Week 2 Psychology Paper and Questions

O R I G I N A L P A P E R

Do you see what I see? Learning to detect micro expressions of emotion

Carolyn M. Hurley

Published online: 11 November 2011

� Springer Science+Business Media, LLC 2011

Abstract The ability to detect micro expressions is an

important skill for understanding a person’s true emotional

state, however, these quick expressions are often difficult

to detect. This is the first study to examine the effects of

boundary factors such as training format, exposure, moti-

vation, and reinforcement on the detection of micro

expressions of emotion. A 3 (training type) by 3 (rein-

forcement) fixed factor design with three control groups

was conducted, in which 306 participants were trained and

evaluated immediately after exposure and at 3 and 6 weeks

post-training. Training improved the recognition of micro

expressions and the greatest success was found when a

knowledgeable instructor facilitated the training and

employed diverse training techniques such as description,

practice and feedback (d’s [ .30). Recommendations are offered for future training of micro expressions, which can

be used in security, health, business, and intercultural

contexts.

Keywords Micro expression � Facial expression � Emotion � Training

Introduction

If facial expressions of emotion were delivered uniformly

each and every time an emotion was elicited, eventually all

of us would be near perfect perceivers of others. However,

pressures to conceal or mask one’s true feelings may result

in emotional displays that are quick or fragmented (called

micro momentary expressions, Haggard and Isaacs 1966;

or micro expressions, Ekman and Friesen 1969). Since

daily life features many pressures to conceal or mask one’s

emotions, as a function of status, culture, context, polite-

ness, and so forth (Ekman 1972), the ability to accurately

perceive and interpret these quick expressions would

improve our interpersonal skills, allowing us to better

understand individuals’ true emotional states.

The ability to ‘‘read’’ others is advantageous for the

average person, but in particular for clinicians and security

practitioners where the ability to understand others can

result in more informed judgments regarding threats to

oneself and others. Practitioners are already utilizing web-

based micro expression (ME) training in security (e.g.,

Department of State, Department of Homeland Security,

Department of Defense) and health contexts, although

testing of these efforts has been largely limited to clinical

populations (e.g., Marsh et al. 2010; Russell et al. 2006,

2008). Identifying effective training methods is imperative,

especially in these critical situations where a superior

understanding of emotion can significantly improve our

national security and quality of life.

The best available research in concealment of emotion

suggests that these masked emotional signals, particularly

MEs, are very difficult to detect (Ekman and Friesen 1969,

1974a; Etcoff et al. 2000; Porter and ten Brinke 2008).

Recent research has found that it is possible to train these

skills in a short period (Matsumoto and Hwang, in press),

This work was submitted in partial fulfillment of a Doctor of

Philosophy degree at the University at Buffalo by the author. Any

opinions, findings, and conclusions or recommendations expressed in

this material are those of the author and do not necessarily reflect the

views of the Transportation Security Administration, the Department

of Homeland Security, or the United States of America. The author

would like to thank Dr.’s Mark Frank and David Matsumoto for loan

of the Micro Expression Training Tool, second edition.

C. M. Hurley (&) Transportation Security Administration, 601 South 12th street,

Arlington, VA 22202, USA

e-mail: [email protected]

123

Motiv Emot (2012) 36:371–381

DOI 10.1007/s11031-011-9257-2

yet few boundary factors that may affect training success

have been explored. This manuscript examines the train-

ability of MEs of emotion, the optimal method of training,

the role of motivating factors, the effect of reinforcement,

and the retention of training materials over a 6-week per-

iod. This will help identify more effective training meth-

ods, which can be used to train individuals—such as those

in national security contexts—who may encounter con-

cealed emotions like MEs.

Background

Micro expressions of emotion

Emotions can be defined as ‘‘short-lived psychological-

physiological phenomena that represent efficient modes of

adaptation to changing environmental demands’’ (Levenson

1994, p. 123). Emotions are automatic responses that are

triggered—aroused in a fraction of a second—by environ-

mental stimuli that alter our attention and organize biological

responses, preparing us to react. Emotions are complex and

involve a number of bodily response systems such as

expression, muscular tonus, voice, and autonomic nervous

system activity (Levenson 1994).

Besides unique internal signals, emotions also generate

external signals—such as facial expressions—that provide

clues of these internal changes. A significant body of lit-

erature has examined the basic emotions of anger, con-

tempt, disgust, fear, happiness, sadness, and surprise,

revealing that each appears to have a characteristic

expression that is universal across cultures (e.g., Ekman

2003; Elfenbein and Ambady 2002). The universal pro-

duction of these facial signals suggests that these emotional

expressions are genetically determined and biology is lar-

gely responsible for establishing which facial movements

are associated with certain emotions (DeJong 1979;

DeMyer 1980).

A ME is a special case of the basic emotional expression,

which was first discovered by Haggard and Isaacs (1966)

while studying clinical interviews. They believed MEs were

caused by an unconscious repression of conflict and that

those expressions occurred too quickly to be seen in real

time. Ekman and Friesen (1969, 1974b) undertook a more

rigorous program of study that fully articulated the nature of

MEs. After examining recorded psychiatric interviews

frame-by-frame they found that MEs were emotional

expressions that ‘‘leaked’’ out when individuals attempted to

inhibit or manage their facial displays (Ekman, 2003). They

concluded that these quick expressions represented signs of

concealed emotion, as uninhibited or naturally occurring

emotional expressions generally last several seconds in

length or more (Hess and Kleck 1990).

The existence of MEs has been verified in studies of

concealment (Porter and ten Brinke 2008) and is relevant to

high-stakes contexts like law enforcement and national

security. For example, if someone is transiting a security

checkpoint and is in possession of illegal drugs, he may

have a fear of discovery. He will in all likelihood try to

hide these feelings, so any emotional clues he produces

may be more subtle then in a context where he is not trying

to manage his behavior. Research has shown that the

ability to detect MEs is related to skill at detecting

deception in high-stakes scenarios (Ekman and O’Sullivan

1991, 2006; Ekman et al. 1999), likely because it is easier

to judge veracity when an observer is able to accurately

understand how the target is feeling. This research

emphasizes the importance of ME recognition skills for

any individual whose profession requires interpersonal

interaction or deception detection.

