of truth and deception. Journal of

 

Criminal Law, Criminology and Police Science, 99, 276-281.

 

Hunter, F.L., S Ash, P. (1973). The accuracy and consistency of

polygraph examiners' diagnoses. Journal Of Police Science and

 

Administration, 1, 370-375.

 

Hunter, J. E. (1982, January). A new desigp for psychologgcal

statistics. Unpublished manuscript, Michigan State University.

 

Hunter, J.E., Schmidt, F.L., S Jackson, G.B. (1982). Meta-

Analysis: Cumulating_research findings across studies. Beverly

 

Hills: Sage.

Hunter, J.E., (1984, March). Personal communication.

Deception Detection - 82

Jones, H.E. (1960). The longitudinal method in the study of

personality. In I. Iscoe S H. W. Stevenson (Eds.), Personality

 

development in Children. Chicago, 111.: University of Chicago
Press.

Kepple, G. (1982). Des1gn and analysis: A researchers handbook

 

(2nd ed.). Englewood Cliffs, NJ: Prentice-Hall.

Kleinmuntz, B., S Szucko, J.J. (1984). Lie detection in ancient
and modern times: A call for contemporary scientific study.
American Psychologigt, 39(7), 766-776.

Kleinmuntz, B., S Szucko, J.J. (1982a) Is the lie detector valid?

Criminal Defense, 9, 13-15.

 

Kleinmuntz, B., S Szucko, J.J. (1982b). On the fallibility of

detection. Law and Society, 11, 85-104.

 

Kleinmuntz, B., S Szucko, J.J. (1984). A field study of the

fallibility of polygraphic lie detection. Nature, 308, 449-

 

450.

Knapp, M.L., Hart, R. P., S Dennis, H. S. (1974). An exploration
of deception as a communication construct. Human Communication
Research, 1(1), 15-29.

Kraut, R.E. (1978). Verbal and nonverbal cues in the perception
of lying. Journal of Personality and Social Psychology,
99(4), 380-391.

Kraut, R. (1980). Humans as lie detectors: Some second thoughts.

Journal of Communication, 99, 209-216.

 

Deception Detection - 83

Kraut, R., S Poe, D. (1980). Behavioral roots of person per-
ception: The deception judgments of customs inspectors and
laymen. Journal Of Personaligy_and Social Psychology, 99991,
784-798.

Kraut, R., S Lewis, S.H. (1984). Some functions of feedback in
conversation. In H. Applegate and J. Sypher (Eds).

Understanding Interpersonal Communication. Sage.

 

Larson, J.A. (1932) Lying and its detection. Chicago: University
of Chicago Press.

Lavrakas, P.J. (1977). Human differences in the ability to
differentiate spoken lies fromggpoken truths. Unpublished
Doctoral Dissertation, Loyola University of Chicago.

Lavrakas, P.J., S Maier, R.A. (1979). Differences in human

ability to judge veracity from audio medium. Journal of

 

Research in Personality, 19, 139-153.

 

Levine, J.M., Romashko, T., S Fleishman, E.A. (1973). Evaluation
of an abilities classification system for integration and
generalizing research findings: An application to vigilance
tasks. Journal of Applied Psychology, 99, 149-157.

Littlepage, G., S Pineault, T. (1978). Verbal, facial, and
paralinguistic cues to the detection of truth and lying.

Personality and Social Psychology Bulletin, 9(3), 461-464.

 

Deception Detection - 84

Littlepage, G.E., S Pineault, M.A. (1979). Detection of deceptive

factual statements from the body and the face. Personality and

 

Social Psychology Bulletin, 9(3), 325-328.

Littlepage, G.E., S Pineault, M.A. (1981). Detection of truthful
and deceptive interpersonal communications across information
transmission modes. The Journal of Social Psycho1ogy,u119,
57-68.

Littlepage, G.E., S Pineault, M.A. (1982). Detection of decep-
tion of planned and spontaneous communications. Unpublished
manuscript, Middle Tennessee State University.

Littlepage, G.E., McKinnie, R., S Pineault, M.A. (1983). Rela-
tionship between nonverbal sensitivities and detection of
deception. Perceptual and Motor Skills, 91, 651-657.

