You are on page 1of 11

Journal of Personality and Social Psychology Copyright 1997 by the American Psychological Association, Inc.

1997, Vol. 72, No. 6, 1429-1439 0022-3514/97/$3.00


The Abi l i ty to Det ect Decei t Generalizes Across Different
Types of Hi gh-Stake Li es
Mark G. Frank
Rut ge r s - - The St at e Uni versi t y of New Jer sey
Paul Ekman
Uni versi t y o f Cal i forni a, San Fr anci sco
The authors investigated whether accuracy in identifying deception from demeanor in high-stake lies
is specific to those lies or generalizes to other high-stake lies. In Experiment 1, 48 observers judged
whether 2 different groups of men were telling lies about a mock theft (crime scenario) or about
their opinion (opinion scenario). The authors found that observers' accuracy in judging deception
in the crime scenario was positively correlated with their accuracy in judging deception in the opinion
scenario. Experiment 2 replicated the results of Experiment 1, as well as P. Ekman and M. O' Sullivan' s
( 1991 ) finding of a positive correlation between the ability to detect deceit and the ability to identify
micromomentary facial expressions of emotion. These results show that the ability to detect high-
stake lies generalizes across high-stake situations and is most likely due to the presence of emotional
clues that betray deception in high-stake lies.
Pr of essi onal l i e cat chers r ecogni ze that some l i e cat chers are
consi st ent l y bet t er j udges of i nt erpersonal decept i on t han others.
For exampl e, Aust ral i an cust oms agent s have not ed that t he same
gr oup of offi cers seems to t op t hei r " cont r aband r e c ove r e d"
ranki ngs each mont h (S. Van Der Kooy, Aust ral i an cust oms
officer, per sonal communi cat i on, June 18, 1993; see al so
Kr aut & Poe, 1980). Moreover, s ome Amer i can pol i ce or gani za-
t i ons ask t hei r consi st ent l y good l i e cat chers t o t rai n ot her inter-
r ogat or s (J. J. Newberry, Al cohol , Tobacco, and Fi r ear ms agent,
per sonal communi cat i on, Apr i l 1992). Even obser vat i ons of
pr of essi onal poker pl ayers have suggest ed that the best pl ayers
ar e char act er i zed by t hei r abi l i t y to r ecogni ze decept i on across
a var i et y of opponent s and si t uat i ons ( Hayano, 1980). Al t hough
this anecdot al evi dence concer ns det ect i on abi l i t i es wi t hi n a
gi ven occupat i on, it suggest s that a per s on' s abi l i t y t o det ect
decei t is not r andom or l i mi t ed to speci fi c peopl e or si t uat i ons
but may be a skill that general i zes across di f f er ent peopl e and
di fferent kinds o f lies.
These observat i ons, however, run cont r ar y to years of psycho-
l ogi cal r esear ch that has suggest ed that t he abi l i t y to det ect
decei t is not general but rat her si t uat i on or person speci fi c. For
exampl e, wi t hi n t he cont ext of a si ngl e decept i on si t uat i on,
This work was originally supported by National Institute of Mental
Health National Research Service Award MH09827 and later by a re-
search grant from the Australian Research Council and Research Scien-
tist Award, MH06092.
We would like to thank David Matsumoto and his emotion research
group at San Francisco State University for their assistance in conducting
Experiment 1 and Colette D' Abreo Read for FACS coding the stimulus
sample.
Correspondence concerning this article should be addressed to Mark
G. Frank, School of Communication, Information, and Library Studies,
Rutgers--The State University of New Jersey, 4 Huntington Street, New
Brunswick, New Jersey 08901-1071, or to Paul Ekman, Department of
Psychiatry, University of California, 401 Parnassus Avenue, San Fran-
cisco, California 94143. Electronic mail may be sent via the Internet to
mgfrank@scils.rutgers.edu or ekmansf@itsa.ucsf.edu.
st udi es have f ound no r el at i onshi p bet ween a l i e cat cher ' s abi l i t y
to j udge accur at el y the t rut hful ness of one st i mul us i ndi vi dual
and his or her abi l i t y t o j udge accur at el y that of a di fferent
st i mul us i ndi vi dual ( Kr aut , 1978, 1980). Moreover, r esear ch
has f ound no r el at i onshi p bet ween accur acy of j udgment and
gender of st i mul us i ndi vi dual ( DePaul o & Rosent hal , 1979b) or
bet ween accur acy of j udgment and cul t ure of st i mul us i ndi vi dual
(i . e. , same cul t ure as j udge vs. di fferent ; Bond, Omar, Mah-
moud, & Bonser, 1990). Taken together, t hese studies demon-
strated that the skill l evel s and charact eri st i cs of the liars s eemed
t o over r i de any pot ent i al i ndi vi dual di fferences in lie cat cher s'
skill. These researchers r easoned that i f there is no r el at i onshi p
bet ween accur acy of j udgment and l i ar charact eri st i cs wi t hi n a
gi ven t ype of lie, there shoul d be no r el at i onshi p bet ween accu-
r acy of j udgment and l i ar charact eri st i cs across di fferent t ypes o f
l i es ( DePaul o, 1994; DePaul o, Zuckerrnan, & Rosent hal , 1980).
However, this same r esear ch showed a st rong r el at i onshi p be-
t ween abi l i t y to det ect di fferent emot i ons and gender of st i mul us
i ndi vi dual ( DePaul o & Rosent hal , 1979b).
Thi s f ai l ur e t o find general i t y in the abi l i t y to det ect decei t
ext ends to t rai ni ng i ndi vi dual s to det ect decei t . Usi ng a par adi gm
in whi ch st i mul us i ndi vi dual s ei t her l i ed or t ol d the truth about
s omeone t hey l i ked and s omeone t hey di sl i ked, researchers
f ound that when j udges wer e pr esent ed wi t h i nf or mat i on about
decept i on pr i or to t he i t em or wer e gi ven f eedback on t hei r
per f or mance aft er seei ng an i t em, t hey i mpr oved t hei r abi l i t y to
det ect t he decept i ons of a gi ven deceiver, but t hey di d not i m-
pr ove t hei r abi l i t y to det ect the decept i ons of di fferent decei vers
( Zucker man, Koestner, & Al t on, 1984). However, usi ng a para-
di gm in whi ch st i mul us i ndi vi dual s wer e shown slides of l and-
scapes and burn vi ct i ms and wer e asked bot h to l i e and to t el l
the truth about how t hey felt, researchers found evi dence that
t rai ni ng i mpr oved decept i on det ect i on abi l i t y across di fferent
decei vers ( deTur ck & Miller, 1990). It seems that t he mai n
di f f er ence bet ween t hese exper i ment s is that the second was
mor e l i kel y to i nduce st rong emot i ons, such as di sgust and fear.
Si gns of t hese emot i ons are pr oduced and r ecogni zed across
1429
1430 FRANK AND EKMAN
cultures and situations and thus would provide context- or situa-
tion-independent evidence for deception (Ekman, 1989, 1992,
1994). However, this evidence for generality in detecting deceit
was limited in deTurck and Miller (1990), because the deceivers
in the study were selected on the basis of their self-monitoring
scores, and more important, there was no independent evidence
for the existence of perceivable emotion on the part of the
deceivers.
A different approach to this question of generality provides
another piece of evidence that is consistent with the situational
specificity conclusion. This approach examined the relation-
ships between stable individual difference measures, such as
self-monitoring (Snyder, 1974), and the ability to detect deceit.
The rationale guiding these studies was that if the ability to
detect deceit was correlated with these temporally and situation-
ally stable individual difference measures, then the ability to
detect deceit must also be temporally and situationally stable.
The actual results of these studies have been contradictory. Some
have reported positive relationships between accuracy at de-
tecting deceit and variables such as social participation, per-
ceived complexity of human nature, social anxiety, and self-
monitoring (DePaulo & Rosenthal, 1979a; Geizer, Rarick, &
Soldow, 1977), whereas others have reported no relationship
between accuracy at detecting deceit and self-monitoring, CPI
scores, or other personality variables (e.g., Kraut & Poe, 1980;
O' Sullivan, Ekman, Friesen, & Scherer, 1992). Overall, these
equivocal results suggest that the ability to detect deceit would
not generalize across situations or lies.
The disparity between the anecdotal observations suggesting
generality and the psychological findings suggesting specificity
might be accounted for by the differences in the structural fea-
tures of the deception situations used to arrive at their respective
conclusions (Frank, 1992). The anecdotal evidence is derived
from very high-stake, real-world deception situations--police
interrogations, customs inspections, and high-stake poker
games- - i n which liars and truth tellers have much to gain or
lose by being judged deceptive. In contrast, the psychological
findings are derived from mostly low-stake, real-world deception
situations--white lies, day-to-day polite lies (DePanlo et al.,
t 980) - - wher e the liars and truth tellers either have little to
gain or lose by being judged deceptive or have little fear of
being caught telling these lies because they are required by
polite society. According to Ekman (1985), this distinction be-
tween high and low stakes for successful or unsuccessful decep-
tion is critical, because the presence of high stakes is central to
a liar feeling strong emotions when lying. It is the presence of
these emotions, such as guilt, fear of being caught, and disgust,
that can betray the liar' s deception when they are leaked through
nonverbal behaviors such as facial expression (Ekman, Frie-
sen, & O'Sullivan, 1988) or voice tone (Ekman, Friesen, &
Scherer, 1976). Given the finding that emotions expressed in
the face are universally recognized (Ekman, 1989, 1992, 1994),
Ekman (1985) has further argued that the extent to which the
stakes elicit emotions that provide clues to deceit in the expres-
sive behavior of the liar, a lie detector who attends to these
behavioral clues would not need to know the specifics of the
situation being evaluated in order to accurately judge deception.
On the basis of this reasoning, he predicted that the ability to
detect deceit would generalize across different types of high-
stake lies.