Facial and micro expression training

Scientists have long endeavored to train people to better

recognize facial expressions. As early as the 1920s

researchers had students study pictorals or verbal descrip-

tions of facial expressions (Allport 1924; Guilford 1929;

Jarden and Fernberger 1926; Jenness 1932). However, the

absence of clear stimulus materials (drawings versus pho-

tographs) and clear identification of expressions limited

this training research. After researchers began to system-

atically study and define the muscle movements inherent in

emotional expressions they were able to create detailed

facial coding systems (e.g., Ekman and Friesen 1978; Izard

1979). This allowed researchers to create standardized sets

of valid emotion training and testing materials (e.g.,

BART, Ekman and Friesen 1974b; PoFA, Ekman and

Friesen 1975; JACFEE, Matsumoto and Ekman 1988;

JACBART, Matsumoto et al. 2000).

The Japanese and Caucasian Brief Affect Recognition

Test (JACBART) was the first published test of micro

expression recognition accuracy (MERA) that was rigor-

ously evaluated (Matsumoto et al. 2000). The JACBART

created the appearance of more dynamic expressions, as

each poser’s neutral face was imposed before and after the

emotional expression face, reducing the after effects of the

stimuli. All expression images were scored with the Facial

Action Coding System (FACS; Ekman and Friesen 1978)

to ensure the same muscle actions occurred for each

emotion and were consistent with universally recognized

expressions (Ekman 2003). Additionally, these images

were tested with an international audience to ensure cross-

cultural agreement (Biehl et al. 1997). Matsumoto and

colleagues provided evidence of internal and temporal

reliability and convergent and concurrent validity for this

test across five studies and found similar accuracy patterns

372 Motiv Emot (2012) 36:371–381

123

even with the differences made to presentation speed and

judgment task (Matsumoto et al. 2000).

This ME testing procedure evolved into a self-instruc-

tional training tool, originally called the Micro Expression

Training Tool (METT; now available as the METT

Advanced at face.paulekman.com and the Microexpression

Recognition Tool [MiX] at www.humintell.com). The

METT is presented as a stand-alone training tool; it offers a

pre-test, a training section, practice examples with feed-

back, a review section, and a post-test. The stimuli used in

these training tools are laboratory produced which provides

the necessary consistency and reliability of expression,

poser, intensity, angle and so forth to provide scientific test

of MERA. However, use of this type of materials limits the

ability to generalize to naturally occurring spontaneous

expression, which have more dynamic features (Naab and

Russell 2007).

Researchers have used versions of the METT to

train department store employees and trial consultants

(Matsumoto and Hwang, in press) and individuals with

Schizophrenia (Marsh et al. 2010; Russell et al. 2006,

2008) to detect MEs. A 2-h instructor led session using the

MiX not only significantly improved Korean department

store employees’ ability to identify MEs (N = 81, 18%

increase), but also led to higher social and communication

skills scores (Matsumoto and Hwang, in press). A similar

experiment using a small group of trial consultants also

showed improvements in accuracy (N = 25, 18%

increase). Further analyses revealed no skill decay over a

2-week period for both groups (Matsumoto and Hwang, in

press).

The METT has also been used to train clinical patients

with emotion recognition deficiencies to more accurately

recognize emotion (Marsh et al. 2010; Russell et al. 2006,

2008). Training individuals with Schizophrenia to read

facial expressions using the METT resulted in a significant

improvement in ME recognition at the post-test (9%

increase, Russell et al. 2006; 18% increase, Russell et al.

2008), illustrating the tool’s robustness to different popu-

lations. These studies support a meaningful training-accu-

racy relationship for identifying MEs, as well as, highlight

some possible social benefits.

Researchers have used other materials to teach others

about facial expressions. Stickle and Pellegreno (1982) and

Elfenbein (2006) used the Pictures of Facial Affect (PoFA,

Ekman and Friesen 1975) to train American students to

recognize emotional expressions (Elfenbein also used a

subset of Chinese posing facial expressions Wang and

Markham 1999). Although both studies reported success

for training, the authors did not report either the pre and

post accuracy scores and within subjects change (Stickle

and Pellegreno 1982) or the baseline recognition accuracy

(Elfenbein 2006). Those limitations inhibit interpretation of

these data. These studies also did not examine the ability to

detect quick expressions—such as MEs—further limiting

the ability to compare these methods to standardized tools

such as the METT or MiX.

Boundary factors to training

While research demonstrates the validity of using

commercial ME training tools to train recognition skills

(Matsumoto and Hwang, in press; Russell et al. 2006,

2008), little research has analyzed the underlying factors

associated with these skill improvements. Training formats

such as simple feedback (Elfenbein 2006), lecture and

practice (Stickle and Pellegreno 1982), and the METT/MiX

(Matsumoto and Hwang, in press) have all improved

expression recognition; but it is unknown which methods

have produced the greatest improvements or had the

greatest retention, due to differences in both testing mate-

rials and measures of effectiveness. It is also unknown

which format and materials are optimal for training indi-

viduals to detect MEs.

These studies revealed that individuals can be trained to

recognize laboratory produced MEs fairly quickly and

effectively, however, retention has only been examined in

one study and only at 2 weeks (Matsumoto and Hwang, in

press). Although training with the METT can improve

individuals’ recognition in as little as a few hours, the

length that this training outlasts the post-test is unknown.

Skill decay is an important variable to examine as many

military or government employees may only be able to

receive ME training once a year or once in a career span.

Another factor to consider is that understanding emo-

tional expressions is a skill that may improve with practice.

People who have repeated exposure to individuals who try

to conceal their emotions or who scrutinize nonverbal

behavior for their jobs—such as law enforcement officers,

judges, clinical psychologists, and secret service person-

nel—are often more accurate judges of how others are

feeling (Ekman and O’Sullivan 1991; Ekman et al. 1999).

Studies that have repeatedly tested the same participants

have found they improved without training (Matsumoto

et al. 2000). This suggests that repeated exposure to the

task or stimuli may serve as a training function as well and

should be examined.

