Design and Validation of a Mental and Social Stress Induction Protocol

Towards Load-invariant Physiology-based Stress Detection

Camille Jeunet

1,2

, Fabien Lotte

and Christian M

uhl

Inria Bordeaux Sud-Ouest / LaBRI, 200 rue de la Vieille Tour, 33405 Talence, France

Laboratoire Handicap & Syst

eme Nerveux, Universit

e Bordeaux Segalen, 146 Rue Leo Saignat, 33076 Bordeaux, France

Keywords:

Mental Stress, Psychosocial Stress, Cognitive Workload, Physiological Computing, ECG, GSR, EEG.

Abstract:

Stress is a major societal issue with negative impacts on health and economy. Physiological computing of-

fers a continuous, direct, and unobtrusive method for stress level assessment and computer-assisted stress

management. However, stress is a complex construct and its physiology can vary depending on its source:

cognitive workload or social evaluation. To study the feasibility of physiology-based load-invariant psychoso-

cial stress-detection, we designed a stress-induction protocol able to independently vary the relevant types of

psychophysiological activity: mental and psychosocial stress. Here, we validate the efﬁcacy of our protocol

to induce psychosocial and mental stress. Our participants (N=24) had to perform a cognitive task associated

with two workload conditions (low/high mental stress), in two contexts (low/high psychosocial stress), during

which we recorded subjects’ self-reports, behaviour, physiology and neurophysiology. Questionnaires showed

that the subjectively perceived level of stress varied with the psychosocial stress induction, while perceived

arousal and mental effort levels vary with mental stress induction. Behaviour and physiology further cor-

roborated the validity of our protocol. Heart rate and skin conductance globally increased after psychosocial

stress induction relative to the non-stressful condition. Moreover, we demonstrated that higher workload tasks

(mental stress) led to decrease in performance and a marked increase of heart rate.

1 INTRODUCTION

Stress is a universal societal issue, affecting both

economy and health. Thus, it is easy to understand

why many people invest in ﬁnding ways to deal with

stress (Regehr et al., 2013): how to help people man-

age their stress is becoming a major preoccupation

in many countries. Computer-assisted stress man-

agement is one way to support coping with stress.

However, it requires reliable stress level assessment

(van den Broek and Westerink, 2012).

Besides psychological questionnaires, many de-

vices are available to assess stress levels. They

measure stress-related physiological markers such as

heart rate, skin conductance or blood pressure, which

are increased during a stressful episode (see Section

1.1). The availability of cheap sensor technology and

small, portable computing devices allows to automat-

ically and continuously monitor the level of stress in

every-day contexts, as during driving (Healey and Pi-

card, 2005) and work (Kusserow et al., 2012), or in

clinical contexts (Hogervorst et al., 2013).

However, there are still several challenges that

have to be addressed to be able to successfully mon-

itor stress levels with physiological sensors. One

of the most notorious is the deﬁnition of the rela-

tionship between physiological measurements and the

psychophysiological construct of stress. For example,

it is known that modiﬁcations of the above physiolog-

ical markers are characteristic of psychosocial stress,

but not exclusively affected by it (Dickerson and Ke-

meny, 2004). For example cardiovascular measures,

like heart rate and its variability, are known to respond

to stress as well as to high cognitive workload or to

exciting situations. The same is true for several fre-

quency bands in the electroencephalogram (EEG): for

example the alpha frequency band has been shown to

covary with stress/relaxation, but is also known to re-

spond strongly to sensory stimulation, attention and

cognitive workload (see Section 1.1). Therefore, the

relationship between measurements and psychophys-

iological constructs is a complex many-to-many map-

ping for which great care has to be taken in the aim of

ﬁnding the right mapping parameters to avoid con-

fusion between episodes of psychosocial stress and

those of high cognitive worload, i.e., of mental stress.

Jeunet C., Lotte F. and Mühl C..

Design and Validation of a Mental and Social Stress Induction Protocol - Towards Load-invariant Physiology-based Stress Detection.

DOI: 10.5220/0004724100980106

In Proceedings of the International Conference on Physiological Computing Systems (PhyCS-2014), pages 98-106

ISBN: 978-989-758-006-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

To enable the creation of systems that are capa-

ble of differentiating activation of body and brain due

to psychosocial stress from those due to mental stress

(associated to high cognitive workload), it is neces-

sary to ensure psychophysiological validity of the in-

ference, as well as the diagnosticity and generality of

the detection system (Fairclough, 2009). To tackle

this essential requirement of physiological comput-

ing, we devised an affect induction protocol that ma-

nipulates mental and bodily activation due to psy-

chosocial and mental stress independently. The data

gathered with it deliver the ground truth necessary to

calibrate reliable stress classiﬁers. Here, we present

the protocol and subjective (self-assessments) and ob-

jective (physiological and behavioural measures) evi-

dence of its validity.