Lykken, D.T. (1978). The psychopath and the lie detector.
Psychophysiology, 19(2), 137-142.

Lykken, D.T. (1979). The detection of deception. Psychological

 

Bulletin, 99(1), 47-53.

Lykken, D.T. (1981). The lie detector and the law. Criminal
Defense, 9, 19-27.

Maier, R.A., S Lavrakas, P.J. (1976). Lying behavior and

evaluation of lies. Perceptual and Motor Skills, 91, 575-581.

 

Maier, N.R.F. (1966). Sensitivity to attempts at deception in an

interview situation. Personnel Psychology, 19, 55-66.

 

Deception Detection - 85

Maier, N.R.F., S Janzen, J.G. (1967). Reliability of reasons used

in making judgments of honesty and dishonesty. Perceptual and

 

Motor Skills, 19, 141-151.

 

Maier, N.R.F., S Thurber, J.A. (1968). Accuracy Of judgments of
deception when an interview is watched, heard, and read.

Personnel Psychology, 11, 23-30.

 

Marston, W.M. (1917). Systolic blood pressure changes in

deception. Journal of Experimental Physics, 1, 117-163.

 

Matarazzo, J.D., Wiens, A.N., Jackson, R.E., S Manaugh, T.S.
(1970). Interviewee speech behavior under conditions of endo-
genously-present and exogenously-induced motivational states.

Journal of Clinical Psychology, 19, 141-148.

 

Mehrabian, A. (1971). Nonverbal betrayal of feeling. Journal of

 

Experimental Research in Personality, 9, 64-73.

 

Mehrabian, A., S Wiener, M. (1967). Decoding of inconsistent
communications. Journal of Personality and Social Psychology,
9(1), 109-114.

Miller, G.R., deTurck, M.A., S Kalbfleisch, P.J. (1983). Self-
monitoring, rehearsal, and deceptive communication. 99999

Communication Research, 19(1), 97-117.

 

Miller, G.R., S Kalbfleisch, P.J. (1982, May). Effect of self-

 

monitoring and Opportunity to rehearse on deceptive success.
Paper presented at the convention of the International

Communication Association, Boston, Mass.

Deception Detection - 86

Mongeau, P. (1984, March). Personal Communication.
Motley, M.T. (1974). Acoustic correlates of lies. Western

Speech, Spr, 81-87.

 

Olson, C.T. (1978). The effect ofyperceived conditions Of

 

interpersonal observation on encoding and decodipgyprocesses

 

duriogydeceitful self:presentations. Unpublished Doctoral

 

Dissertation, Columbia University.
Parker, R.J. (1978). ége,isex, and the ability‘to detect decep-

tion through nonvocal cues. Unpublished Doctoral Dissertation,

 

Fresno Campus, California School of Professional Psychology.

Podlesny, J.A., S Raskin, D.C. (1977). PsychOphysiological
measures and the detection Of deception. Psychological
Bulletin, 99, 782-799.

Potamkin, G.G. (1982). Heron addicts and nonaddicts: The use and

 

detection of nonverbal deception clues. Unpublished doctoral

 

dissertation, California School of Professional Psychology.
Raskin, D.C. (1978). Scientific assessment of the accuracy of

detection of deception: A reply to Lykken. Psychophysiology,

 

g2), 143-147.
Raskin, D.C., S Hare, R.D. (1978). Psychopathy and detection of

deception in a prison pOpulation. Psychophysiology, 19(2),

 

Raskin, D.C., S Podlesny, J.A. (1979). Truth and deception: A

reply to Lykken. Psycho1ogical Bulletin, 99(1), 54-59.

 

Deception Detection - 87

Raskin, D.C., S Podlesny, J.A. (1978). Effectiveness of techni-
ques and physiological measures in the detection of deception.
Psychophysiology, 19(4), 344-359.

Reid, J.E., S Inbau, F.E. (1977). Truth and deception: The poly-

graph ("lie detection") technique (2nd ed.). Baltimore:

 

Williams S Wilkins.
Riggio, R.E., S Friedman, H.S. (1983). Individual differences and

cues to deception. Journal of Personality and Social

 

Psychology, 99(4), 899-915.
Rosenthal, R. (1978). Combining results of independent studies.
Psychological Bulletin, 99, 185-193.