Motivation can also influence a person’s ability to learn

material. Even though micro expression training may

improve MERA for all individuals, those who are more

motivated may learn and retain more material. Motivation

to learn is positively related to skill acquisition (Colquitt

et al. 2000), deeming it an important area for investigation.

It is important to examine individuals’ motivation to learn

Motiv Emot (2012) 36:371–381 373

123

both at the start and completion of each testing phase, as

motivation may be affected by external factors such as the

quality or content of the training or assignment to the

training or control group. Any differences must be con-

trolled for to insure that any gains made post-training can

be properly attributed to the training.

Overall, the previously published studies raise questions

regarding the optimal method of training, the role of

exposure and motivating factors, and the persistence of

training effects over time. It is important to examine these

boundary factors that may reduce skill loss so that

researchers can identify more effective training techniques.

The METT is an ideal instructional tool for testing these

differences. This training can be self-administered or

administered by an instructor in a group setting and pro-

vides enough stimulus materials to examine skill retention.

This will allow us to assess these factors in an existing and

well-used training.

Based on the above literature review, which found sig-

nificant improvements in MERA with different iterations

of the METT training (Matsumoto and Hwang, in press;

Russell et al. 2006, 2008), the following set of specific

hypotheses are proposed:

H1 ME Training will significantly improve participants’

MERA and result in greater skill retention, opposed to the

control conditions, which will experience no change in

MERA.

Although training by feedback alone has significantly

improved expression recognition skills (Elfenbein 2006),

ME recognition is an advanced skill which requires

understanding of subtle differences among expressions.

Thus,

H2 An instructor-led, multi-faceted ME training condi-

tion will produce the greatest increases in MERA, opposed

to ME training conditions that are self-led, or only provide

feedback to participants.

Any increased exposure to training material should also

provide an advantage to the exposed group. Thus,

H3 Reinforcement will significantly improve retention of

MERA.

Previous studies have assumed that a comparison group

assigned to do nothing during the training time serves as an

adequate control for examining training effects. Factors

such as mere exposure to stimuli or motivation to learn

could affect ME post-test scores or moderate effectiveness

of training. Thus, three control groups will also be exam-

ined to answer the following research question:

RQ1 What is the effect of motivation and simple expo-

sure on MERA?

Method

Participants

Three hundred thirty four (334) participants were recruited

from large introductory communication courses. An in-

class announcement advertised the study as ‘‘an evaluation

of students’ nonverbal communication skills’’ and inter-

ested students signed up for three 1-h appointments through

an online sign-up system. Participants who completed the

study received 3 h of research credit in partial fulfillment

of their 5 h departmental requirement.

Design

The study employed a 3 (training type—instructor feed-

back; instructor feedback plus description; or self led) by 3

(reinforcement—none; at time 2 only; or at time 3 only)

fixed factor design with three control groups (traditional

control; control with additional exposure of items; or

control with a motivating lecture). The four times at which

participants’ accuracy at judging MEs was assessed (pre-

training, immediately after training, 3 weeks later, and

6 weeks later) was treated as a within-subject independent

variable. The dependent variable was the participants’

accuracy on the various ME tests. Participants were ran-

domly assigned to each condition.

Conditions

Participants in the control conditions received no training

to serve as comparison groups to the training manipula-

tions. Participants in the ‘‘traditional’’ control condition

occupied themselves for the length of the manipulation and

were not exposed to any other emotional expression items.

Participants in the ‘‘exposure’’ control condition were

exposed to the same stimulus items (photographs of facial

expressions) as the training conditions during the manip-

ulation period, but received no feedback or other infor-

mation to facilitate their judgment. Participants in the

‘‘motivating lecture’’ control condition were provided with

a lecture on the importance of accurately perceiving and

interpreting human emotion based on the work of Ekman

(2001, 2003), but were not exposed to any other facial

expression material.

Training techniques previously published were com-

bined to allow for a fair comparison and evaluation of these

different training methods. Participants in both the ‘‘feed-

back only’’ training and ‘‘full instruction’’ training condi-

tions received the METT training led by an instructor

highly knowledgeable in the area of facial expressions of

emotion (the author). The difference was that the feedback

only training manipulation consisted solely of feedback

374 Motiv Emot (2012) 36:371–381

123

regarding the MEs of emotion (available in the practice

section of the METT), whereas in the full instruction

training manipulation the instructor also discussed subtle

differences among expressions (according to the ‘‘training’’

and ‘‘review’’ sections of METT) and answered questions

raised by participants. Participants in the ‘‘self-led’’ train-

ing condition also received training via the METT. These

participants led themselves through the training, feedback,

and review sections of the METT on a personal computer

(monitored by the instructor). The self-led training group

was exposed to the same materials as the full instruction

training group except the instructor was not allowed to

answer questions or discuss subtle differences to mirror a

true self-led training environment. The length of time was

standardized (25 min) for all six manipulations (both

training and control).

Reinforcement was manipulated by randomly assigning

the training participants to either receive or not receive

re-training at their second and third appointments.

Refreshers were identical in format to participants’ original

training conditions (i.e., feedback only, full instruction, or

self-led) although the instruction time was reduced to

15 min. Trained participants were randomly assigned into

one of three Refresher conditions: approximately one-third

received no refresher training, one-third received refresher

training at time 2, and one-third received refresher training

at time 3.

Stimulus materials

The second version of the METT was used for the testing

and training of MERA. The laboratory produced METT

expression items involve full-face flash displays that show

a subject’s neutral expression, a quick expression flash

(1/15th of a second), and then a return to the subject’s

neutral face. The METT training is divided into five sec-

tions: (1) a 14-item pre test (anger, contempt, disgust, fear,

happiness, sadness, and surprise, each shown twice), (2) a

training section in which each of the universal expressions

are introduced and described, (3) a 42-item practice sec-

tion, (4) a review section, and (5) a 28-item post-test (the

same seven emotions shown four times). Elements of this

training program were manipulated to form the stimulus

materials used to assess MERA as well as functioned as the

training materials in the training manipulations. To enable

three post-training assessment periods, a pilot test was

conducted to evaluate the difficulty of the expression items

so they could be grouped into equivalent post-tests. (The

third post-test was also used to assess MERA at the pre-test

period.) The 42 ME items taken from the pre-test and post-

test sections of the METT were shown separately to 12

communication undergraduates, who judged each of these

items at the speed of 1/15th second. These 42 items were

then divided into three sets to create three MERA post-

tests, each having two examples of each emotion. Paired

samples t-tests revealed no significant differences in test

difficulty among the three tests. The mean difficulty for

each of these tests based on the pilot data was 0.63 (post-

test 1), 0.66 (post-test 2), and 0.62 (post-test 3).