This paper is organized as follows: concepts of

stress and related work on physiological computing as

well as our hypotheses are presented in the remainder

of the introduction, Section 2 presents material and

methods, then results are presented in Section 3 and

discussed in Section 4; ﬁnally, we conclude our work

with a perspective on future studies in Section 5.

1.1 Stress Responses

The psychophysiological concept of ’stress’ was in-

troduced in 1936 by (Selye, 1936) to describe ’the

non-speciﬁc response of the body to any demand for

change’. In that sense, it is an organism’s response

to an environmental situation or stimulus perceived

negatively - called a ’stressor’ - which can be real

or imagined, that taxes the capacities of the subject,

and thus has an impact on the body’s homeostasis

(that is to say that the constants of the internal en-

vironment are modiﬁed). To face the demand (i.e.,

to restore homeostasis), two brain circuitries can be

activated during a ’stress response cascade’ (Sinha

et al., 2003; Dickerson and Kemeny, 2004; Taniguchi

et al., 2009): the sympatho-adrenomedullary axis

(SAMa, also called the noradenergic circuitry) and

the hypothalamus-pituitary gland-adrenal cortex axis

(HPAa). On the one hand, the SAMa induces the re-

lease of noradrenaline which allows immediate phys-

ical reactions (such as increased heart rate and skin

conductance, or auditory and visual exclusion phe-

nomena) associated with a preparation for violent

muscular action (Dickerson and Kemeny, 2004). On

the other hand, HPAa activation (which is lower) re-

sults in the releasing of cortisol, the purpose of which

is to redistribute energy in order to face the threat.

Thus, more energy is allocated to the organs that need

it most (brain and heart), while non-necessary organs

for immediate survival (reproductive, immune and di-

gestive systems) are inhibited. This stress response

cascade ends when homeostasis is restored. However,

stress can be of different types, such as physical, psy-

chological or psychosocial (Dickerson and Kemeny,

2004), each kind of stress being associated with a spe-

ciﬁc response (contrary to what has previously been

said in Selye’s deﬁnition of stress). Indeed, physi-

cal stress, induced by extreme temperatures or phys-

ical pain for example, is associated with an increase

of heart rate (Loggia et al., 2011), skin conductance

(Boucsein, 1992; Buchanan et al., 2006) and subjec-

tive stress ratings but with only a weak cortisol re-

sponse (Dickerson and Kemeny, 2004). These results

suggest that this kind of stress induces an activation

of the SAMa but only a weak activation of the HPAa.

Mental (or psychological) stress, associated with dif-

ﬁcult cognitive tasks (i.e., high workload), uncontrol-

lability or negative emotions triggers a weak release

of cortisol (weak HPAa activation), but strong effects

on heart rate and skin conductance (strong SAMa ac-

tivation) (Boucsein, 1992; Reinhardt et al., 2012). Fi-

nally, psychosocial stress, triggered by a social eval-

uation threat (that is to say a situation in which the

person’s own estimated social value is likely to be de-

graded) added to uncontrollability (in particular dur-

ing the Trier Social Stress Task (TSST) (Kirschbaum

et al., 1993)) has been shown to induce a strong acti-

vation of both the SAMa (Hellhammer and Schubert,

2012) and the HPAa (Dickerson and Kemeny, 2004).

Thanks to research in the ﬁeld of neurosciences,

and consistently with the notion of neural response

systems, we know that stress also has strong elec-

troencephalographic correlates. One of the most

prominent correlates of psychosocial stress is found

in the alpha band, and speciﬁcally in brain asymme-

try. Tops et al. (Tops et al., 2006) proposed that cor-

tisol administration (which simulates a stressful sit-

uation) leads to a global decrease of cortical activ-

ity (except for the left frontal cortex in which ac-

tivity is increased). However, other studies (Lewis

et al., 2007; Hewig et al., 2008) showed that stress

was associated with a higher activity in the right hemi-

sphere, and that right hemisphere activation was cor-

related with negative affect. For Crost and colleagues

(Crost et al., 2008), the explanation of these conﬂict-

ing results would be that an association between EEG-

asymmetry and personality characteristics may only

be observed in relevant situations to the personality

dimensions of interest. As for the other physiologi-

cal responses, different types of stress share similar

neurophysiological responses. For mental stress for

example, we know that the alpha band plays a role

as well (Brouwer et al., 2012). Given the overlap in

the response characteristics between different types of

DesignandValidationofaMentalandSocialStressInductionProtocol-TowardsLoad-invariantPhysiology-basedStress

Detection

stress, it is essential to explore these types of stress in

context, that is to say in one experiment.