Rotkin, H.G. (1980). Information used in detecting deception.

 

Unpublished doctoral dissertation, New York University.

Rovira, M.L. (1982). Detection of deception: A signal detection

 

theory analysis. Unpublished doctoral dissertation, The

 

Catholic University of America.

Sakai, D.J. (1981). Nonverbal communication in the detection of

 

deception amopg women and men. Unpublished doctoral dis-

 

sertation, University Of California, Davis.

Sereno, T.J.P. (1981). Children's honesty revisited: An explora-

 

tion of deceptive communication in preschoolers. Unpublished

 

doctoral dissertation, Bowling Green State University.

Deception Detection - 88

Slowik, S.M., S Buckley, J.P. (1975). Relative accuracy of poly-
graph examiner diagnoses from respiration, blood pressure, and
GSR recordings. Journal of Police Science and Administration,
9, 300-309.

Stiff, J.B., S Miller, G.R. (1984). Interrogation,ylevel of
message exposure, and yjudgpents of honest and deceit: Toward
a more interactive model of communication. Paper presented
at the Annual Western Speech Communication Association
meeting, Seattle.

Streeter, L.A., Krauss, R.M., Geller, V., Olson, C., S Apple, W.
(1977). Pitch changes during attempted deception. Journal of
Personality and Social Psycho1pgy,‘99(5), 345-350.

Summers, W.G. (1939). Science can get the confession. Fordham

Law Review, 9, 355-354.

 

Snyder. M. (1974). Self-monitoring of expressive behavior.
Journal of Personality and Social Psychology, 99, 526-537.

Szucko, J.J., S Kleinmuntz, B. (1981). Statistical versus
clinical lie detection. American Psychologist, 99, 488-496.

Trovillo, P.V. (1939a). A history of lie detection. The Journal
of Criminal Law and Criminology, 99(6), 848-875.

Trovillo, P.V. (1939b). A history of lie detection. The Journal

of Criminal Law and Criminology, 99(1), 104-119.

Deception Detection - 89

Wicklander, D., S Hunter, F. (1975). The influence of auxiliary

sources of information in polygraph diagnosis. Journal of

 

Police Science and Administration, 9, 405-409.

 

Wilson, S.J. (1975). Channel differences in the detection of

 

deception. Unpublished doctoral dissertation, Florida State
University. (University Microfilms No. 76-2726)
Yohman, J.R. (1978). The guilty knowledge technique in lie

detection. Biological Psychology Bulletin, 9(3), 96-103.

 

Zuckerman, M., Amidon, M.D., Bishop, S.E., S Pomrantz, S.D.
(1982). Face and tone of voice in the communication of

deception. Journal of Personality and Social Psychology,

 

33(2), 347-357 .
Zuckerman, M., DePaulo, B.M., S Rosenthal, R. (1981). Verbal and

nonverbal communication of deception. Advances in Experimental

 

Social Psychology, 19, 1-49.

 

Zuckerman, M., Kernis, M.R., Driver, R., S Koestner, R. (1984).
Segmentation of behavior: Effects of actual deception and

expected deception. Journal Of Personality and Social

 

Psychology, 99, 1173-1182.
Zuckerman, M., Koestner, R., S Alton, A. O. (1984). Learning to

detect deception. Journal of Personality and Social

 

Psychology, 99(5), 345-350.

Deception Detection - 90

Zuckerman, M., Koestner, R., S Colella, M. J. (in press) Learning
to detect deception from three communication channels.

Journal of Nonverbal Behavior.

 

Deception Detection - 91

Footnotes

1The studies reviewed here required persons to Observe
communicators both lying and telling the truth at equal
frequencies.