Procedure

Time 1

This study was conducted over an 8-week period and was

approved by the University’s Institutional Review Board.

Participants were scheduled in small groups for hour-long

sessions at three points in the semester. Participants were

randomly assigned to one of the six conditions and each

condition was run separately. One instructor (the author)

led all sessions. After arrival, participants completed an

informed consent document and then completed a demo-

graphic questionnaire and personality indexes. Then the

instructor provided an overview of the experiment and

explained the ME test procedure to the group. The format

and procedure of each ME test was identical. At this point

in the experiment the pre-test was administered according

to the procedure described below.

Before each test, participants were asked to indicate

their confidence in their ability to perform well, as well as

their motivation to correctly identify the ME items. Con-

fidence was measured using a 1 (Very poor) to 7 (Very

good) rating to the question: How well do you think you

will do at recognizing the upcoming facial expressions of

emotion? Motivation was assessed using a 1 (Not Moti-

vated) to 7 (Very Motivated) rating response to the ques-

tion: How motivated are you to recognize people’s

emotional expressions?

Next, participants viewed the fourteen-item ME test.

Each item was projected on a blank wall in the research

room at the speed of 1/15th of a second. Participants were

given approximately 10 s to judge each expression by

circling the appropriate response on the provided answer

sheet (choices included anger, contempt, disgust, fear,

happiness, sadness, surprise, and none of the above). After

all items had been judged, participants indicated their

confidence in their judgments. Post-confidence was mea-

sured using a 1 (Very poor) to 7 (Very good) rating to the

question: How well do you think you did at recognizing

these facial expressions of emotion?

After the pre-test was completed, the next 25 min served

as the manipulation period for the experiment. Control

participants received no training, and training participants

received ME training in one of three styles described pre-

viously. After the manipulation, all participants completed

the fourteen-item ME post-test (1) according to the

Motiv Emot (2012) 36:371–381 375

123

procedure described above. After the post-test participants

were reminded of their next research appointment, and

dismissed from the research space.

Time 2

Exactly 3 weeks after the first session, participants returned

to the research space. At this time participants in the

training conditions were randomly assigned as a group to

one of the three Reinforcement conditions: none, refresher

at time 2, or refresher at time 3. Participants assigned to a

refresher at time 2 received 15 min of training based on

their original training condition. After the manipulation

participants completed the fourteen-item ME post-test (2)

according to the procedure described previously. After all

participants completed the post-test they were reminded of

their next research appointment, and dismissed from the

research space.

Time 3

Exactly 6 weeks after the original training, participants

returned to the research space. Participants assigned to a

refresher at time 3 received 15 min of training based on

their original training condition. After the manipulation

participants completed the fourteen-item ME post-test (3)

according to the procedure described previously. After the

post-test, all participants completed a questionnaire

exploring how this study had impacted their lives. Last,

participants were debriefed regarding the purpose of the

study, provided research credit, and dismissed from the

research space.

Results

Participants

A total of 334 students participated at Time 1, with a 92%

completion rate (N = 306). Analyses were conducted to

determine if there were any differences in the demographic

makeup (age, gender, and ethnicity) of the 306 final subject

sample and the 28 participants who did not complete the

study. These analyses revealed no significant demographic

differences between the group who completed the study

and the group that dropped out. From hence forth, only the

final sample (N = 306) is discussed.

The participants were 174 female (57%) and 132 male

(43%) undergraduates with an average age of 20.13

(SD = 3.08) years. Participants were mostly Caucasian

(70.9%), but there were also participants who identified

themselves as Asian or Pacific Islander (11.1%), African or

Caribbean (8.8%), Hispanic (6.2%), Middle Eastern

(1.6%), or another ethnic background (1.4%). Participants

were mostly sophomores (38.2%) and juniors (32.7%),

although some seniors (15.0%) and freshman (13.1%) also

participated (1.0% did not list class year).

Motivation

In this study, participants were asked to rate how motivated

they were on a one item scale (1 = Not Motivated to

7 = Very Motivated). Independent samples t tests were

conducted to examine motivation differences between

untrained participants and trained participants. One sig-

nificant difference was uncovered for the pre-test,

t (304) = -2.133, p = .034, d = -.24, suggesting that

trained participants (M = 5.51, SD = 1.07) were more

motivated to succeed than the controls (M = 5.23,

SD = 1.04) before the manipulation. At this point in the

experiment participants had not received any information

regarding the training manipulation so the cause of the

greater motivation level is unknown. There were no sig-

nificant differences in motivation between controls

(M = 5.18, SD = 1.06) and training (M = 5.40, SD =

1.19) participants after the manipulation was introduced. A

one-way ANOVA was conducted to examine change in

motivation at Time 1. No significant differences were

uncovered; suggesting that assignment to a training group

did not significantly increase motivation to perform well in

this paradigm.

Pearson correlations were computed to examine the

relationship between motivation and accuracy. Motivation

was not significantly related to accuracy at any test for

control participants. For trained participants, motivation

was significantly positively related to accuracy at post-test

1, r (212) = .191, p = .005, and post-test 3, r (212) =

.157, p = .021, revealing that trained participants who

were more motivated to succeed were more accurate on

these tests. Since motivation was not significantly related to

accuracy at the pre-test—the only test in which groups

differed in motivation—it was dropped as a potential

covariate in ensuing analyses.

Confidence

In the current study confidence in judgment was measured

on a one-item scale (1 = Very poor to 7 = Very good)

both before and after each ME test. Pearson correlations

were computed to examine the relationship between con-

fidence and accuracy for both trained and control partici-

pants. All but two relationships were significant (Table 1).