To date, no experiment has tested this idea directly by showing
the same group of observers two different samples of liars culled
from two different high-stake deception situations. 1 However,
two studies have provided indirect evidence consistent with Ek-
man' s (1985) generality hypothesis. In the first study, under-
graduate judges were shown videotapes of both high- and low-
motivated, and hence high- and low-aroused, stimulus partici-
pants who lied and told the truth about their opinions, attitudes,
and feelings on a variety of topics (DePaulo, Lanier, & Davis,
1983). The experimenters motivated the stimulus participants
by telling them either (a) that their success in deceiving was
related to career success and that their behavior would be care-
fully scrutinized (high motivation) or (b) that their deceptions
were simply a little game and that their behavior would not be
scrutinized (low motivation). The results showed that the high-
motivation stimulus participants were more accurately judged
from their nonverbal behavior, whereas the low-motivation par-
ticipants were more accurately judged from their verbal behav-
ior. However, there was no independent measure of expressed
emotion in this study, and a manipulation check showed that
the high- and low-motivation participants did not differ on their
self-reports of tension while responding.
In the second study, professional lie catchers such as agents
of the Secret Service, federal polygraphers, judges, and psychia-
trists, as well as students, were shown videotapes of highly
motivated nursing students who were attempting to convince an
interviewer that the films they were watching made them feel
pleasant, when one of the films was pleasant and the other one
featured gory scenes of an amputation and scenes of burn vic-
tims (Ekman & O'Sullivan, 1991 ). The experimenters motivated
the nurses by telling them that their success in convincing an
interviewer that the gory films they were watching were actually
pleasant would be related to their success in their nursing careers
(see Ekman & Friesen, 1974, for more details). In this high-
stake, emotionally arousing deception situation, not only could
many of the observers detect deceit significantly better than
chance but those observers who were most accurate reported
using a deception detection strategy that was based on examin-
ing the nonverbal behavior of the stimulus participants, whereas
those observers who were least accurate reported using a decep-
tion detection strategy that was based on examining only the
verbal behavior of the stimulus participants. The highly accurate
observers' strategy was successful in this deception situation
because they judged individuals who--because they faced very
high stakes for successful and unsuccessful deception--dis-
played facial and vocal signs of emotion when lying (Ekman &
O'Sullivan, 1991; Ekrnan, O'Sullivan, Friesen, & Scherer,
1991).
Taken together, these studies show that high-stake deceptions
are more likely to produce clues to deception in a person' s
Although one could argue that the faking positive and faking negative
lies used by DePaulo and Rosenthal (1979b) are different types of lies,
they are both derived from the same deception situation: describing
people. Our study featured two different situations: a lie about an action
one just took and a lie about opinions. However, we acknowledge that
this can be a slippery definitional issue.
DETECTING DECEIT 1431
nonver bal behavi or, whi ch, r esear ch has shown, is mor e l i kel y
t han verbal behavi or to cont ai n i nf or mat i on about the emot i onal
state of t he person ( Ekman & Fri esen, 1969; Zucker man, De-
Paul o, & Rosent hal , 1981 ). Moreover, t he Ekman and O' Sul l i -
van ( 1991 ) st udy s howed t hat pr of essi onal s wer e abl e t o j udge
decei t accur at el y in a hi gh- st ake decept i on si t uat i on t hat t hey
wer e not at all f ami l i ar with, whi ch is consi st ent wi t h Ekma n' s
( 1985) pr oposal that accur acy when det ect i ng di f f er ent ki nds
of l i es is not cont i ngent on knowl edge of t he det ai l s of a si t uat i on
i f t he stakes are suffi ci ent l y hi gh to ar ouse emot i ons.
However, none of t he af or ement i oned st udi es exami ned ob-
ser ver s' abi l i t i es to j udge decept i on across mor e t han one t ype
of lie, l et al one mor e t han one t ype of hi gh- st ake lie. Thus, we
pr opose t o make a par si moni ous t est of l i e cat cher s' abi l i t i es to
j udge decept i on across di f f er ent t ypes o f l i es by compar i ng t hei r
abi l i t i es to j udge accur at el y a sampl e o f liars in one hi gh- st ake
decept i on si t uat i on wi t h t hei r abi l i t y t o j udge accur at el y a di ffer-
ent sampl e of liars in a di f f er ent hi gh- st ake decept i on situation.
We pr edi ct ed that t he abi l i t y to det ect decei t woul d gener al i ze
across hi gh- st ake l i es such t hat a person who scor ed hi gh in
j udgi ng one hi gh- st ake l i e woul d scor e hi gh when j udgi ng a
di f f er ent t ype of hi gh- st ake lie.
E x p e r i me n t 1
Overview: Creating Ecologically Valid High-Stake
Scenarios
Ther e ar e a number of st ruct ural feat ures of decept i on situa-
t i ons that must be r ecr eat ed in t he l abor at or y to gener al i ze l abo-
rat ory resul t s to t he real wor l d (Frank, 1992; Podl esny &
Raski n, 1977). For exampl e, to r ecr eat e t he st ruct ural feat ures
of a cr i mi nal i nvest i gat i on in t he l aborat ory, one must have hi gh
s t a k e s - - t h a t is, the si t uat i on must cont ai n not onl y a r ewar d
f or successf ul decept i on but al so a puni shment f or unsuccessf ul
decept i on. The l i ar who is bel i eved obt ai ns f r eedom as wel l as
what ever i l l - got t en gai n he or she may have acqui r ed in the
commi s s i on o f t he cri me. The l i ar who is di sbel i eved faces
severe c o n s e q u e n c e s - - f o r exampl e, he or she can go to j ai l or
even be execut ed. In addi t i on, t hese stakes must appl y not onl y
to the l i ar but also, in a sl i ght l y di f f er ent way, to t he truth teller.
A t rut h t el l er who is bel i eved obt ai ns his or her f r eedom, but a
t rut h t el l er who is di sbel i eved faces the same consequences as
t he l i ar who is di sbel i eved. Newspaper account s of deat h r ow
i nmat es who are r el eased when the act ual ki l l er conf esses i l l us-
trate that in many r eal - l i f e si t uat i ons, not onl y ar e di sbel i eved
liars puni shed, but so are di sbel i eved t rut hful peopl e.
For this exper i ment , we cr eat ed t wo hi gh- st ake decept i on
si t uat i ons t hat we fel t woul d be ver y di fferent f r om each ot her
yet still cont ai n many of t he st ruct ural feat ures descr i bed above.
The first decept i on si t uat i on i nvol ved t he mock t heft of money,
and t he second i nvol ved f al si f yi ng one ' s opi ni on about a current
event issue. We adapt ed t he first scenar i o f r om a mock cr i me
par adi gm used in pol ygr aph r esear ch (e. g. , Hor owi t z, 1989).
In our ver si on of t he mock cr i me par adi gm ( what we r ef er to
as the crime scenario), part i ci pant s wer e t ol d that t hey and a
second pa r t i c i pa nt - - a c t ua l l y a c o n f e d e r a t e - - wo u l d have t he
oppor t uni t y to t ake $50 cash f r om i nsi de a bri efcase. The part i ci -
pant and t he conf eder at e woul d ent er t he r oom cont ai ni ng t he
br i ef case one at a time. The person who ent er ed first coul d
choose whet her to t ake t he money; t he person who ent ered sec-
ond woul d have t o t ake t he money i f it was there and, o f course,
coul d not t ake it i f it was not there. 2 We used this pr ocedur e
f or t wo reasons. First, we want ed to mi r r or what usual l y happens
in an act ual c r i me - - t h a t is, in the real wo r d , a cr i me al r eady
has been commi t t ed bef or e an i nvest i gat i on is begun. We want ed
t he part i ci pant to know that this " c r i me " had been commi t t ed
and that it was ei t her that part i ci pant or t he ot her i ndi vi dual who
had t aken the money. Second, our pi l ot part i ci pant s r epor t ed
that t hey wer e ver y suspi ci ous o f t he whol e pr ocedur e because
each knew t hat he was t he onl y pl ausi bl e suspect i f the money
was mi ssi ng. The part i ci pant s never saw the conf eder at e, but
t hey heard his voi ce over an i nt er com and heard hi m shuffl i ng
about t he r oom wi t h t he br i ef case when it was his t urn to enter.
No part i ci pant expr essed suspi ci on about t he t rue i dent i t y of
the confederat e. The si ze of the r ewar d was based on pi l ot
st udi es that suggest ed this amount o f money was suffi ci ent to
i nduce about hal f t he part i ci pant s t o choose t o l i e and suffi ci ent
to mot i vat e t he l i ar to want to succeed. These rewards r epr esent
t hose si t uat i ons in l i f e in whi ch there is mor e to be g a i n e d - -
al bei t in the short r u n - - f r o m l yi ng t han f r om bei ng truthful.
Al l part i ci pant s knew that aft er this cr i me had been commi t -
ted, t hey woul d be i nt er r ogat ed concer ni ng t he theft. Al l part i ci -
pant s wer e i nst ruct ed t o deny that t hey had taken t he money.
Part i ci pant s ( al l o f whom wer e men) wer e t ol d that i f t hey t ook
the money and wer e abl e to convi nce t he i nt er r ogat or that t hey
had not taken it, t hey coul d keep the $50. I f a part i ci pant t ook
t he money and t he i nt er r ogat or j udged hi m as l yi ng, that part i ci -
pant was t ol d that he woul d be puni shed by bei ng f or ced to
f or f ei t not onl y t he bonus money but al so his $10 per hour
par t i ci pat i on fee. Moreover, hal f of t he part i ci pant s wer e ran-
doml y assi gned to face an addi t i onal puni shment i f j udged l yi ng:
They wer e t ol d t hey woul d have to sit on a col d, met al chai r
i nsi de a cr amped, dar kened r oom l abel ed omi nousl y XXX, wher e
t hey woul d have to endure anywher e f r om 10 to 40 r andoml y
sequenced, 110-deci bel st art l i ng blasts o f whi t e noi se over t he
cour se of 1 hr. 3 These part i ci pant s wer e gi ven a sampl e of this
2 What this means is that half of the participants were allowed to
choose whether to take the money and half were assigned to take it or
not take it. This manipulation did not affect any of the analyses reported
in this article; for example, observers were no more accurate at judging
the participants who chose whether to take the money. However, one
reviewer raised the point that any relationship for accuracy across sce-
narios might he due to observers' being able to judge which participants
would choose to lie rather than due to observers' being able to judge
deception from behavioral clues. We were able to rule out this explana-
tion by finding a nonsignificant correlation between observers' accuracy
(in Experiments 1 and 2 combined) for judging those who could choose
in the crime scenario and for judging those who could choose in the
opinion scenario ( r = .14, p > .10, n = 78). Ironically, that was the
only nonsignificant correlation for accuracy across choice and deception
scenario. Therefore, the choice manipulation is not discussed further.