To summarise, stress is a process composed of ﬁve

phases: -1- an environmental demand that disrupts

homeostasis, -2- a personal perception of this demand

by the person , -3- a stressor-speciﬁc activation of the

stress-response, -4- behavioural consequences of this

activation, -5- a return to homeostasis once the threat

is overcome, or exhaustion if the body has not enough

resources to face the demand. Depending on the type

of stressor, different types of stress occur, sharing

mechanisms and overlapping in response characteris-

tics. To identify valid psychophysiological inferences

in every-day stress detection systems, these types of

stress have to be experimentally delineated.

1.2 Physiological Computing and Stress

Several works in the domain of physiological com-

puting have addressed the measurement of various

types of stress from physiological and neurophysi-

ological sensors. For example, Healey has shown

that during different driving phases, various levels

of stress can be automatically infered from physio-

logical sensors, assessing cardiovascular and respi-

ratory responses (Healey and Picard, 2005). Hoger-

vorst and colleagues showed that in the context of an

impending eye surgery, physiological signals such as

heart rate can be used to classify stress (Hogervorst

et al., 2013). Khosrowabadi and colleagues showed

that stress emerging during an examination period

can be identiﬁed from the EEG (Khosrowabadi and

Quek, 2011). Riera found that neurophysiological

signals can differ between the different levels of stress

elicited in an interview setting (Riera et al., 2012).

As these studies show, the stress response can

be automatically detected using information recorded

from the somatic, autonomous, and central nervous

systems, which therefore pose a potential basis for

stress detection systems. However, as layed out in

the former section, physiological stress responses due

to psychosocial stress and those due to mental de-

mand resemble each other. This is especially rele-

vant for the construction of stress detection systems,

since in every-day life and work mental demand and

psychosocial stress are often co-occuring: mental task

demand might lead to psychosocial stress due to the

associated evaluation and psychosocial stress often is

related to certain (demanding) tasks that have to be

conducted. Therefore, care has to be taken to ensure

the validity of the psychophysiological inference of-

fered by a physiology-based stress detection method.

As stressed by Fairclough (Fairclough, 2009), psy-

chophysiological validity can be ensured by a number

of approaches, including standardised experimental

stimuli and tasks for the induction of the psychophys-

iological constructs in question, and the assessment of

subjective and objective measures, such as observable

behaviour or physiological signals known to respond

to the psychophysiological construct in question. Re-

turning to the above mentioned studies, no distinction

between the type of stress involved has been made.

However, such destinction is critical for systems that

are supposed to be sensitive to one speciﬁc type of

stress, while being invariant for others. That is why

it is important to devise a protocol that allows to vary

psychosocial and mental stress independently.

Our long-term goal is the development of a system

that can be used during private or professional work,

which is able to differentiate psychophysiological ac-

tivation due to stress compared to that due to cogni-

tive workload (mental stress). Therefore, we devised

a stress induction protocol that would take the follow-

ing requirements into account: 1) enable recording

during actual tasks, not during resting periods (eco-

logical validity), 2) manipulate the context of the task,

but maintain same sensory stimulation and task dur-

ing measurement (avoiding confounds), 3) introduce

different levels of cognitive workload during stress-

ful and non-stressful conditions (enabling the creation

of a classiﬁer speciﬁc to psychosocial stress and in-

variant to mental stress). In line with Fairclough’s re-

quirements, we combine two recognised experimental

paradigms to induce psychosocial and mental stress

and validate this protocol via subjective ratings and

objective measures from behaviour and physiology.

1.3 Hypotheses

To ensure the validity and efﬁciency of the affec-

tive manipulation (i.e., psychosocial stress) within

our stress induction protocol, we analyzed partici-

pants’ self-assessments (State-Trait Anxiety Inven-

tory, arousal scale of the Self-Assessment Maneken)

to see if subjective experience of anxiety and arousal

was increasing in the stressful condition of our proto-

col relative to the control condition. For a more ob-

jective evaluation, we also analysed the physiological

(heart rate and skin conductance) differences accord-

ing to the condition (stress vs. no-stress). Accord-

ing to the literature provided in Section 1.1, we were

expecting an increase of the heart rate and skin con-

ductance during the stressful condition. To ensure the

validity and efﬁcacy of the cognitive workload ma-

nipulation (i.e., mental stress), we analysed partici-

pants’ self-assessments (Rating Scale Mental Effort,

arousal scale of the Self-Assessment Maneken) ex-

pecting to ﬁnd an increase in perceived cognitive load

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

100

and arousal during high cognitive workload compared

to low cognitive workload. Furthermore, we expected

a decrease of task performance with higher workload.

Finally, we also hypothesised an increase of heart rate

with high workload (see Section 1.1).

2 METHODS

2.1 Participants

12 female and 12 male participants were recruited for

our experiment. The participants were between 18

and 54 years old, with a mean age of 24.7 (SD: 7.9),

and all except 4 were right-handed. Educations varied

between high school degree and Ph.D., with a mean

education of high school + 3.1 years (SD: 2.4). To be

admitted, people had to be at least 18 and had to sign

an informed consent. Furthermore, we had criteria

of non inclusion: bad vision, heart, neurological and

psychological diseases, and affective troubles. More-

over, people were asked to schedule on a day and at

a time at which they expected to feel good, not too

tired. Finally, we asked them not to drink coffee and

tea less than two hours before the experiment.