The only negative relationship occurred for trained par-

ticipants at the pre-test, all other relationships were posi-

tive. This suggests that people’s perceptions regarding their

376 Motiv Emot (2012) 36:371–381

123

MERA were not so different from their objective ability

after MEs had been defined.

Training effects

H1 predicted a significant main effect for training, such

that trained participants would improve in accuracy post

manipulation at Time 1 and retain this improvement,

whereas controls would experience no change in accuracy.

A mixed model ANOVA was conducted to examine the

differences in accuracy across time within each of the six

conditions. Mauchly’s test indicated that the assumption of

sphericity had been violated, v2 (5) = 14.849, p = .011, therefore degrees of freedom were corrected using

Huynh–Feldt estimates of sphericity (e = .994). There was a significant main effect for time, F (2.983, 894.954) =

104.967, p \ .001, g2 = .259. Pairwise comparisons uncov- ered that accuracy was significantly different (p \ .001) for all tests except between post-test 2 and post-test 3, showing

that accuracy improved from pre-test to post-test 1 and

post-test 2.

There was also a significant main effect for condition,

F (5, 300) = 4.994, p \ .001, g2 = .077. Pairwise com- parisons indicated that the full instruction training condi-

tion was significantly more accurate than the traditional

control and motivating lecture control conditions, but was

not significantly different from the control group with

exposure, or feedback only training, or self-led training

conditions.

A significant interaction was revealed for time by con-

dition, F (14.916, 894.954) = 5.421, p \ .001, g2 = .083. To further explore this interaction, one-way ANOVAs

were conducted at each test (pre-test, post-test 1, post-test

2, and post-test 3) to examine between subject differences.

There were no significant differences at the pre-test,

revealing that all groups began at approximately the same

skill level. A significant difference was revealed at post-test

1, F (5, 300) = 7.561, p \ .001. Bonferroni post hoc tests revealed that the three control conditions were significantly

less accurate than the two instructor-led training condi-

tions. There were no significant differences between the

control conditions and the self-led training condition. There

was a significant main effect at post-test 2, F (5, 300) =

3.388, p = .005. Bonferroni post hoc tests revealed that the

full instruction training participants were significantly

more accurate than the traditional control participants. At

post-test 3 there was a significant main effect for condition,

F (5, 300) = 8.328, p \ .001. Bonferroni post hoc tests revealed that full instruction training participants were

significantly more accurate than the traditional control,

motivating lecture control, feedback only training, and self-

led training participants. There was no significant differ-

ence between the full instruction training condition and the

exposure control condition at post-test 3.

To further explore the within subjects differences,

paired samples t-tests were conducted for each of the six

conditions to examine accuracy change over time. A total

of 3 comparisons (pre-test vs. post-test 1, post-test 1 vs.

post-test 2, and post-test 2 vs. post-test 3) were conducted

for each condition. The significant differences are outlined

in the Table 2. Between the pre-test and post-test at Time

1, all three training conditions significantly increased in

accuracy (feedback only: ?14.19%; full instruction:

?19.52%; and self-led ?10.66%), and two of the control

conditions experienced no significant increase (traditional:

?1.43%; and exposure: -0.76%), revealing support for

H1. Surprisingly one of the control conditions (motivating

Table 1 Relationship between confidence and accuracy

Condition Confidence Accuracy

Pre-test Post-test 1 Post-test 2 Post-test 3

Control Pre- -.010 .170 .223* .260*

Post- .404*** .408*** .337** .598***

Training Pre- -1.77** .246*** .215** .155*

Post- .388*** .535*** .316*** .254***

* p \ .05; ** p \ .01; *** p \ .001

Table 2 Within subjects comparisons for accuracy from test to test

Condition Pre-test

versus

post-test 1, t

Post-test 1

versus

post-test 2, t

Post-test 2

versus

post-test 3, t

Control

Traditional .560 3.477** -2.065*

Exposure -.205 2.554* 1.745

Motivating lecture 2.143* 3.266** -1.929

Training

Feedback only 7.346*** 2.908** -1.586

Full instruction 8.757*** 3.106** 1.430

Self-led 5.635*** 6.289*** -3.898***

* p \ .05; ** p \ .01; *** p \ .001

Motiv Emot (2012) 36:371–381 377

123

lecture: ?6.15%) also significantly increased from the pre-

test to post-test 1, t (28) = 2.143, p = .041, d = -.40.

Between post-test 1 and post-test 2, all of the conditions

significantly improved in accuracy, suggesting a possible

exposure or practice effect to the stimuli, or that the

material shown in post-test 2 was easier than the other

tests, although pilot testing suggested that all three tests

were equivalent in difficulty. Between post-test 2 and post-

test 3, both the traditional control condition (-5.95%) and

the self-led training condition (-6.91%) significantly

decreased.

H1 was partially supported, as the combined training

participants outperformed control participants and more

specifically the full instruction training participants out-

performed most controls on all tests. However, the moti-

vating lecture control group also significantly improved

after the manipulation, suggesting an effect for the moti-

vating lecture. Additionally, all control groups significantly

improved from post-test 1 to post-test 2. This pattern of

results is illustrated in Fig. 1.

Type of training

H2 predicted that the full instruction training would result

in greater improved accuracy compared to feedback only

and self-led trainings. Independent samples t tests (one-

tailed) were conducted to examine the differences in

improvement from the pre-test to post-test 1 for the three

training types. Tests revealed that the full instruction

condition (?19.52%) improved significantly more than

both the self-led condition (?10.66%), t (138) = 3.021,

p \ .005, d = .51, and the feedback only condition

(?14.19%), t (143) = 1.811, p \ .05, d = .30, supporting H2. There were no significant differences between the

accuracy change of the feedback only and self-led

conditions.