3 We did not have to threaten all participants with this punishment to
make the situation high stake; the gain or loss of $50 and the participation
fee seemed to he a high enough stake to produce the facial signs of
negative emotion. We take the advice of Levenson (1988), who argued
that the criterion for study of emotional phenomena should be the inde-
pendently confirmed presence of the emotion and not the attempts to
elicit that emotion. There was no difference in the presence of emotion,
1 4 3 2 FRANK AND EKMAN
puni s hme nt pr i or t o engagi ng i n t he t ask, but no par t i ci pant
act ual l y r ecei ved t he puni s hme nt af t er war d. Mos t pr i or i nvest i -
gat or s ( i nc l udi ng us ) ha ve not puni s he d t he l i ar who i s det ect ed
by t he i nt er r ogat or ( Me hr a bi a n, 1971, i s an except i on; he us ed
mi l d el ect r i c s hoc k as a puni s hme nt t o t he l i ar who was c a ught ) .
We f el t t hes e puni s hme nt s wer e appr opr i at e be c a us e i n mos t
r eal - l i f e si t uat i ons, not onl y i s a r ewar d r e duc e d or e l i mi na t e d
whe n s ome one i s c a ught l yi ng, but a puni s hme nt is a dmi ni s t e r e d
t o t he uns ucces s f ul liar. It i s t hi s puni s hme nt t hat i s cent r al t o
a l i ar ' s f ear of be i ng c a ught ( Ek ma n , 1985) .
We t ol d e a c h par t i ci pant t hat i f he t ol d t he t r ut h and t he
i nt er r ogat or j u d g e d h i m t o be t r ut hf ul , he woul d r ecei ve a $10
bonus on t op of hi s par t i ci pat i on fee. I f a par t i ci pant t ol d t he
t r ut h but t he i nt er r ogat or j u d g e d hi m as l yi ng, he woul d not get
t he $10 bonus or t he par t i ci pat i on f ee and woul d f ace t he Same
puni s hme nt as t he gui l t y pa r t i c i pa nt who ha d be e n j u d g e d as
l yi ng. The r eas on we di d t hi s i s t hat of t en i n hi gh- s t ake si t ua-
t i ons, bot h t r ut hf ul and decei t f ul peopl e evi dence fear: I n t he
t r ut hf ul per s on i t i s t he f ear of be i ng puni s he d i f mi s t akenl y
j u d g e d t o be l yi ng ( c a l l e d t he " Ot h e l l o e r r o r " by Ekman,
1985 ) , and i n t he decei t f ul per s on i t i s t he f ear of be i ng puni s he d
i f j udge d correct l y.
We adapt ed t he s econd de c e pt i on s cenar i o ( we r ef er t o t hi s
as t he opinion scenario) f r om t he f al s e opi ni on pa r a di gm (e. g. ,
Me hr a bi a n, 1971) . In our ver si on, par t i ci pant s wer e gi ven a
ques t i onnai r e t hat as s es s ed bot h t he di r ect i on and ma gni t ude of
t hei r opi ni ons on soci al i ssues (e. g. , " S h o u l d c onvi c t e d col d-
bl oode d mur der er s be e x e c u t e d ? " and " S h o u l d s moki ng be
b a n n e d f r om al l encl os ed publ i c p l a c e s ? " ) on a 1 (strongly
agree) t o 7 (definitely disagree) r at i ng scal e. The opi ni on t hat
t he par t i ci pant f el t mos t s t r ongl y a b o u t I a n d hence was mos t
cer t ai n a b o u t i wa s s el ect ed as t he one t o be de s c r i be d t r ut h-
f ul l y or falsely. A f al se des cr i pt i on me a nt t r yi ng t o c onvi nc e t he
i nt er r ogat or t hat t he par t i ci pant hel d an opi ni on on t hat i s s ue
exact l y oppos i t e t o t he pa r t i c i pa nt ' s t r ue opi ni on. The st akes f or
t hi s scenar i o, par t i ci pant s wer e t ol d, wer e i dent i cal t o t hose of
t he c r i me scenar i o: Tr ut h t el l er s who wer e bel i eved r ecei ved a
$10 bonus , l i ars who wer e bel i eved r ecei ved a $50 bonus , a nd
l i ars or t r ut h t el l er s di s bel i eved l ost al l mone y and s ome f a c e d
t he addi t i onal puni s hme nt . We a dmi ni s t e r e d t he opi ni on ques-
t i onnai r e t o par t i ci pant s bef or e t hey r ecei ved t he f ul l i ns t r uct i ons
f or ei t her of t hes e pa r a di gms t o ens ur e t hat t hey woul d not l i e
a bout t hei r opi ni ons whe n t hey c ompl e t e d t hi s ques t i onnai r e.
Al l par t i ci pant s wer e i ns t r uct ed t o c l a i m t hat t hey wer e hones t l y
de s c r i bi ng t hei r opi ni ons .
Me ~ o d
Participants. Participants were 20 men, age 18-28, from the S a n
Francisco area who volunteered for a study entitled "Communication
Skills Experiment." They were told that they would be paid $10 per
hour for participation, with the possibility o f gaining more money de-
pending on performance.
as assessed by the facial measurement, between those who were threat-
ened with the noise punishment and those who were not. There was
also no difference in detectability between these participants, so this
difference is not discussed further.
The observers who judged the participants were 32 female and 17 male
undergraduates from San Francisco State University who volunteered for
a study entitled " Can You Spot a Li ar?" They received course credit
for their participation.
Deception procedure. All participants engaged in the crime scenario
first. The participant was instructed to enter a room in the laboratory
where the money-filled briefcase was; hal f of the time the confederate
went first, and hal f of the time the participant went first. The participant
was instructed that the confederate would be interrogated before him
in the crime scenario and after hi m in the opinion scenario. Using this
procedure allowed us to inform the participants that the interrogator
knew the money was taken but did not know whether it was the partici-
pant or the confederate who took it.
After having the opportunity to take the money or not, the participant
returned to the original room. From there he was led by the experimenter
to the punishment room to await the arrival of the interrogator. The
participant stayed in the punishment room while the interrogator ostensi-
bly interrogated the confederate. After 6 - 7 rain, the experimenter led
the participant to the interrogation room. The interrogation room was
approximately 3 x 4 m and featured two chai r s - - one comfortable,
padded chair and one metal chair with no armrests, which was anchored
to the floor. Next to the metal chair was a stand-alone lamp, and at a 120
angle was the second, more comfortable chair, where the interrogator sat.
The experimenter led the participant to the metal chair and introduced
him to the interrogator. The interrogator was described as an expert in
judging deception. The interrogator did not know whether the participant
had taken the money or whether the participant was the first or second
to enter the room. The participant was then interrogated for approxi-
mately 5 - 7 min on a series of 20 scripted questions. After that, the
interrogator told the participant that he would examine his notes and
return in a few minutes for a second interrogation. The interrogator then
left the room.
At this point, the experimenter reentered and gave the participant the
instructions for the opinion scenario. These instructions stated that the
participant was to choose whether to lie about his opinion on the social
issue that was written on top of the instruction form. (Earlier, the experi-
menter had examined the participant' s responses to the opinion question-
naire and selected the topic that the participant indicated that he felt
most strongly about.) The participant was told that he would have only
2 min to prepare himself for the next interrogation. The instructions also
reminded the participant of the stakes for successful and unsuccessful
deception. After 2 min, the interrogator returned and asked the partici-
pant a series of 20 scripted questions parallel to those asked in the crime
interrogation. The interrogator reminded the participant that he would
be interrogating the "ot her participant" (the confederate) on his opinion
after the participant had finished. When the opinion interrogation was
complete, the participant was asked to fill out another questionnaire,
which included questions that probed his suspicions about the experi-
ment in general, how successful he felt he was at convincing the interro-
gator, and how much he feared the consequences of being judged decep-
tive. When the participant finished, he was told that the experiment was
over and that he had been judged either innocent or guilty on the crime
interrogation and judged either honest or deceptive on the opinion inter-
rogation. Participants who were judged to be honest were paid whatever
bonus was applicable, and participants who were judged to be deceptive
were told that they would not receive any bonus but would not have to
face any punishment. All participants, regardless of judgment, received
their $10 per hour participation fee and then were debriefed.
Stakes check. We designed bot h deception scenarios to have high
stakes, and as such they should have been able to elicit strong emotions
on the part of the participants. We assessed the presence of emotion by
applying the Facial Action Coding System (FACS; Ekman & Friesen,
1978) to score the participants' facial expressions during the interroga-
tions. FACS is a comprehensive system that measures all visible facial
DETECTING DECEIT 1 4 3 3
muscle movements and not j ust those presumed to be related to emotion.
FACS was chosen because it is nonobtrusive and has been used to verify
the physiological presence of emotion in a number of studies (e.g.,
Ekman, Davidson, & Friesen, 1990; Ekman et al., 1988; Levenson,
Ekman, & Friesen, 1990). In this high-stake situation, we expected that
the deceptive participants would feel strong negative emotions, such as
fear, disgust, and anger (e.g., fear of getting caught, Ekman, 1985;
disgust at oneself for lying; and anger at having one' s integrity chal-
lenged). We used a computerized dictionary that offered a priori predic-
tions about which configuration of FACS scores represented various
facial expressions of emotion; these predictions were based on particular
combinations of facial movements that have been judged by observers
across cultures to represent various emotions (see Ekrnan & Friesen,
1975, for more details about the facial configurations). We applied the
FACS coding to the participants after we had edited the stimulus materi-
als (see bel ow).