2.2 Material

For our recordings, we used the following sensors

(see Figure 1): electroencephalogram (EEG, 28+8 -in

a 10/20 system without T7, T8, Fp1 and Fp2- active

electrodes -g.LADYbird- with g.USBamp), electro-

cardiogram (ECG, 2 active electrodes), electromyo-

gram (EMG, 2 active electrodes), electrooculogram

(EOG, 4 active electrodes), breath belt (SleepSense),

pulse-sensor (g.PULSEsensor), and a galvanic skin

response sensor (g.GSRsensor). The workload task

was designed in the Presentation software (Neurobe-

havioral Systems, www.neurobs.com/presentation)

and EEG signals were recorded and visually inspected

with Open ViBE (Renard et al., 2010).

Subjects were ﬁrst asked to sign an informed con-

sent and to ﬁll in three questionnaires. The ﬁrst one

concerned personal characteristics (such as gender,

age and education). The second and the third ones

were State-Trait Anxiety Inventory (STAI) form Y-A

and Y-B (Spielberger et al., 1970) (see 2.2.1 for more

details). Then, all the sensors were put into place and

a three minute baseline was recorded. From this point,

as we wanted to counterbalance the conditions (not

to have any order effect), we set up four scenarios

(see Figure 2) composed of two blocks each, sepa-

rated by a STAI form Y-A questionnaire. Therefore,

we randomly began with either relaxation or stress

Figure 1: Illustration of our set up.

induction, and we randomly started with either low

workload (0) or high workload (2). During both the

ﬁrst and the second block, the subject did each work-

load condition six times (low/high) (6x2x2=24min

per block), with a short break after six tasks. Once

the two blocks were completed, the subject ﬁlled in

the STAI form Y-A one last time before the sensors

were taken off. Finally, the participant was debriefed

about the aim of the experiment.

2.2.1 Neuropsychological Stress Evaluation

In order to measure the level of anxiety of the sub-

jects, the ”State Trait Anxiety Inventory” (Spielberger

et al., 1970) was used. This test is composed of two

scales of 20 propositions each: STAI form Y-A and

STAI form Y-B. STAI form Y-A score increases in a

stressful situation. It is a good indicator for transi-

tory modiﬁcations of the level of stress. STAI form

Y-B evaluates the clinical anxiety (trait), and thus al-

lows to recognize generally anxious people (who have

higher scores).

2.2.2 Psychosocial Stress and Relaxation

Inductions

In order to induce stress, a 15 minutes long stress-

induction protocol based on the Trier Social Stress

Task (or TSST, (Kirschbaum et al., 1993)) which is

a validated protocol, was set up. This stress induction

protocol is composed of three parts and requires the

participation of three people, ”the committee”, who

are presented as being body language experts. In the

ﬁrst part, a member of the committee tells the subject

he has ﬁve minutes to prepare a fake job interview

for a position as a teacher. During the second part,

the committee asks the person to do this job interview

and to speak about himself for 5 minutes. They tell

the subject that he is ﬁlmed (for a future behavioural

DesignandValidationofaMentalandSocialStressInductionProtocol-TowardsLoad-invariantPhysiology-basedStress

Detection

101

Figure 2: Chronology of our Experimental Protocol: each row represents a different scenario (R=Relax; S=Stress; 0=0-back;

2=2-back).

analysis) and they take notes. The committee acts

serious and neutral/unresponsive towards the subject.

The third part is an arithmetic task lasting 3 minutes.

The subject has to count backwards from 2083 to 0

by steps of 13 and to begin again at any mistake or

hesitation. At the end of this protocol, in order to

keep the stress level high, the comity tells the sub-

ject he will be ﬁlmed during the workload tasks and

that he will have to do another interview, which will

be longer, and a self-evaluation after it. Furthermore,

during the experiment, participants have to perform

cognitive tasks, followed by feedback corresponding

to their performance. During the stress condition, the

score shown to the participant was automatically de-

creased by 5 to 10%. Thus, this protocol includes psy-

chosocial stress and uncontrollability in order to max-

imise the chances of triggering a stress response in all

participants (Dickerson and Kemeny, 2004). On the

other hand, the goal of the relaxation induction was

to create a condition in which participants would be

able to relax. In this last part, they were allowed to ei-

ther rest in silence, or select relaxing music or videos

(Krout, 2007).

2.2.3 Workload Tasks

We used the n-back task (Kirchner, 1958) to manipu-

late mental stress, as it was easy to modify the work-

load without changing any other parameter (such as

visual stimulation or motor behaviour). Indeed, 0-

back (low workload) and 2-back (high workload) are

very similar. In both, sixty white letters (font-size

48) appear (during 500ms) one after the other (with a

1500ms break between them) on a black background.