The role of refreshers

H3 predicted that refreshers would aid in retention of

training material. A mixed model ANOVA showed a sig-

nificant main effect for time, F (2, 410) = 23.338, p \ .001, g2 = .102 (Table 3). Pairwise comparisons revealed that accuracy was significantly higher for post-tests 2 and 3

compared to post-test 1 (p \ .001). There was also a sig- nificant main effect for condition, F (2, 205) = 4.017,

p = .019, g2 = .038. Pairwise comparisons indicated that the full instruction condition was significantly more accu-

rate than the self-led condition. There was no main effect

for refresher type.

There was a significant interaction for time by condition,

F (4, 410) = 4.300, p = .002, g2 = .040. This interaction was previously explored and reported, and revealed that full

instruction condition improved significantly more than the

self-led and feedback only conditions. The two-way inter-

actions for time by refresher, and condition by refresher, and

the three-way interaction for time by refresher by condition,

were not significant.

Paired samples t tests (post-test 1 vs. post-test 2, post-

test 2 vs. post-test 3, and post-test 1 vs. post-test 3) were

conducted to evaluate MERA retention for each of the

three refresher manipulations within the three training

groups. In the feedback only condition, the time 2 refresher

group significantly improved in accuracy after the

refresher, t (21) = 3.346, p = .003, d = .71 (post-test 1 to

post-test 2). T tests also revealed that accuracy significantly

decreased from post-test 2 to post-test 3, t (21) = -3.186,

p = .004, d = -.68. No other significant differences,

Fig. 1 Accuracy across time and conditions

Table 3 Within subjects differences for ME accuracy over time

Training type Refresher

type

N Accuracy

Post-test

1 (%)

Post-test

2 (%)

Post-test

3 (%)

Feedback only None 27 74.60 78.70 78.57

At time 2 22 70.45 80.30 72.40

At time 3 25 83.43 86.67 85.14

Full instruction None 24 81.55 85.07 91.96

At time 2 25 78.00 87.33 89.43

At time 3 22 78.25 84.47 88.31

Self-led None 25 72.29 85.33 74.57

At time 2 20 71.79 80.83 76.43

At time 3 24 72.32 88.89 83.93

378 Motiv Emot (2012) 36:371–381

123

including improvements for the time 3 refresher group,

were uncovered.

In the full instruction condition, the time 2 refresher

group significantly increased in accuracy from post-test 1

to post-test 2, t (24) = 2.402, p = .024, d = .48, and post-

test 1 to post-test 3, t (24) = 3.578, p = .002, d = .72. For

the time 3 refresher group, paired samples t tests revealed

that accuracy significantly increased from post-test 1 to

post-test 3, t (21) = 3.241, p = .004, d = .69. No other

significant differences were uncovered.

In the self-led condition, the no refresher group signif-

icantly increased in accuracy from post-test 1 to post-test 2,

t (24) = 3.496, p = .002, d = .70. Accuracy at post-test 3

was significantly less than at post-test 2, t (24) = -4.064,

p \ .001, d = -.81. This was the only non-refresher group that performed significantly different on one of the post-

tests. For the time 2 refresher group, paired samples t tests

revealed success for the refresher in significantly increas-

ing accuracy from post-test 1 to post-test 2, t (19) = 2.323,

p = .031, d = .52. For the time 3 refresher group, accu-

racy significantly increased from post-test 1 to post-test 2,

t (23) = 5.175, p \ .001, d = 1.06, and from post-test 1 to post-test 3, t (23) = 3.646, p = .011, d = .74, revealing an

increase prior to the refresher. No other significant differ-

ences were uncovered.

These inconsistent results do not support H3. The time 2

refresher groups did experience significant increases in

accuracy after their refresher, but this result did not outlast

the time period. Actually, all groups increased at Time 2

(although not all changes were significant) suggesting that

the refresher may not have caused increases but rather by

some condition of the ME test.

Discussion

This study provides the first data comparing different

methods of training micro expressions, the effects of

motivation and exposure on recognition, and skill retention

over three points in time. As predicted, the training was

successful. At the pre-test, there were no significant dif-

ferences in accuracy based on condition, but after the

manipulation trained participants performed better on post-

tests 1, 2 and 3 (76, 84, and 82% respectively) than controls

(64, 75, and 73% respectively). Further, having an expert

present to guide participants through the subtle differences

among these expressions and answer questions was an

advantage over the other tested training methods. The full

instruction training condition continually provided consis-

tent results: it was the only training condition that was

significantly more accurate than one or more control con-

ditions on all three post-tests.

These results revealed that the best method for using the

METT in a short session was to fully explore all sections of

the program including the training and review, have a

knowledgeable instructor describe subtle differences

between the expressions, and practice identifying the ME

items and provide feedback to trainees. This suggests that

feedback paired with additional training techniques may

produce a more effective training manipulation. Although

the METT has been designed as a self-instructional tool to

train emotion recognition, this method of training was

considerably less effective compared to an instructor-led

training.

Two explanations for this finding may be that the full

instruction training provided both more material than the

self-led training (instructor answered questions) as well as

enthusiasm for the topic. This study does not definitively

show which factor provided significant benefits. However,

the surprising finding that the motivating lecture control

group also significantly increased at after the manipulation

(?6%) suggests that the instructor’s enthusiasm may have

provided motivation to concentrate or attend closer to the

post-test. Further research should examine the both the role

of content and instructor in training ME skills, as there may

be some ideal combination of content and charisma that

produces the greatest effects.

In this experiment, the same instructor was used for all

trials in attempt to keep the presentation consistent and

eliminate the possibility of attention or motivation biases

caused by the instructor’s appearance or presentation style.

However, this is also a limitation of the study since the

instructor was aware of the hypotheses and experimental

design, which could have unintentionally affected por-

trayals within the experiment.

Research examining confidence of judgment has gen-

erally not found a relationship between one’s confidence in

judgment and the accuracy of that judgment (DePaulo et al.

1997; Patterson et al. 2001). The current study revealed a

clear relationship between confidence and MERA: before

individuals were introduced to the concept of MEs, their

confidence and accuracy were not calibrated, but after

individuals had seen MEs, they became calibrated such that

accurate judges were more confident and less accurate

judges were less confident. Previous studies examining

emotion recognition have not found this strong a link. In

this study, participants’ confidence in interpreting the MEs

was significantly positively correlated to their accuracy

post-testing period, even though they had never received

feedback on their performance. This may suggest a very

parsimonious means to determine whether trainees under-

stand the material—the trainer merely has to ask. The

novelty of this finding suggests that this relationship should

be verified in subsequent research. Of particular importance

Motiv Emot (2012) 36:371–381 379

123

would be to replicate this finding with naturally occurring

MEs, which may be more difficult to spot.