Creation of stimulus materials. Each participant' s mock crime and
opinion interrogations were videotaped from behind a hidden panel. The
participant knew beforehand that this interrogation was going to be
videotaped, but this hidden camera technique kept the camera from
distracting the participant. Overall, 12 out of 20 participants lied in the
crime scenario (whi ch meant that 7 out of the 10 participants who
entered the room first chose to take the money), and 9 out of 20 lied
in the opinion scenario. Each participant' s responses to the first six
questions of each interrogation were edited and assembled onto two
separate videotapes;4 one videotape consisted of the crime interrogations
for 10 participants, and the second videotape consisted of the opinion
interrogations for the remaining 10 participants. We assigned partici-
pants to each videotape based on three criteria. First, no participant
could appear on more than one tape; otherwise, observers' deception
accuracy across different types of lies would be confounded by their
accuracy for different stimulus participants. Second, each videotape had
to contain 5 men lying and 5 men telling the truth so that observers who
simply guessed would get only 50% correct. Third, the number of opin-
ions discussed on each videotape could be no more than three to reduce
confusion, and of these three, approximately even numbers of partici-
pants had to be lying and telling the truth about bot h the affirmative and
negative perspectives on each opinion (i.e., equal numbers lying and
telling the truth about supporting the death penalty, and equal numbers
lying and telling the truth about opposing the death penalty). There was
only one split of the 20 participants that satisfied all three criteria. Each
participant' s videotape excerpts were approximately 1 rain in length and
showed hi m in facial close-up with full audio. The interrogator also
could be he a r d- - but not seen- - aski ng his questions. The duration of
the crime videotape was approximately 12 min, and the duration of the
opinion videotape was approximately 14 min.
Judgment procedure. We showed the two deception detection video-
tapes to 49 observers, in counterbalanced order, in classroom settings
ranging from 7 to 11 observers per group. For the crime video, we told
observers that they would see 10 men who were being interrogated about
their involvement in a theft of money. For the opinion video, we told
observers that they would see 10 men who were being interrogated about
their opinion on a current event topic. We then asked observers to judge
whether each person they saw was lying or telling the truth by circling
the word truthful or the word lying on a response form after viewing
each participant' s segment. We encouraged observers to record their
judgments during the 10 s of silence that occurred between each segment.
We told the observers that anywhere from one fourth to three fourths
of the people they would see in each video were lying. We did this
because our experience has shown that there are always a few observers
in any group who circle truthful or lying every time without actually
closely observing the videotape because they think the test is a trick in
which all the people are actually truthful or all are deceptive.
We also asked observers to assess on the response form their confi-
dence in their ability to detect deception bot h before and after taking
the test. The pretest question was "How good do you think you are in
being able to tell i f another person is l yi ng?" The posttest question was
" I n this video, how well do you think you did in telling who was
l yi ng?" Both questions were rated on a 1- 5 scale, with 1 indicating
very poor (or very poorly) and 5 indicating very good (or very well).
Design. The independent variables were type of lie (crime or opin-
ion, a within-subject compari son), gender of observer, and order of
videotape presentation (between-subjects comparisons). The dependent
variable was the observer' s accuracy score for each videotape. Our
hypothesis was that observers who excelled at detecting deceit when
judging the crime video would also excel at detecting deceit when judg-
ing the opinion videotape. We made no predictions about the overall
accuracy level of the judgments. Given the past null findings on the
relationship between observer confidence and accuracy (e.g., DePaulo &
Pfeifer, 1986; Ekman & O' Sullivan, 1991 ), we predicted no relationship
between accuracy and self-rated ability or confidence in detecting deceit.
Re s ul t s
Stakes confirmation. A key a s s umpt i on ma de i n our par a-
di gms was t ha t t hes e hi gh- s t ake s cenar i os woul d be s ucces s f ul
i n pr oduc i ng e mot i on on t he pa r t of t he l i ars. A FACS- t r ai ned
s cor er s cor ed t he 10 i t ems f or e a c h t ape, and a s econd scor er
checked t he r el i abi l i t y f or appr oxi mat el y 20% of t he s ampl e
( t he scor er s a gr e e d on over 76% of t he s c or i ng) . We t hen r an
t hes e FACS scor es t hr ough a di ct i onar y cr eat ed by Ek ma n and
Fr i es en t o get a pr i or i pr edi ct i ons o f e mot i on f r om f aci al expr es -
si on. We f ound t ha t 80% of par t i ci pant s i n t he c r i me vi de o and
80% of t he par t i ci pant s i n t he opi ni on vi de o c oul d b e s ucces s -
f ul l y cl assi f i ed as l i ars or t r ut h t el l er s on t he bas i s of t he pr es ence
or a bs e nc e of f ear or di s gus t ( a n d i ncons i s t ent wi t h r ect a- anal yt i -
cal f i ndi ngs s uch as t hos e of Zuc ke r ma n et al., 1981, we di d
not fi nd t hat s mi l i ng di f f er ent i at ed t r ut h f r om decept i on; see
Fr a nk & Ekma n, 1997, f or mor e de t a i l s ) . Wh e n we c ons i de r e d
bot h vi deos, t he pr e s e nc e o f f ear or di s gus t accur at el y cat ego-
r i zed 90% o f t he l i ars, a nd t he a bs e nc e of t hese e mot i ons accu-
r at el y cat egor i zed 70% of t he t r ut ht el l er s. The per cent age of
accur at e cat egor i zat i ons , whi c h we der i ved sol el y f r om f aci al
expr es s i on, c ompa r e s qui t e f avor abl y wi t h t he 86% accur acy
r at e de mons t r a t e d by Ek ma n et al. ( 1 9 9 0 ) , who us e d bot h f aci al
e xpr e s s i on and voi c e pi t ch t o cl as s i f y t r ut h t el l er s and l i ars. Our
r esul t s cl ear l y s how t hat t he st i mul us par t i ci pant s wer e f eel i ng
s t r ong e mot i ons whe n l yi ng dur i ng bot h scenar i os.
Preliminary analyses. One obs er ver di d not f ol l ow i ns t r uc-
t i ons a nd was r e move d f r om t he anal ysi s. We der i ved e a c h ob-
s e r ve r ' s accur acy s cor e f or e a c h vi de o by count i ng t he n u mb e r
of cor r ect j udgme nt s he or she ma de out of t he 10. For t he sake
4 The six questions for the crime interrogation were as follows: ( a)
Describe exactly what happened, what you saw and did, when you were
in that room. ( b) Describe for me what your thoughts were when you
entered that room. ( c) Do you know how much money wa s - - o r was
supposed to b e - - i n the envelope? ( d) Did you take the money from
the envelope? ( e) Did you bring the money with you into this room?
and ( f ) Are you lying to me now7 The six questions for the opinion
interrogation were as follows: ( a) What is your position on this current
event issue? ( b) Why is it that you believe what you do on this issue?
( c) How long have you had this opinion? ( d) Is this your true opinion?
( e) You di dn' t j ust make up this opinion a few minutes ago? and ( f )
Are you lying to me now?
1434 FRANK AND EKMAN
of clarity, in the t ext and t abl es t hese numbers ar e conver t ed to
t he per cent age of cor r ect scores, but all anal yses wer e per f or med
on t he r aw numbers. We f ound no mai n effect s or i nt eract i ons
f or or der of present at i on of vi deot apes on accuracy, all Fs (1,
42) < 1.52, ns. Li kewi se, we found no mai n or i nt eract i on
effect s f or gender o f obser ver on accuracy, all Fs ( 1, 42) < 0. 96,
ns, so we col l apsed across t hese vari abl es.
Main analysis. An obser ver ' s accur acy scor e coul d range
f r om 0% to 100% cor r ect in i ncr ement s of 10%. Becaus e each
vi deo f eat ur ed 5 men l yi ng and 5 men t el l i ng the truth, observers
who si mpl y guessed woul d aver age five i t ems cor r ect ( 50%
accur acy) . Thus, we cat egor i zed t hose observers who scor ed
60% or bet t er on a gi ven vi deo as hi gh scorers f or that vi deo
and t hose who scored 50% or less as l ow- s c or e r s ) We chose
t hese cut offs f or t wo reasons. Fi rst , t he exper i ment al l i t erat ure
r epor t ed that observers rarel y surpass 60% accur acy when de-
t ect i ng decei t ( DePaul o et al., 1980). Second, a seri es of one-
sampl e t tests that set # at five i t ems cor r ect showed that six
i t ems cor r ect in ei t her scenar i o was si gni fi cant l y di f f er ent f r om
five i t ems cor r ect at t he t wo- t ai l ed p < .05 level.
The number of hi gh and l ow scores in each scenar i o is pre-
sent ed in Tabl e 1. Thi s t abl e shows that there is a st rong associ a-
t i on bet ween bei ng a hi gh scorer on the cr i me vi deo and bei ng
a hi gh scorer on t he opi ni on vi deo, X 2( 1, N = 48) = 6. 15, p
< .02. Sevent y- ei ght per cent of the hi gh scorers on t he cr i me
vi deo wer e al so hi gh scorers on t he opi ni on vi deo, and 70% of
the hi gh scorers on the opi ni on vi deo wer e hi gh scorers on t he
cr i me vi deo. Li kewi se, 67% of the l ow scorers on the opi ni on
vi deo wer e l ow scorers on t he cr i me vi deo and 57% vi ce versa.
Thus, it appears per f or mance when j udgi ng one hi gh-st ake de-
cept i on is r el at ed to per f or mance when j udgi ng a di fferent hi gh-
stake decept i on.
A Pearson cor r el at i on r eveal ed that this r el at i onshi p was lin-
ear ; a compar i son of all 48 obser ver s' accur acy rates f or de-
t ect i ng decept i on in t he cr i me vi deot ape wi t h t hei r per f or mance
in the opi ni on vi deot ape showed a si gni fi cant posi t i ve cor r el a-
t i on ( r = .48; p < .001, one- t ai l ed) . Thi s means that t hose
i ndi vi dual s who excel l ed at det ect i ng one t ype of l i e t ended to
excel in det ect i ng a di fferent t ype of lie, t hose who per f or med
at chance l evel s f or one wer e at chance f or another, and t hose
who di d poor l y for one t ype o f l i e di d poor l y f or t he other.