Thus, each task lasts two minutes. Among these let-

ters, 25% are targets. In both tasks, when the let-

ter that appears is a target, the subject has to do a

left click, and a right click otherwise (like this, there

will not be a big difference at the motor level be-

tween targets and non target letters: this motor dif-

ference could have induced unexpected differences in

the EEG). Thus, for the 0-back task, that is to say the

low workload condition, the target is the letter ”X”:

each time an ”X” appears, the subject has to do a left

click, and in all the other cases he has to do a right

click. For the 2-back task, the high workload condi-

tion, the subject has to do a left click when the letter

that appears is the same as two letters before (for ex-

ample, if the sequence ”C A C” appeared, the second

”C” would be a target). At the end of each task, the

subject has to report his perceived level of arousal (on

a scale from 1 to 9 of the Self-Assessment Maneken,

SAM) and the spent mental effort (Rate Scale of Men-

tal Effort, RSME) (Brouwer et al., 2012). Finally, a

screen with the subject’s performance at the prior task

(number of targets correctly identiﬁed) appears.

2.3 Variables and Factors

In this study, we analysed the effects of two factors

(with two possible values each) on different depen-

dent variables (DV). The ﬁrst factor is the psychoso-

cial stress level: condition ”stress” (after the stress

induction) vs. condition ”no-stress” (after relaxation

induction); the second factor is the mental stress level:

low workload (low-WL, 0-back task) vs. high work-

load (high-WL, 2-back task). Thus, we had four

experimental conditions: stress/low-WL, stress/high-

WL, no-stress/low-WL and no-stress/high-WL. The

DVs are: STAI form Y-A score, subjective arousal

evaluation, subjective task difﬁculty evaluation, tasks

performance, heart rate and skin conductance.

2.4 Data Analyses

2.4.1 Self-assessment Data

To investigate the effect of psychosocial stress induc-

tion on the state-trait anxiety inventory, we computed

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

102

(a) STAI (b) Arousal (c) RSME

Figure 3: Mean and standard error of mean of subjective stress level assessments : (a) STAI form Y-A scores, (b) SAM arousal

scale and (c) RSME. (a) shows signiﬁcant increase of perceived stress during the stressful condition compared to the baseline

and the relax condition. (b) and (c) show an increase of perceived arousal and mental effort for the 2-back compared to 0-back

task.

an ANOVA with the self-ratings in the 3 factor-levels

(baseline, after relaxation, after stress induction). To

test the effect of psychosocial and mental stress ma-

nipulation, we conducted 2 (stress) × 2 (workload)

ANOVAs for the averaged-over-blocks ratings (one

value per subject for stress/low workload, stress/high

workload, relaxation/low workload, relaxation/high

workload blocks) on the arousal scale of the SAM and

on the RSME.

2.4.2 Behavioural Data

To investigate the effects of the experimental manip-

ulations on behaviour, we calculated the performance

per block based on the number of true positive (T P),

true negative (T N), false negative (FN), and false pos-

itive (FP) responses resulting from the button presses

within the n-back task (left click for targets, right

click for non-targets) using the following equation:

Per f = (T P + T N) ÷ (T P + T N + FP + FN). As for

ratings, we analysed the data then in a 2 (stress) × 2

(workload) ANOVA.

2.4.3 Physiological Data

Physiological responses were analysed with respect

to heart rate (HR) and galvanic skin response (GSR).

Before applying statistical methods, the GSR data

were pre-processed by extracting the GSR value (µS)

for each block and then averaging these values over

the blocks described above. The ECG signal was

band-pass ﬁltered between 5 and 200 Hz, applying

a notch-ﬁlter 48-52 Hz to reduce power line noise,

before HR for each of the blocks was extracted us-

ing the BioSig toolbox

. As for the former analyses,

we then analysed the data in a 2 (stress) × 2 (work-

http://biosig.sourceforge.net/

load) ANOVA. We are reporting data as signiﬁcant if

p < 0.05 and as trend if p < 0.1.

3 RESULTS

3.1 Subjective Indicators

3.1.1 State-trait Anxiety Inventory

Each subject ﬁlled in three ”STAI form Y-A” (state)

questionnaires: at the beginning (STAI

) of the

experiment, and after each of the stress conditions

(stress: STAI

; no-stress: STAI

) (see Figure 3(a)).

Three datasets were excluded due to incompleteness.