In this study a possible repeated exposure effect was

found, as untrained individuals improved without train-

ing—and most markedly in the repeated exposure control.

If practice—or exposure—improves performance, this

suggests additional training time would be beneficial. One

limitation of the current study was the limited time spent

training (25 min) and refreshing (15 min) ME skills. Per-

haps this is the reason why the reinforcement sessions were

ineffective. This data suggests that exposure is an impor-

tant element for learning, and future studies should explore

increased exposure and training time manipulations.

A limitation of the current study was all conditions

performed significantly better on post-test 2 (as compared

to the pre-test and post-test 1), which suggests that the

stimuli utilized in post-test 2 may have been easier than the

other tests. Pilot tests were conducted to ensure that the ME

items were divided into equally difficult post-tests, and

while not significant the pattern revealed by the means

suggests that post-test 2 was slightly easier. Although the

availability of a subject pool was prohibitive in this sense,

future research should counterbalance the order of these

tests to insure that accuracy is due to the manipulation, not

the ease of any particular test.

Another limitation was the nature of the MEs used to

both test and train recognition. These MEs were full-face

but very quick expressions of emotion that were imbedded

within a poser’s neutral expression. Research has shown

that spontaneous expressions are more difficult to interpret

than posed expressions, as often naturally occurring

expressions blend with other emotions or expressions

(Naab and Russell 2007). Naturally occurring MEs may not

engage every action unit, do not occur at the same speed or

intensity every time, may be masked or covered by other

expressions, and accuracy of recognition may be affected

by lighting, angle, attention, and the observer’s cognitive

load (Porter and ten Brinke 2008). Length of expression

itself is an additional variable that was not examined in this

study; it is possible that good ME perceivers could identify

MEs that last for varying lengths of time, as they would

naturally occur. While this study presented a necessary first

step to understand boundary factors, these results should be

repeated using more ecologically valid stimuli.

It is important to explore this training on samples of real

world practitioners—such as law enforcement, behavior

detection officers, physicians, clinical groups, and negoti-

ators. However, this training is only useful to real world

practitioners if it relates to a real world skill. One study

under review (Frank et al. 2011) found that the METT

training resulted in significantly better ability to catch ‘‘real

life micros’’ for Coast Guard officers. Although these

officers almost doubled their abilities to accurately read

these MEs, their average post training accuracy (38%) was

much lower than ME accuracy for posed photographs seen

both in their study (78%) and this current study (81%

across all post-tests). The posed faces selected for training

with the METT are not representative of all of the facial

expressions encountered daily, and therefore future

research is required to determine whether these are the best

materials for training spontaneous expression recognition.

There is a need to correctly recognize and interpret a

person’s true feelings in any number of interpersonal,

health, business, legal, and social contexts. The detection

of concealed or masked emotions is invaluable in law

enforcement and national security settings, medical con-

texts and the corporate world—where better understanding

of our suspects, patients, or partners can allow us to make

more informed decisions about a person’s true feelings and

intent. In any context, the ability to recognize emotional

displays can make us more effective perceivers of others,

which can enhance the quality of our interpersonal rela-

tionships and reduce the potential for misunderstanding.

This study was among the first to evaluate the specific

features and use of the METT, a facial expression training

program currently in use in security and health contexts.

These findings validate use of the METT for improving

MERA, suggest the training persists at 6 weeks, and further

provide the optimal way to deploy that training. Previous

work suggests that this type of training will translate to real

time spontaneously expressed MEs, but this particular tool

requires further testing to conclusively demonstrate its utility

across the wide variety of situations seen in daily life.

References

Allport, F. H. (1924). Social psychology. Boston: Houghton Mifflin. Biehl, M., Matsumoto, D., Ekman, P., Hearn, V., Heider, K., Kudoh,

T., et al. (1997). Matsumoto and Ekman’s Japanese and

Caucasian facial expressions of emotion (JACFEE): Reliability

data and cross-national differences. Journal of Nonverbal Behavior, 21, 3–21. doi:10.1023/A:1024902500935.

Colquitt, J. A., LePine, J. A., & Noe, R. A. (2000). Toward an

integrative theory of training motivation: A meta-analytic path

analysis of 20 years of research. Journal of Applied Psychology, 85, 678–707.

DeJong, R. N. (1979). The neurologic examination. Hagerstown, MD: Harper & Row.

DeMyer, W. (1980). Technique of the neurologic examination. New York: McGraw-Hill.

DePaulo, B. M., Charlton, K., Cooper, H., Lindsay, J. J., &

Muhlenbruck, L. (1997). The accuracy–confidence correlation

in the detection of deception. Personality and Social Psychology Review, 1, 346–357. doi:10.1207/s15327957pspr0104_5.

Ekman, P. (1972). Universals and cultural differences in facial

expression of emotion. In J. Cole (Ed.), Nebraska symposium on motivation (Vol. 19, pp. 207–283). Lincoln: University of Nebraska Press.

380 Motiv Emot (2012) 36:371–381

123

Ekman, P. (2001). Telling Lies: Clues to deceit in the marketplace, politics, and marriage. New York: W. W. Norton & Co.

Ekman, P. (2003). Emotions revealed: Recognizing faces and feelings to improve communication and emotional life. New York: Henry Holt & Co.

Ekman, P., & Friesen, W. V. (1969). Nonverbal leakage and cues to

deception. Psychiatry, 32, 88–106. Ekman, P., & Friesen, W. V. (1974a). Detecting deception from the

body or the face. Journal of Personality and Social Psychology, 29, 124–129. doi:10.1037/h0036006.

Ekman, P., & Friesen, W. V. (1974b). Nonverbal behavior and

psychopathy. In R. J. Friedman & M. Katz (Eds.), The psychology of depression: Contemporary theory and research (pp. 3–31). Washington, D.C.: Winston and Sons.