Subsidiary analyses. Ther e wer e no di fferences in overal l
accur acy bet ween obser ver s' per f or mance in t he cr i me vi deo
( M = 58%) and t hei r per f or mance in the opi ni on vi deo ( M =
Table 1
Number of Observers Who Scored High and Low on Judging
Deception Across Deception Scenarios in Experiment 1
Opinion scenario score
Crime scenario score High Low Total
High 21 6 27
Low 9 12 21
Tot ~ 30 18 48
Note. A high scorer judged deception at 60% accuracy and above. A
low scorer judged deception at 50% accuracy and below. Contingency
table X2(I, N = 48) = 6.15, p < .02.
59%) , F( 1 , 42) < 1. These l evel s of accur acy are at the hi gh
end of the t ypi cal range of det ect i on accur acy r epor t ed in ot her
st udi es of decept i on det ect i on ( see r evi ews by DePaul o,
St one, & Lassiter, 1985; Zucker man & Driver, 1985), and bot h
the cr i me vi deo accuracy, t ( 47) = 3. 78, p < .01, and the opi ni on
vi deo accuracy, t ( 47) = 3.61, p < .01, wer e great er than chance
( 50%) . The accur acy scores f or observers ranged f r om 10% to
90% in a nor mal di st ri but i on f or the opi ni on vi deo, and 10%
to 80% in a nor mal di st ri but i on f or t he cr i me vi deo.
We al so found no r el at i onshi p bet ween obser ver s' pret est
conf i dence in t hei r abi l i t i es to det ect l i es and t hei r act ual accu-
racy at det ect i ng lies f or ei t her vi deo ( f or cr i me, r = .10; f or
opi ni on, r = - . 1 5 ; bot h ns) , nor di d we find any rel at i onshi p
bet ween obser ver s' post t est conf i dence in t hei r per f or mance
aft er each vi deo and accur acy ( f or cr i me, r = . 13; f or opi ni on,
r = .13; bot h ns) . Thus, t hose who wer e mor e accur at e at
det ect i ng decei t di d not necessar i l y t hi nk t hey wer e any bet t er
than t hose who wer e less accurat e, ei t her bef or e engagi ng in the
task or aft er compl et i ng the task. However, we di d find that
pret est conf i dence and post t est conf i dence wer e si gni fi cant l y
cor r el at ed for the cr i me vi deo such that t hose who wer e confi -
dent pr i or to the cr i me vi deo t ended to be mor e confi dent aft er
that vi deo, and t hose l ess conf i dent bef or e the vi deo r emai ned
less confi dent af t er war d ( r = .53, p < . 001) . Ther e was no
such p r e - p o s t r el at i onshi p f or j udgi ng the opi ni on vi deo ( r =
.15, ns) . However, t hose who wer e mor e confi dent pr i or to
j udgi ng t he cr i me vi deo wer e al so mor e conf i dent pr i or to j udg-
ing the opi ni on vi deo ( r = .30, p < . 05). Thus, al t hough this
finding is not rel at ed to t hei r accuracy, obser ver s seem to have
a fai rl y r el i abl e vi ew of t hei r abi l i t i es to det ect decept i on.
We al so f ound a si gni fi cant decr ease in obser ver s' conf i dence
over the cour se of t he vi deos; obser ver s' conf i dence scor es aver-
aged 3.11 bef or e t he tests and 2.91 aft erward, F ( 1, 43) = 4. 13,
p < .05. However, we f ound that observers wer e no mor e confi -
dent on one test than on the ot her ( f or cr i me, M = 2. 99; f or
opi ni on, M = 3. 02) , F ( 1 , 43) < 1. We al so found that t he
obser ver s' conf i dence f or each t est di d not i nt eract wi t h t hei r
rat i ngs made bef or e or after the test, F ( I , 43) = 3. 01, ns.
Di s c us s i on
A per s on' s abi l i t y to det ect lies in one hi gh-st ake decept i on
si t uat i on was cor r el at ed wi t h hi s or her abi l i t y to do so in a
di fferent hi gh- st ake decept i on situation. Thi s is the first evi dence
that the abi l i t y to det ect decei t general i zes across hi gh-st ake
lies. We al so f ound that det ect i on accur acy and conf i dence in
one ' s abi l i t y wer e not cor r el at ed; t hese null resul t s r epl i cat e
ot her findings that conf i dence and det ect i on accur acy are not
rel at ed (e. g. , DePaul o & Pfeifer, 1986; Ekman & O' Sul l i van,
1991). However, conf i dence bef or e and aft er the vi deo was
correl at ed, even t hough it dr opped si gni fi cant l y duri ng t he vi deo.
Demonst r at i ng general i t y across t hese t wo t ypes of lies is
especi al l y not ewor t hy gi ven that the observers made their j udg-
ment s on t wo scenari os that di f f er ed on a number of i mpor t ant
5 We did not divide observers into groups of above chance, at chance,
and below chance because it would have created unanalyzably small
cell sizes.
DETECTING DECEIT 1435
di mensi ons that woul d wor k agai nst the general i t y hypot hesi s.
First, t hey di f f er ed on t he cont ent of t he lie. I n the cr i me sce-
nari o, part i ci pant s coul d t el l t he t rut h about ever yt hi ng except
whet her t hey had taken t he money; in the opi ni on scenari o,
part i ci pant s had t o concoct a coher ent fal se opi ni on and reasons
f or i t - - a consi der abl y mor e cogni t i vel y c ompl e x task. Second,
t hese scenari os di f f er ed in t he amount of t i me avai l abl e to fabri -
cat e t he lie. In the cr i me scenari o, decept i ve part i ci pant s had at
l east 8 mi n to creat e t hei r al i bi s; in t he opi ni on scenari o, decep-
tive part i ci pant s had onl y about 2 mi n to f or mul at e a posi t i on
opposi t e t o a posi t i on t hat t hey fel t ext r emel y st rongl y about.
Thi r d, t hese scenar i os di f f er ed in t hei r or der such t hat all part i ci -
pants went t hr ough t he cr i me scenar i o first and t he opi ni on
second. Thus, fact ors t hat mi ght af f ect t he par t i ci pant ' s perfor-
mance, such as fat i gue, fami l i ari t y, and ot her ef f ect s due to
order, woul d be apparent onl y in the opi ni on scenari o. Finally,
each vi deot ape f eat ur ed an ent i rel y di fferent sampl e of men wi t h
ver y di f f er ent abi l i t i es to l i e ( consi st ent wi t h Ekman et al.,
1991, and Kraut , 1980) ; thus, the finding of cr oss- si t uat i onal
general i t y is st ri ki ng when one consi der s t he addi t i onal var i ance
pr oduced by t hese di f f er ent abi l i t i es to decei ve.
Ther e are a number of si mi l ari t i es across t hese scenari os that
may have wor ked in f avor of the hypot hesi s. For exampl e, the
same i nt er r ogat or f ol l owed t he same l i ne of quest i oni ng in bot h
i nt errogat i ons, the part i ci pant s f aced the same stakes in bot h
scenari os, and the part i ci pant s wer e all me n - - wh o , r esear ch
has shown, di f f er f r om women in t hei r decept i on cl ues ( DePaul o
et al., 1985). However, t hese si mi l ar i t i es mi ght al so wor k
agai nst the general i t y hypot hesi s because t he si mi l ari t i es woul d
enabl e t he part i ci pant t o pr act i ce his demeanor or to change a
st rat egy that di d not s eem successf ul t o the part i ci pant in t he
first i nt errogat i on. We cannot pr eci sel y measur e what ef f ect
havi ng one scenar i o f ol l ow anot her had on each part i ci pant . I f
we pr es ume what is most l i k e l y - - t h a t it had di f f er ent ef f ect s on
di fferent pa r t i c i pa nt s - - t hi s woul d wor k agai nst the general i t y
hypot hesi s by i nt r oduci ng mor e var i ance to par t i ci pant s'
behavi or.
A final met hodol ogi cal i ssue i nvol ves t he smal l number of
st i mul us i t ems used in bot h tests ( 10 per vi de o) . Al t hough some
may ar gue that this smal l si ze does not al l ow an adequat e t est
of t he general i t y hypot hesi s, this smal l sampl e si ze al so wor ks
agai nst t he finding of stable i ndi vi dual di f f er ences because of
t he pr obl em of r est r i ct ed range of accur acy scores ( 0 - 1 0 ) .
It appears as i f the r eason we wer e abl e to find evi dence f or
general i t y was that we wer e abl e t o creat e t wo real i st i c hi gh-
stake decept i on scenar i os in whi ch we coul d document the ex-
pr essi on of negat i ve emot i on on t he part of the liars. Acr oss
bot h vi deos, we wer e abl e to cl assi f y accur at el y 80% of the
part i ci pant s sol el y on the basis of faci al expr essi ons of emot i on.
Al t hough t he pr evi ous l i t erat ure has used real i st i c par adi gms,
it appears as i f researchers used real i st i c l ow- st ake par adi gms
that in all l i kel i hood di d not generat e t he hi gh l evel s of emot i on
that occur r ed i n our par adi gm. The pr esumed l ack of st rong
emot i ons i n t hose par adi gms meant that it is most l i kel y the
verbal i n f o r ma t i o n - - t h a t is, t he wor ds part i ci pant s use to de-
scri be t hei r si t uat i ons, bel i efs, or a c t i o n s - - t h a t bet r ays decep-
tion. Thi s verbal i nf or mat i on, whi ch is most easi l y under the
vol i t i onal cont r ol of part i ci pant s ( Ekma n & Fri esen, 1969;
Zucker man et al., 1981 ), must necessar i l y be t i ed to the feat ures
of t he situation. Thus, it makes sense that t hese l ower stake
par adi gms woul d find evi dence consi st ent wi t h the not i on that
t he abi l i t y to det ect decei t is si t uat i on speci fi c or at l east over-
whel med by t he i ndi vi dual di f f er ences in t he l i ars' abi l i t i es to
decei ve ( DePaul o et al., 1980; Kraut , 1980). Yet we must not e
that t hose who st udi ed l ow- st ake l i es have not t o dat e done an
exper i ment compar abl e to this one; t hey have not shown obser v-
ers t wo di fferent ki nds of l i e si t uat i ons to det er mi ne whet her
accur acy is general or si t uat i on-speci fi c in l ow- st ake lies.