An ANOVA (N = 21) with ’baseline’, ’stress’, and

’relax’ factor levels showed a signiﬁcant difference

of perceived anxiety between the conditions (F(2,20)

= 3.6225, p ¡ 0.05, η

= 0.108). We conducted a

post-hoc analysis using directed paired t-tests accord-

ing to the hypothesis that subjectively perceived stress

increases due to the stress induction procedure rela-

tive to baseline and relaxation condition. The results

show the expected effect (see Figure 3(a)). STAI

= 37.5, SD

= 12.6) scores are signiﬁcantly higher

(t(20) = 2.87, p = 0.005) than STAI

scores (M

30.1, SD

= 4.6) and also signiﬁcantly higher (t(20)

= 2.37, p = 0.014) than STAI

scores (M

= 32.2,

= 8.6). We found no difference between STAI

and STAI

((t(20) = 1.27, p = 0.11).

3.1.2 Perceived Arousal and Mental Effort

We asked the subjects after each block to rate their

arousal on the respective scale of the Self-Assessment

Maneken (see Figure 3(b)) and to rate the invested

mental effort on the Rating Scale Mental Effort (see

DesignandValidationofaMentalandSocialStressInductionProtocol-TowardsLoad-invariantPhysiology-basedStress

Detection

103

(a) Performance (b) GSR (c) Heart Rate

Figure 4: Mean and standard error of mean of objective stress level assessments : (a) Performance, (b) Galvanic Skin Response

and (c) Heart Rate. (a) depicts a lower performance in 2-back compared to 0-back tasks. (b) shows that GSR signiﬁcantly

increased only for the ”stress” condition compared to the ”relax” condition. (c) depicts a signiﬁcant increase of heart rate

during high workload compared to low-workload tasks.

Figure 3(c)). Two datasets were excluded due to in-

completeness. We submitted the data of each scale

to a 2 (stress) × 2 (workload) ANOVA. Regard-

ing subjectively perceived arousal, we only found a

main effect of the workload manipulation (F(1,21) =

4.444, p = 0.047, η

= 0.175) with higher perceived

arousal for the 2-back task (M

2−back

= 4.7, SD

2−back

= 1.4) compared to the 0-back task (M

0−back

= 4.3,

0−back

= 1.7). Regarding the subjectively per-

ceived workload, we only found a main effect of the

workload manipulation (F(1,21) = 63.216, p ¡ 0.0001,

= 0.751) with higher perceived effort for the 2-

back task (M

2−back

= 48.1, SD

2−back

= 11.5) com-

pared to the 0-back task (M

0−back

= 28.6, SD

0−back

= 12.9). Although the interaction between workload

and stress did not reach signiﬁcance (F(1,21) = 2.345,

p = 0.141), an exploratory post-hoc paired t-test (t(21)

= 2.7, p = 0.013) showed greater perceived mental ef-

fort during low workload when exposed to psychoso-

cial stress than during ’relaxation’ condition.

3.2 Objective Indicators

3.2.1 Performance

For the analysis of task performance the accuracy for

each task block (see Figure 4(a)) was submitted to a 2

(stress) × 2 (workload) ANOVA. Two datasets were

excluded due to incompleteness. As for the subjective

indicators of perceived arousal and effort, we found

a main effect of workload manipulation (F(1,21) =

65.251, p ¡ 0.0001, η

= 0.757) with higher accuracy

for the simple 0-back task (M

0−back

= 97.3, SD

0−back

= 2.0) compared to the hard 2-back task (M

2−back

91.1, SD

2−back

= 4.8).

3.2.2 Physiological Sensors

We submitted mean skin conductance level and heart

rate for each block to a 2 (stress) × 2 (workload)

ANOVA. Four datasets were excluded due to incom-

pleteness. For heart rate analysis a further dataset

was excluded due to its low signal quality. As for

GSR (see Figure 4(b)), we found an increase of skin

conductance (F(1,19) = 4.4806, p = 0.048, η

0.191), indicating higher sympathetic arousal during

the ”stress condition” (M

= 3.83 , SD

= 2.05) com-

pared to the ”relax” condition (M

= 3.52, SD

2.07). Skin conductance was also observed to in-

crease (although not signiﬁcantly) during the high

workload condition compared to the low workload

condition. For HR (see Figure 4(c)), we found a trend

towards an increase of heart rate (F(1,18) = 3.2123,

p = 0.089, η

= 0.151), indicating higher sympa-

thetic arousal during the stress condition (M

= 79.41

, SD

= 10.23) compared to the relax condition (M

= 78.30, SD

= 10.08). Furthermore, we found a

highly signiﬁcant effect of workload manipulation on

HR (F(1,18) = 36.1431, p ¡ 0.0001, η

= 0.667),

with a higher heart rate for the more challenging 2-

back task (M

2−back

= 80.4, SD

2−back

= 9.89) com-

pared with the relatively easy 0-back task (M

0−back

= 77.27, SD

0−back

= 10.19).