2Atmiyanandana, 1976; Bauchner, Brandt, S Miller, 1977;
Brandt, Miller, S Hocking, 1980a; Brandt, Miller, S Hocking,
1980b; Brandt, Miller, S Hocking, 1982; Ekman S Friesen, 1974;
Fay S Middleton, 1941; Harrison, Hwalek, Raney, S Fritz, 1978;
Hemsley, 1977; Hemsley S Doob, 1979; Hocking, Bauchner, Kaminski,
S Miller, 1979; Lavrakas, 1977; Littlepage S Pineault, 1979;
Littlepage S Pineault, 1981; Littlepage S Pineault, 1982;
Littlepage, McKinnie, S Pineault, 1983; Maier S Janzen, 1965;
Maier S Lavrakas, 1976; Maier S Thurber, 1968; Matarazzo, Wiens,
Jackson, S Manaugh, 1970; Miller, deTurck, S Kalbfleisch, 1983;
Motley, 1974; Parker, 1978; Potamkin, 1982; Rovira, 1988; Sakai,
1981; Sereno, 1981; Stiff S Miller, 1984; Wilson, 1975;
Zuckerman, Koestner, S Alton, 1984; Zuckerman, Kernis, Driver, S
Koestner, 1984; Zuckerman, Koestner, S Colella, in Press.

3Atmiyanandana, 1976; Bauchner, Brandt, S Miller, 1977;
Brandt, Miller, S Hocking, 1980a; Brandt, Miller, S Hocking,
1980b; Brandt, Miller, S Hocking, 1982; Comadena ,1981; DePaulo,
Davis, S Lanier, 1980; DePaulo S Jordan, 1982; DePaulo, Jordan,
Irvine, S Laser, 1982; DePaulo, Lanier, S Davis, Submitted;

DePaulo, Lassiter, S Stone, 1982; DePaulo S Rosenthal, 1979;

 

Deception Detection - 92

DePaulo, Rosenthal, Green, S Rosenkrantz, in Press;DePaulO,
Rosenthal, Rosenkrantz, S Green, Submitted; DePaulo, Stone, S
Lassiter, 1984; Dollinger, Reader, Marnett, S Tylenda, 1983;
Ekman S Friesen, 1969; Ekman S Friesen, 1974; Exline, Thibaut,
Hickey, S Gumpert, 1970; Pay S Middleton, 1941; Feldman, 1976;
Feldman, 1979; Feldman, Jenkins, S Popoola, 1979; Feldman S
White, 1980; Fugita, Hogrebe, S Wexley, 1980; Geis S Moon, 1981;
Geizer, Rarick, S Soldow, 1977;Harrison, Hwalek, Raney, S Fritz,
1978; Hemsley, 1977; Hemsley S Doob, 1978; Hemsley S Doob, 1979;
Hildreth, 1953; Hocking, Bauchner, Kaminski, S Miller, 1979;
Kraut, 1978; Kraut S Lewis, 1984; Kraut S Poe, 1980; Lavrakas,
1977; Littlepage S Pineault, 1978; Littlepage S Pineault, 1979;
Littlepage S Pineault, 1981; Littlepage S Pineault, 1982;
Littlepage, McKinnie, S Pineault, 1983; Maier, 1966; Maier S
Janzen, 1965; Maier S Lavrakas, 1976; Maier S Thurber, 1968;
Matarazzo, Wiens, Jackson, S Manaugh, 1970; Miller, deTurck, S
Kalbfleisch, 1983; Motley, 1974; Olson, 1978; Parker, 1978;
Pollman, 1982; Potamkin, 1982; Rotkin, 1980; Rovira, 1988; Sakai,
1981; Sereno, 1981; Stiff S Miller, 1984; Streeter, Krauss,
Geller, Olson, S Apple, 1977; Wilson, 1975; Zuckerman, Amidon,
BishOp, S Pomrantz, 1982; Zuckerman, Koestner, S Alton, 1984;
Zuckerman, Kernis, Driver, S Koestner, 1984; Zuckerman, Koestner,

S Colella, in Press.

Deception Detection - 93

4Mean accuracy ratings for study 11 are the combined ratings
from both factual and emotional lies/truths. Studies 6, l4, and
2 do not have audio.

5Mean accuracy ratings for study 11 are the combined ratings
from both factual and emotional lies/truths.

6Mean accuracy ratings for study 11 are the combined ratings
from both factual and emotional lies/truths. The combined head
and body with full audio observation condition for study 2 represents
accuracy estimates made by subjects viewing and listening to

stimulus subjects through a two way mirror.

1mm”:nun:mum:mummrmwuwmuml

 

106824 1 82