E x p e r i me n t 2
In this hi gh- st ake par adi gm, faci al expr essi ons of emot i on
bet r ayed decept i on, as Ekman ( 1985) predi ct ed. Thi s suggest s
that obser ver s who are profi ci ent at r eadi ng faci al expr essi ons
of emot i on shoul d be bet t er det ect ors of decei t . Ekman and
O' Sul l i van ( 1991 ) f ound in fact that observers who mor e accu-
rat el y r ecogni zed faci al expr essi ons of emot i on pr esent ed in a
way t o r es embl e mi cr omoment ar y faci al expr essi ons ( Ekman &
Fri esen, 1969; Haggar d & Isaacs, 1966) wer e al so mor e accu-
rat e in j udgi ng decept i on. However, t hei r study f eat ur ed part i ci -
pants who wer e wat chi ng fi l ms desi gned to el i ci t emot i ons, and
so t hei r par adi gm may have been st rongl y wei ght ed t owar d t he
i mpor t ance of r ecogni zi ng emot i ons. Our par adi gm is a mor e
pr ot ot ypi cal hi gh-st ake decept i on situation, in whi ch peopl e are
havi ng emot i ons caused by l yi ng, as compar ed wi t h l yi ng about
emot i ons ( Ekman & Frank, 1993). Thus, we at t empt ed to repl i -
cat e Ekman and O' Sul l i va n' s ( 1991 ) finding by showi ng phot o-
graphs of f aci al expr essi ons of emot i on at q2~ s and then de-
t er mi ni ng whet her accur acy in this t ask was cor r el at ed wi t h
accur acy in j udgi ng the t wo decept i on scenari os used in Exper i -
ment 1. Exper i ment 2 al so pr ovi ded an oppor t uni t y to r epl i cat e
t he general i t y findings we obt ai ned in Exper i ment 1.
Me t h o d
Participants. The observers were 13 male and 17 female undergrad-
uates from San Jose State University who received course credit for
participating in a study entitled "Can You Spot a Liar?"
Materials. We used the same deception detection materials in this
experiment as in Experiment 1. We assessed the ability to accurately
judge microexpressions of emotion with a 40-item microexpression test
videotape. This test consists of 40 slides of facial expressions of emo-
t i on-speci fi cal l y, anger, contempt, disgust, fear, happiness, sadness,
and surprise (taken from Japanese and Caucasian Facial Expressions
of Emotion, Matsumoto & Ekman, 1988)--t hat were flashed tachisto-
scopically onto a screen for ~125s. These tachistoscopic presentations
were videotaped, and the videotape served as the microexpression test.
Observers were instructed to circle the correct emotion term from a list
of seven for each of the 40 items presented.
Procedure. The observers saw the same two deception videotapes,
in counterbalanced order, as the observers did in Experiment 1. They
were run in classroom settings in groups of 5, 5, 6, and 14. We also
asked observers to rate their pre- and posttest confidence on the same
measures as used in Experiment 1. Observers viewed and judged which
of the 10 men in the crime video were lying and which were truthful
and then did the same for the opinion video (or vice versa). After judging
both deception videotapes, all observers judged the microexpression test.
We again predicted that an observer's level of accuracy in distinguish-
ing the liars from the truth tellers in the crime video would be related
to his or her performance in the opinion video. We also predicted a
1436 FRANK AND EKMAN
positive correlation between observers' performance on the deception
videos and their performance on the microexpression test. Finally, we
again predicted no relationship between confidence and accuracy.
Res ul t s
As in Exper i ment 1, we f ound no ef f ect f or gender of obser ver
or for or der of vi deot ape present at i on, all Fs ( 1 , 28) < 1, ns,
so we col l apsed across t hese vari abl es. We di vi ded observers
i nt o hi gh ( 60% or hi gher ) and l ow ( 50% or l ower ) scorers usi ng
the rat i onal e out l i ned in Exper i ment 1. A chi - squar e t abul at i ng
the hi gh scorers in one or bot h vi deos and the l ow scorers in
one or bot h vi deos shows t he same si gni fi cant pattern as r epor t ed
in Exper i ment 1; that is, t hose who wer e hi gh scorers in one
vi deo scenar i o wer e mor e l i kel y to be hi gh scorers in the other,
X2(1, N = 30) = 5. 46, p < .025, one-t ai l ed.
Table 2 shows that 79% of the observers who wer e cl assi fi ed
as hi gh scorers in the cr i me vi deo wer e al so cl assi fi ed as hi gh
scorers in the opi ni on vi deo ( and vi ce ver sa) . Li kewi se, 64%
of the obser ver s who wer e cl assi fi ed as l ow scorers in one of
the vi deos wer e cl assi fi ed as l ow scorers in the second. As in
Exper i ment 1, we found a si gni fi cant posi t i ve cor r el at i on be-
t ween an obser ver ' s per f or mance in det ect i ng lies in t he cr i me
vi deo and his or her per f or mance in the opi ni on vi deo ( r = .31;
p < .05, one- t ai l ed) .
Means and subsidiary analyses. As in Exper i ment 1, we
f ound no overal l mean di f f er ence in accur acy f or det ect i ng lies
in the opi ni on vi deo ( M = 58%) versus the cr i me vi deo ( M =
6 1 %) , F( 1 , 28) = 1.38, ns, al t hough accur acy scores wer e
agai n great er than chance f or the opi ni on vi deo, t ( 29) = 3. 53,
p < .01, and the cr i me vi deo, t ( 29) = 4. 04, p < .01. Except
f or a si gni fi cant r el at i onshi p bet ween pret est conf i dence and
accur acy for the opi ni on vi deo ( r = .39, p < . 05) , there was
general l y no rel at i onshi p bet ween pre- and post t est conf i dence
and accur acy at det ect i ng l i es in the cr i me vi deo ( f or pretest, r
= .20, ns; f or posttest, r = .05, ns) and the opi ni on vi deo ( f or
posttest, r = .03, ns) . Agai n there was a si gni fi cant rel at i onshi p
bet ween pre- and post t est conf i dence f or the cr i me vi deo ( r =
.41, p < .01 ) and this t i me al so f or t he opi ni on vi deo ( r = .31,
p < . 05) . We al so f ound a si gni fi cant rel at i onshi p bet ween
conf i dence pr i or to the cr i me vi deo and conf i dence pr i or to
j udgi ng t he opi ni on vi deo ( r = .59, p < .001 ). Unl i ke in Exper i -
ment 1, this t i me we f ound no change in pre- and post t est
conf i dence and no di fferences in conf i dence bet ween the cr i me
Tabl e 2
Number o f Observers Who Scored High and Low in Judging
Deception Across Deception Scenarios in Experi ment 2
Opinion scenario score
Crime scenario score High Low Total
High 15 4 19
Low 4 7 11
Total 19 11 30
Note. A high scorer judged deception at 60% accuracy and above. A
low scorer judged deception at 50% accuracy and below. Contingency
table X2(1, N = 30) = 5.46, p < .025.
and opi ni on vi deos, all Fs ( 1, 29) < 3.78, ns. These resul t s for
conf i dence r epl i cat e t he pattern f ound in Exper i ment 1; that is,
observers seem to have fai rl y r el i abl e bel i ef s about t hei r abi l i t i es
to det ect decept i on, i ndependent of t hei r actual ability.
Microexpression test. We scor ed obser ver s' responses on
the mi cr oexpr essi on test as 1 f or each cor r ect r esponse and 0
f or each i ncor r ect r esponse so that an obser ver ' s accur acy score
coul d range f r om 0 to 40. As predi ct ed, the obser ver s' scores
on the mi cr oexpr es s i on test cor r el at ed si gni fi cant l y wi t h t hei r
overal l ( combi ned cr i me and opi ni on vi de o) accur acy ( r = .35,
p < .04, one- t ai l ed) . When obser ver s' scores on the cr i me and
opi ni on vi deos wer e cor r el at ed separat el y wi t h t he mi cr oexpr es-
si on test, the cr i me vi deo accur acy cor r el at ed si gni fi cant l y wi t h
the mi cr oexpr es s i on test ( r = .34, p < .04, one- t ai l ed) , but the
opi ni on vi deo accuracy, al t hough also a posi t i ve rel at i onshi p,
di d not cor r el at e si gni fi cant l y ( r = .20, p = . 15 ). Al t hough this
suggest s that the abi l i t y to accur at el y j udge emot i on was mor e
i mpor t ant in det ect i ng decei t f or the cr i me scenar i o than f or the
opi ni on scenari o, an r-to-z t r ansf or mat i on test compar i ng the
cor r el at i on bet ween mi cr oexpr es s i on test accur acy and cr i me
vi deo accur acy ( r = . 34) and bet ween mi cr oexpr es s i on test
accur acy and opi ni on vi deo accur acy ( r = . 20) f ound that t hese
cor r el at i ons di d not di ffer f r om each ot her (z = .55, ns) . Finally,
the mi cr oexpr essi on test i t sel f di d appear to be a r el i abl e mea-
sure of abi l i t y to det ect mi cr oexpr es s i on of emot i on ( Cr on-
bach' s a = .82; spl i t - hal f rel i abi l i t y = . 84) .
Di s c us s i on
These resul t s r epl i cat e the finding f r om Exper i ment 1 that a
per s on' s abi l i t y to det ect l i es does not appear to be si t uat i onal l y
speci fi c but is a st abl e skill that general i zes across di fferent lies
t ol d in di fferent hi gh-st ake decept i on scenari os. An obser ver
who was abl e to accur at el y det ect lies in a cr i me scenar i o in
whi ch t he liar deni ed an al l egat i on was al so abl e to accur at el y
det ect lies in an opi ni on scenar i o in whi ch the l i ar at t empt ed
to creat e a coher ent and def ensi bl e opi ni on to whi ch he was
vehement l y opposed. Of cour se the conver se is al so true; t hose
observers who wer e poor at det ect i ng decei t in one scenar i o
wer e poor in the other.