4 DISCUSSION

We set out to devise an experimental protocol al-

lowing to manipulate psychosocial and mental stress

independently and presented evidence for its valid-

ity. The score of the STAI (anxiety state inven-

tory) showed a signiﬁcant response to psychosocial

stress manipulation, as STAI scores were signiﬁcantly

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

104

higher in the ’stress’ condition compared to ’base-

line’ and ’relaxation’ conditions. Given that the form

was ﬁlled in at the end of the ’stress’ and ’relax-

ation’ conditions, after the workload tasks, the still

increased scores in the ’stress’ condition are evidence

for a high psychosocial stress level throughout the

whole ’stress’ condition. Furthermore, regarding the

arousal scale of the SAM, we found no evidence for

a higher perceived arousal during the ’stress’ condi-

tion. The lacking effect of the stress manipulation

on that scale might be explained by a domination of

the workload manipulation on arousal, as it was given

with the rating scale on mental effort after each task.

Furthermore, a post-hoc analysis showed that in the

low-workload condition (i.e., 0-back task), perceived

mental effort (on the RSME scale) was higher in the

’stress’ condition than in the ’relaxation’ condition.

This result suggests that perceived mental effort in-

creases with psychosocial stress level, and thus in-

dicates that more cognitive resources are needed to

perform a task in a highly stressful situation. It is

in line with the dynamic adaptability theory of stress

(Szalma and Teo, 2012) which states that psychoso-

cial stress induces an increase in perceived workload.

The fact that this pattern is not observable in the high

workload condition (i.e., 2-back task) may be due to

a ceiling effect: the fact that the task itself induces

a high level of perceived mental effort decreases the

probability of increased mental effort due to psy-

chosocial stress. This ceiling effect could be respon-

sible of the apparent (but not signiﬁcant) interaction

in Figure 3(c). Regarding the mental stress manipu-

lation, the arousal scale showed increased arousal for

the high workload condition, mirroring the higher per-

ceived mental effort for this condition on the RSME

scale. Moreover, the performance decreased for the 2-

back task. These results deliver further evidence for

the validity of the mental stress manipulation, since

they show that the 2-back task is much more difﬁcult

(requires a signiﬁcantly highler workload level) than

the 0-back task.

Furthermore, the physiological signals delivered

evidence for the validity of the psychosocial as well as

of the mental stress manipulation. The elevated skin

conductance level and a trend towards a higher heart

rate during the stressful task-context corroborated the

effect found for the subjective stress ratings. Finding

elevated sympathetic nervous system activity after the

stress induction phase (TSST) and throughout the task

period is strong evidence that we were able to stretch

the phase of increased psychosocial anxiety over the

interview itself, using instructions and camera setup

(Hellhammer and Schubert, 2012). Furthermore, a

higher heart rate during difﬁcult (2-back) compared

to simple (0-back) task conditions supported the va-

lidity of the mental stress induction via task difﬁculty.

Finally, we found no signiﬁcant interaction effects,

neither for subjective (STAI, arousal, RSME) nor for

objective indicators (heart rate, GSR). This is conve-

nient, as it practically excludes the possibility of dif-

ferences between conditions others than the intended

ones (i.e., psychosocial and mental stress) and thus

may indicate that the goal has been achieved: we de-

signed and validated a stress induction protocol allow-

ing to vary mental and psychosocial stress indepen-

dently.

5 CONCLUSIONS

The presented results show that we managed to design

and validate a stress induction protocol which allows

the experimenter to vary mental stress and psychoso-

cial stress independently. However, although we de-

vised a strict protocol to avoid confusion with other

behavioural parameters as well as order effects, this

protocol remains very simple and not very ecologi-

cal. That is why future work will focus on continuous

classiﬁcation (instead of only two classes: stress/relax

and easy/difﬁcult) which would be more relevant for

real-life applications. Finally, two other goals will

have to be achieved : 1) EEG data analyses will be

performed in order to study the feasability of an EEG-

based (mental and psychosocial) stress classiﬁcation;

2) the ﬁnal goal of this project is to design and val-

idate a load-invariant physiology-based stress detec-

tion protocol, which will be used to help people man-

age their stress at home and at work.

REFERENCES

Boucsein, W. (1992). Electrodermal Indices of Emotion

and Stress. In Electrodermal Activity, chapter 3, pages

369–391. Springer New York.

Brouwer, A. M., Hogervorst, M. A., van Erp, J. B. F., Heffe-

laar, T., Zimmerman, P. H., and Oostenveld, R. (2012).

Estimating workload using EEG spectral power and

ERPs in the n-back task. International Journal of Neu-

ral Engineering, 9.

Buchanan, T. W., Etze, J. A., Adolphs, R., and Tranel, D.

(2006). The inﬂuence of autonomic arousal and se-

mantic relatedness on memory for emotional words.

International Journal of Psychophysiology, 61:23–26.

Crost, N. W., Pauls, C. A., and Wacker, J. (2008). Defen-

siveness and anxiety predict frontal EEG asymmetry

only in speciﬁc situational contexts. Biological Psy-

chology, 78:43–52.