Ekman, P., & Friesen, W. V. (1975). Pictures of facial affect instrument. Palo Alto, CA: Consulting Psychologist Press.

Ekman, P., & Friesen, W. V. (1978). Facial action coding system. Palo Alto, CA: Consulting Psychologists Press.

Ekman, P., & O’Sullivan, M. (1991). Who can catch a liar? American Psychologist, 46, 189–204. doi:10.1037/0003-066X.46.9.913.

Ekman, P., & O’Sullivan, M. (2006). From flawed self-assessment to

blatant whoppers: The utility of voluntary and involuntary

behavior in detecting deception. Behavioral Sciences & The Law, 24, 673–686. doi:10.1002/bsl.729.

Ekman, P., O’Sullivan, M., & Frank, M. G. (1999). A few can catch a

liar. Psychological Science, 10, 263–266. doi:10.1111/1467- 9280.00147.

Elfenbein, H. A. (2006). Learning in emotion judgments: Training

and the cross-cultural understanding of facial expressions.

Journal of Nonverbal Behavior, 30, 21–36. doi:10.1007/s10919- 005-0002-y.

Elfenbein, H. A., & Ambady, N. (2002). On the universality and

cultural specificity of emotion recognition: A meta-analysis.

Psychological Bulletin, 128, 203–235. doi:10.1037/0033-2909. 128.2.203.

Etcoff, N. L., Ekman, P., Magee, J. J., & Frank, M. G. (2000). Lie

detection and language comprehension. Nature, 405, 139. doi: 10.1038/35012129.

Frank, M. G., Matsumoto, D. M., Ekman, P., Kang, S., & Kurylo, A.

(2011). Improving the ability to recognize micro-expressions of

emotion. Manuscript under review. Guilford, J. P. (1929). An experiment in learning to read facial

expression. The Journal of Abnormal and Social Psychology, 24, 191–202. doi:10.1037/h0069973.

Haggard, E. A., & Isaacs, K. S. (1966). Micromomentary facial

expressions as indicators of ego mechanisms in psychotherapy.

In L. A. Gottschalk & A. H. Auerbach (Eds.), Methods of research in psychotherapy (pp. 154–165). New York: Appleton Century Crofts.

Hess, U., & Kleck, R. E. (1990). Differentiating emotion elicited and

deliberate emotional facial expressions. European Journal of Social Psychology, 20, 369–385. doi:10.1002/ejsp.2420200502.

Izard, C. E. (1979). The maximally discriminative facial movement coding system. Unpublished manuscript; Instructional Resources Center, University of Delaware.

Jarden, E., & Fernberger, S. (1926). The effects of suggestion on the judgment of facial expressions of emotion. American Journal of Psychology, 37, 565–570. doi:10.2307/1414917.

Jenness, A. (1932). The effects of coaching subjects in the recognition

of facial expressions. Journal of General Psychology, 7, 163–178.

Levenson, R. W. (1994). Human emotion: A functional view. In P.

Ekman & R. J. Davidson (Eds.), The nature of emotion: Fundamental questions (pp. 123–126). New York: Oxford University Press.

Marsh, P. J., Green, M. J., Russell, T. A., McGuire, J., Harris, A., &

Coltheart, M. (2010). Remediation of facial emotion recognition

in schizophrenia: Functional predictors, generalizability, and

durability. American Journal of Psychiatric Rehabilitation, 13, 143–170. doi:10.1080/15487761003757066.

Matsumoto, D., & Ekman, P. (1988). Japanese and Caucasian facial expressions of emotion (JACFEE) [Slides]. San Francisco, CA: Intercultural and Emotion Research Laboratory, Department of

Psychology, San Francisco State University.

Matsumoto, D., & Hwang, H. S. (in press). Evidence for training the

ability to read microexpressions of emotion. Motivation and Emotion. Retrieved from http://www.humintell.com/wp-content/ uploads/2011/05/Uncorrected-Proof2.pdf.

Matsumoto, D., LeRoux, J., Wilson-Cohn, C., Raroque, J., Kooken,

K., Ekman, P., et al. (2000). A new test to measure emotion

recognition ability: Matsumoto and Ekman’s Japanese and

Caucasian brief affect recognition test (JACBART). Journal of Nonverbal Behavior, 24, 179–209. doi:10.1023/A:10066681 20583.

Naab, P. J., & Russell, J. A. (2007). Judgments of emotion from

spontaneous facial expressions of New Guineans. Emotion, 7, 736–744. doi:10.1037/1528-3542.7.4.736.

Patterson, M. L., Foster, J. L., & Bellmer, C. (2001). Another look at

accuracy and confidence in social judgments. Journal of Nonver- bal Behavior, 25, 207–219. doi:10.1023/A:1010675210696.

Porter, S., & ten Brinke, L. (2008). Reading between the lies:

Identifying concealed and falsified emotions in universal facial

expressions. Psychological Science, 19, 508–514. doi:10.1111/j. 1467-9280.2008.02116.x.

Russell, T. A., Chu, E., & Phillips, M. L. (2006). A pilot study to

investigate the effectiveness of emotion recognition in schizophre-

nia using micro-expression training tool. British Journal of Clinical Psychology, 45, 579–583. doi:10.1348/014466505X90866.

Russell, T. A., Green, M. J., Simpson, I., & Coltheart, M. (2008).

Remediation of facial emotion perception in schizophrenia:

Concomitant changes in visual attention. Schizophrenia Research, 103, 248–256. doi:10.1016/j.schres.2008.04.033.

Stickle, F. E., & Pellegreno, D. (1982). Training individuals to label

nonverbal facial cues. Psychology in the Schools, 19, 384–387. doi: 10.1002/1520-6807(198207)19:3\384:AID-PITS2310190321[ 3.0.CO;2-A.

Wang, L., & Markham, R. (1999). The development of a series of

photographs of Chinese facial expressions of emotion. Journal of Cross-Cultural Psychology, 30, 397–410. doi:10.1177/0022022 199030004001.

Motiv Emot (2012) 36:371–381 381

123

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.