We al so r epl i cat ed Ekman and O' Sul l i va n' s ( 1991) finding
that the abi l i t y to accurat el y det ect l i es is rel at ed to the abi l i t y
to accur at el y r ecogni ze mi cr omoment ar y faci al expr essi ons of
emot i on, thus ext endi ng t hei r findings to decept i on scenari os
besi des t hose that di rect l y i nvol ve conceal i ng negat i ve emot i ons.
These findings are consi st ent wi t h Ekma n' s ( 1985) reasoni ng
that hi gh-st ake si t uat i ons arouse emot i ons that can oft en bet ray
decept i on and that t he abi l i t y to r ecogni ze t hose emot i ons wi l l
ai d the l i e detector.
Finally, as in Exper i ment 1, we found a general pattern sug-
gest i ng no r el at i onshi p bet ween an obser ver ' s accur acy and his
or her rat ed conf i dence in his or her abi l i t y to det ect l i e s - -
ei t her bef or e or after compl et i on of t he l i e det ect i on t a s k - - e v e n
t hough obser ver s' pre- and post t est conf i dence was rel at i vel y
unchanged by the task. Al t hough we di d find one si gni fi cant
rel at i onshi p bet ween pret est conf i dence and accur acy for the
opi ni on vi deo, this cor r el at i on may have been due to chance
gi ven that we ran ei ght di fferent c onf i de nc e - a c c ur a c y combi na-
t i ons and obt ai ned onl y one si gni fi cant correl at i on. Moreover,
DETECTING DECEIT 1437
a number of other studies also have not found this relationship
(e.g., DePaulo & Pfeifer, 1986; Ekman & O' Sullivan, 1991).
General Di scussi on
Taken together, these two studies provide the first experimen-
tal evidence that a person' s ability to detect deceit may not be
situationally specific but instead may be a skill that generalizes
across different high-stake lies. It also appears that related to
this ability is the ability to accurately distinguish among micro-
momentary facial expressions of emotion.
At first glance, this generality finding seems to contradict
years of psychological literature, which has concluded that the
ability to detect deceit is specific to situations (e.g., DePaulo et
al., 1980; Kraut, 1980). This contradiction is resolved when
one notes that our study was able to show independent evidence
for the existence of strong negative emotions in the facial expres-
sions of 90% of the liars and the absence of these negative
emotions in the facial expressions of 70% of the truth tellers.
Thus, if most of the liars across two situations are showing
signs of negative emotion, and most of the truth tellers across
the two situations are not, then clearly lie catchers who observe
the presence or absence of these emotions are going to be consis-
tently better detectors of deceit across both situations than are
lie catchers who do not attend to these clues. The fact that
observers who were best able to distinguish among different
emotions shown at tachistoscopic speed were also the best detec-
tors of deceit in this experiment, where the liars showed facial
signs of negative emotions, strongly supports this explanation.
Thus, we were able to find evidence for generality in the ability
to detect deceit while others have not because our paradigm
successfully generated signs of emotion, whereas all but one of
the other studies have reported no evidence for such visible
clues. The only other study that documented visible and auditory
clues that betrayed deception reported findings consistent with
ours, that is, that the best detectors of deceit attended more to
nonverbal rather than verbal clues to deceit (Ekman & O' Sulli-
van, 1991). Finally, consistent with this finding is research
showing that liars who were motivated, and hence more aroused,
were more accurately judged from their nonverbal behavior and
that liars who were not motivated were more accurately judged
from their verbal behavior (DePaulo et al., 1983).
Our study was the first to find evidence that the ability to
detect the emotions involved in deceit is related to one' s ability
to detect deception across different high-stake situations. This
is consistent with earlier research, which has shown that within
a single high-stake deception situation, lie catchers who were
better able to detect leaked negative emotion were better able
to detect leaked positive emotion; moreover, those who were
better at detecting leakages by men were better able to detect
leakages by women (DePaulo & Rosenthal, 1979b).
The consistent pattern of these findings for emotion and de-
ception, obtained in different laboratories, suggests that the gen-
erality of the ability to detect deceit would not be limited to
some strange quirk in our choice of paradi gms--any two para-
digms that are able to generate strong emotions in the liars
should show evidence for generality. We make this assertion on
the basis of the large literature that has shown that across situa-
tions, people, and cultures, facial expressions of emotion not
only appear similar but also are recognized at levels greater
than chance (Ekrnan 1989, 1992, 1994). This is consistent with
Ekman' s (1985) argument that under high-stake deception situa-
tions, people are more likely to feel strong emotions when lying;
the extent to which these signs of emotions betray deception is
the extent to which a lie catcher would not have to know much
about the situation in order to infer accurately the presence or
absence of deception. Conversely, if strong emotions are not
elicited, then one would expect deception to be betrayed mostly
through "thinking" clues (words, factual descriptions, pauses,
long speech latencies, speech errors, etc.; see DePaulo et al.,
1985, for a review). Because thinking clues appear to be more
specific to particular situations than do emotional clues, one
would expect that in deception situations that do not arouse
strong emotions, the ability to detect deceit should be specific
to the situations.
These results highlight the fact that judging deception is a
two-step process (e.g., Bond et al., 1992; Burgoon & Walther,
1990; DePaulo et al., 1980; Ekman, 1985; Ekman & Friesen,
1969; Kraut, 1978). The first step in the process is to recognize a
sign, a clue, a behavior that violates expectations, or an emotion
displayed by a target person that is at odds with his or her verbal
line. The second step in the process is to interpret those clues
accurately. For example, is the person feeling anxious because
he or she is lying, or is that person truthful but afraid of being
disbelieved? What a high-stake, emotion-eliciting paradigm
does is to make more evident the signs of emotion, thus facilitat-
ing the recognition phase of the judgment. Our results show
that people who are good at spotting these clues, as measured
through the microexpression test, stand a better chance of suc-
cessfully completing the interpretation phase of the judgment
process, and if the situation is such that the clues to emotion
are correlated with deception, then an observer who can recog-
nize these emotion signs will outperform an observer who fails
to recognize these signs. Alternatively, if these emotional signs
are not present or are situation specific, then any advantage
brought about by higher skill in recognition of emotion will be
nullified, and one would expect situational specificity in the
ability to detect deception. This is exactly what the psychologi-
cal literature, which has focused on lower stake situations, has
reported over the years. To use a sports analogy, one cannot
win a championship without first making the playoffs; thus one
cannot accurately detect deception without first noticing some
quirk in the deceiver's behavior.
These results also highlight the notion that deception detec-
tion ability may not be a unitary construct, such as mathematics
ability or reading skill (cf. Knapp & Comadena, 1979). Because
there is no singular sign of deceit, and for some people or
situations, there are no signs of deceit at all, there can be no
singular strategy that would be perfectly successful in detecting
deceit. In other words, there is no universal algorithm for de-
tecting deception (Ekman, 1985; Zuckerman et al., 1981 ). Thus,
when we discuss deception detection ability, we must take into
account that the term is shorthand for a number of different
skills and abilities--some related, and some not.
Our results and the results of others suggest that one such
skill would be the recognition of emotion, as assessed by our
microexpression test or by some other test, such as the Profile
of Nonverbal Sensitivity (Rosenthal, Hall, DiMatteo, Rogers, &
1438 FRANK AND EKMAN
Archer, 1979). Thi s is by no means the onl y skill i nvol ved in
det ect i ng decei t ; ot her possi bl e skills are verbal compr ehensi on,
verbal reasoni ng, l ogi c, and so forth. Thi s suggest s that any
st rat egy to uncover the l i nk bet ween mor e gl obal personal i t y
traits such as ext r aver si on or consci ent i ousness and the det ect i on
of decei t is r el evant onl y to the ext ent that it capt ur es or is
compos ed of mor e basi c- l evel skills such as r ecogni t i on of
emot i on.
I f it is t hese basi c skills, such as the abi l i t y to r ecogni ze
fl eet i ng emot i ons, that i mpact decept i on det ect i on ability, then
one coul d pr edi ct that peopl e who are f or ced to t hor oughl y
devel op t hese s k i l l s - - b e c a u s e of l i fe ci r cumst ances or what -
e v e r - - s h o u l d out per f or m ot hers when det ect i ng decei t . We have
pr el i mi nar y evi dence that one such group, i ndi vi dual s wi t h l eft
hemi spher e brai n damage, who cannot pr ocess speech and mast
rel y excl usi vel y upon nonverbal behavi or to assess communi ca-
tions f r om others, tends to out per f or m normal , i nt act gr oups
when det ect i ng hi gh- st ake decei t ( Et cof f , Ekman, Frank,
Magee, & Torreano, 1992). We are f ol l owi ng up this wor k wi t h
si mi l arl y i mpai r ed groups.
Finally, t hese studies have i mpl i cat i ons f or i mpr ovi ng peo-
pl e ' s abi l i t i es to det ect decei t . The decept i on det ect i on t rai ni ng
studies have shown modes t yet st at i st i cal l y si gni fi cant i mpr ove-
ment s in accur acy ( usual l y bet ween 5% and 10% i ncreases in
accur acy; deTurck & Miller, 1990; Zucker man et al., 1984).
However, rarel y do the groups r ecei vi ng t rai ni ng surpass 70%
accuracy. Yet in Exper i ment 1, 19% of t he part i ci pant s scored
at or above 70% accur acy on bot h tests, and 6% scor ed at or
above 80% on bot h ( i n Exper i ment 2, the numbers wer e 13%
and 3%, r espect i vel y) . Thi s suggest s t wo, possi bl y cont r adi c-
tory, appr oaches to i mpr ovi ng the decept i on det ect i on abi l i t i es
of personnel in or gani zat i ons that deal wi t h hi gh- st ake decep-
t i ons in psychi at ri c, cust oms, and l aw enf or cement situations.
The first appr oach woul d be to sel ect t hose i ndi vi dual s who have
shown consi st ent abi l i t y to det ect l i es to be the or gani zat i ons'
i nt ervi ewers, rather than spend the r esour ces to train all i ndi vi du-
als in the or gani zat i on to be bet t er det ect ors of decei t . The
second appr oach woul d be to train t hese i ndi vi dual s to r ecogni ze
not decept i on but emot i on. The utility of each of t hese ap-
pr oaches coul d be det er mi ned empi ri cal l y, and t he exact skills,
st rat egi es, and exper i ences that make s omeone a good or poor
det ect or of decei t coul d be i sol at ed.
Finally, decept i on occurs in day- t o- day l i fe in many si t uat i ons
and under a vari et y of ci r cumst ances. In or der to ful l y under-
stand the pr ocesses of decept i on and det ect i on of decei t , re-
searchers must creat e real i st i c par adi gms that cover bot h hi gh-
and l ow- st ake situations. Researchers have been qui t e successf ul
at document i ng t he i nt erpersonal and si t uat i onal i nt eract i on pro-
cesses i nvol ved in day-t o-day, l ow- st ake sorts of decept i on situa-
tions and the i mpor t ant quest i ons t hey address about communi -
cat i on and human nat ure (e. g. , DePaul o, Kashy, Ki rkendol ,
Wyer, & Epst ei n, 1996; DePaul o et al., 1980). Our r esear ch is
now begi nni ng to unr avel the i nt eract i on pr ocesses i nvol ved in
hi gh-st ake decept i on situations, in whi ch t he successf ul det ec-
tion of decei t may be cri t i cal to heal t h and publ i c safety.
Re f e r e n c e s
Bond, C. E, Jr., Omar, A., Mahmoud, A., & Bonser, R. N. (1990). Lie
detection across cultures. Journal of Nonverbal Behavior, 14, 189-
204.
Bond, C. E, Jr., Omar, A., Pitre, U., Lashley, B. R., Skaggs, L. M., &
Kirk, C.T. (1992). Fishy-looking liars: Deception judgment from
expectancy violation. Journal of Personality and Social Psychology,
63, 969-977.
Burgoon, J. K., & Walther, J. B. (1990). Nonverbal expectancies and
the evaluative consequences of violations. Human Communication
Research, 17, 232-265.
DePaulo, B. M. (1994). Spotting lies: Can humans learn to do better?
Current Directions in Psychological Science, 3, 83-86.
DePaulo, B. M., Kashy, D. A., Kirkendol, S. E., Wyer, M. M., & Epstein,
J. A. (1996). Lying in everyday life. Journal of Personality and
Social Psychology, 70, 979-995.
DePaulo, B. M., Lanier, K., & Davis, T. (1983). Detecting the deceit of
the motivated liar. Journal of Personality and Social Psychology, 45,
1096-1103.
DePaulo, B. M. , & Pfeifer, R. L. (1986). On-the-job experience and
skill at detecting deception. Journal of Applied Social Psychology,
16, 249-267.
DePaulo, B. M., & Rosenthal, R. (1979a). Ambivalence, discrepancy,
and deception in nonverbal communication. In R. Rosenthal (Ed.),
Skill in nonverbal communication (pp. 204-248). Cambridge, MA:
Oelgeschlager, Gunn, & Hain.
DePaulo, B. M., & Rosenthal, R. (1979b). Telling lies. Journal of Per-
sonality and Social Psychology, 37, 1713-1722.
DePaulo, B. M. , Stone, J., & Lassiter, D. (1985). Deceiving and de-
tecting deceit. In B. R. Schlenker (Ed.), The self and social life (pp.
323-370). New York: McGraw-Hill.
DePaulo, B. M., Zuckerman, M., & Rosenthal, R. (1980). Humans as
lie detectors. Journal of Communication, 30, 129-139.
deTurck, M. A., & Miller, G. R. (1990). Training observers to detect
deception: Effects of self-monitoring and rehearsal. Human Communi-
cation Research, 16, 603-620.
Ekman, P. (1985). Telling lies: Clues to deceit in the marketplace,
politics, and marriage. New York: Norton.
Ekman, P. (1989). The argument and evidence about universals in facial
expressions of emotion. In H. Wagner & A. Manstead (Eds.), Hand-
book of psychophysiology: The biological psychology of the emotions
and social processes (pp. 143-164). New York: Wiley.
Ekman, P. (1992). Facial expressions of emotion: New findings, new
questions. Psychological Science, 3, 34-38.
Ekman, P. (1994). Strong evidence for universals in facial expression:
A reply to Russell' s mistaken critique. Psychological Bulletin, 115,
268-287.
Ekman, P., Davidson, R. J., & Friesen, W.V. (1990). The Duchenne
smile: Emotional expression and brain physiology: II. Journal of
Personality and Social Psychology, 58, 342-353.
Ekman, P., & Frank, M. G. (1993). Lies that fail. In M. Lewis & C.
Saarni (Eds.), Lying and deception in everyday life (pp. 184-200).
New York: Guilford Press.
Ekman, P., & Friesen, W.V. (1969). Nonverbal leakage and clues to
deception. Psychiatry, 32, 88-105.
Ekman, P., & Friesen, W. V. (1974). Detecting deception from body or
face. Journal of Personality and Social Psychology; 29, 288-298.
Ekman, P., & Friesen, W. V. (1975). Unmasking the face: A guide to
recognizing emotions from facial clues. Englewood Cliffs, NJ: Pren-
tice Hall.
Ekman, P., & Friesen, W. V. (1978). The facial action coding system.
Palo Alto, CA: Consulting Psychologists Press.
Ekman, P., Friesen, W. V., & O'Sullivan, M. (1988). Smiles when lying.
Journal of Personality and Social Psychology, 54, 414-420.
Ekman, P., Friesen, W. V., & Scherer, K. (1976). Body movement and
voice pitch in deceptive interaction. Semiotica, 16, 23-27.
Ekman, P., & O'Sullivan, M. (1991). Who can catch a liar? American
Psychologist, 46, 913-920.
DETECTING DECEIT 1439
Ekman, P., O'Sullivan, M., Friesen, W. V., & Scherer, K. ( 1991 ). Invited
article: Face, voice, and body in detecting deceit. Journal of Nonver-
bal Behavior, 15, 125-135.
Etcoff, N. L., Ekman, P., Frank, M. G., Magee, J., & Torreano, L. ( 1992,
August). Detecting deception: Do aphasics have an advantage? Paper
presented at the 1992 Annual Convention of the International Society
for Research in Emotion, Pittsburgh, PA.
Frank, M. G. (1992). Commentary: On the structure of lies and decep-
tion situations. In S. Ceci, M. DeSimone Leichtman, & M. B. Putnick
(Eds.), Cognitive and social factors in early deception (pp. 127-
146). Hillsdale, NJ: Erlbaum.
Frank, M. G., & Ekman, P. (1997). Facial behavior betrays deception.
Manuscript in preparation.
Geizer, R. S., Rarick, D. L. , & Soldow, G. E (1977). Deception and
judgmental accuracy: A study in person perception. Personality and
Social Psychology Bulletin, 3, 446-449.
Haggard, E. A., & Isaacs, K. S. (1966). Micromomentary facial expres-
sions as indicators of ego mechanisms in psychotherapy. In L. A.
Gottschalk and A. H. Aaerbach (Eds.), Methods of research in psycho-
therapy (pp. 154-165). New York: Appleton Century Crofts.
Hayano, D. M. (1980). Communicative competency among poker play-
ers. Journal of Communication, 30, 113-120.
Horowitz, S. (1989). The role of control questions in the physiological
detection of deception. Unpublished doctoral dissertation, University
of Utah, Salt Lake City.
Knapp, M. L., & Comadena, M. E. (1979). Telling it like it isn't: A
review of theorY and research on deceptive communications. Human
Communication Research, 5, 270-285.
Kraut, R. E. (1978). Verbal and nonverbal cues in the perception of
lying. Journal of Personality and Social Psychology, 36, 380-391.
Kraut, R. E. (1980). Humans as lie detectors: Some second thoughts.
Journal of Communication, 30, 209-216.
Kraut, R. E.., & Poe, D. (1980). Behavioral roots of person perception:
The deception judgments of customs inspectors and laymen. Journal
of Personality and Social Psychology, 39, 784-798.
Levenson, R. W. (1988). Emotion and the autonomic nervous system:
A prospectus for research on autonomic specificity. In H. L. Wagner
(Ed.), Social psychophysiology and emotion: Theory and clinical
applications (pp. 17-42). New York: Wiley.
Levenson, R. W., Ekman, P., & Friesen, W. V. (1990). Voluntary facial
action generates emotion-specific autonomic nervous system activity.
Psychophysiology, 27, 363-384.
Matsumoto, D., & Ekman, P. (1988). Japanese and Caucasian facial
expressions of emotion (JACFEE) [Slide set, available from first
author]. San Francisco State University.
Mehrabian, A. ( 1971 ). Nonverbal betrayal of feeling. Journal of Experi-
mental Research in Personality, 5, 64-73.
O'Sullivan, M., Ekman, P., Friesen, W. V,, & Scherer, K. (1992). Judg-
ing honest and deceptive behavior. Unpublished manuscript.
Podlesny, J. A., & Raskin, D. (1977). Physiological measures and the
detection of deception. Psychological Bulletin, 84, 782-799.
Rosenthal, R., Hall, J. A., DiMatteo, M. R., Rogers, R. L. , & Archer,
D. (1979). Sensitivity to nonverbal communication: The PONS test.
Baltimore: Johns Hopkins University Press.
Snyder, M. (1974). The self-monitoring of expressive behavior. Journal
of Personality and Social Psychology, 30, 526-537.
Zuckerman, M., DePaulo, B. M., & Rosenthal, R. (1981). Verbal and
nonverbal communication of deception. In L. Berkowitz (Ed.), Ad-
vances in experimental social psychology (Vol. 14, pp. 1- 59) . San
Diego, CA: Academic Press.
Zuckerman, M., & Driver, R. E. ( 1985 ). Telling lies: Verbal and nonver-
bal correlates of deception. In W. A. Siegman & S. Feldstein (Eds.),
Multichannel integration of nonverbal behavior (pp. 129-147).
Hillsdale, NJ: Erlbaum.
Zuckerman, M., Koestuer, R. E., & Alton, A. (1984). Learning to detect
deception. Journal of Personality and Social Psychology, 46, 519-
528.
Recei ved Apr i l 19, 1994
Revi si on r ecei ved Jul y 1, 1996
Accept ed Jul y 5, 1996