Dickerson, S. S. and Kemeny, M. E. (2004). Acute stres-

sors and cortisol responses: a theoretical integration

DesignandValidationofaMentalandSocialStressInductionProtocol-TowardsLoad-invariantPhysiology-basedStress

Detection

105

and synthesis of laboratory research. Psychological

Bulletin, 130(3):355–391.

Fairclough, S. H. (2009). Fundamentals of physiological

computing. Interacting with Computers, 21:133–145.

Healey, J. and Picard, R. (2005). Detecting Stress During

Real-World Driving Tasks Using Physiological Sen-

sors. IEEE Transactions on Intelligent Transportation

Systems, 6(2):156–166.

Hellhammer, J. and Schubert, M. (2012). The physiological

response to Trier Social Stress Test relates to subjec-

tive measures of stress during but not before or after

the test. Psychoneuroendocrinology, 37(1):119–24.

Hewig, J., Schlotz, W., Gerhard, F., Breitenstein, C.,

Luerken, A., and Naumann, E. (2008). Associations

of the cortisol awakening response (CAR) with corti-

cal activation asymmetry during the course of an exam

stress period. Psychoneuroendocrinology, 33:83–89.

Hogervorst, M. A., Brouwer, A.-M., and Vos, W. K.

(2013). Physiological Correlates of Stress in Indi-

viduals About to Undergo Eye Laser Surgery. In 5th

IEEE International Conference on Affective Comput-

ing and Intelligent Interaction, Geneva, Switzerland.

IEEE Computer Society Press.

Khosrowabadi, R. and Quek, C. (2011). A Brain-Computer

Interface for classifying EEG correlates of chronic

mental stress. In Proceedings of International Joint

Conference on Neural Networks 2011 (IJCNN), vol-

ume 138632, pages 757–762.

Kirchner, W. K. (1958). Age differences in short-term re-

tention of rapidly changing information. Journal of

Experimental Psychology, 55(4):352–358.

Kirschbaum, C., Pirke, K.-M., and Hellhammer, D. H.

(1993). The ”Trier Social Stress Test”: a Tool for

Investigating Psychobiological Stress Responses in a

Laboratory Setting. Neuropsychobiology, 28:76–81.

Krout, R. E. (2007). Music listening to facilitate relaxation

and promote wellness: Integrated aspects of our neu-

rophysiological response to music. The Arts in Psy-

chotherapy, 34:134–141.

Kusserow, M., Amft, O., and Troster, G. (2012). Monitoring

Stress-arousal in the Wild - Measuring the Intangible.

IEEE Pervasive Computing, 12(2):28 – 37.

Lewis, R. S., Weekes, N. Y., and Wang, T. H. (2007).

The effect of a naturalistic stressor on frontal EEG

asymmetry, stress, and health. Biological Psychology,

75:224–239.

Loggia, M. L., Juneau, M., and Bushnell, M. C. (2011).

Autonomic responses to heat pain: Heart rate, skin

conductance, and their relation to verbal ratings and

stimulus intensity. Pain, 152(3):592–8.

Regehr, C., Glandy, D., and Pitts, A. (2013). Interventions

to reduce stress in university students: A review and

meta-analysis. Journal of Affective Disorders, 148:1–

11.

Reinhardt, T., Schmahl, C., W

ust, S., and Bohus, M. (2012).

Salivary cortisol, heart rate, electrodermal activity and

subjective stress responses to the Mannheim Multi-

component Stress Test (MMST). Psychiatry research,

198(1):106–11.

Renard, Y., Lotte, F., Gibert, G., Congedo, M., Maby, E.,

Delannoy, V., Bertrand, O., and L

ecuyer, A. (2010).

OpenViBE: An Open-Source Software Platform to

Design, Test, and Use Brain-Computer Interfaces in

Real and Virtual Environments. Presence: Teleopera-

tors and Virtual Environments, 19(1):35–53.

Riera, A., Soria-Frisch, A., Albajes-Eizagirre, A., Ci-

presso, P., Grau, C., Dunne, S., and Rufﬁni, G.

(2012). Electro-physiological data fusion for stress

detection. Studies in health technology and informat-

ics, 181:228–32.

Selye, H. (1936). A syndrome produced by diverse nocuous

agents. Nature, 138:32.

Sinha, R., Talih, M., Malison, R., Cooney, N., Anderson,

G.-M., and Kreek, M.-J. (2003). Hypothalamic-

pituitary-adrenal axis and sympatho-adreno-

medullary responses during stress-induced and

drug cue-induced cocaine craving states. Psychophar-

macology, 170:62–72.

Spielberger, C. D., Gorsuch, R. L., and Lushene, R. E.

(1970). Manual for the State-Trait Anxiety Inventory.

Consulting Psychologists Press, Palo Alto.

Szalma, J. L. and Teo, G. W. L. (2012). Spatial and tem-

poral task characteristics as stress: A test of the dy-

namic adaptability theory of stress, workload, and per-

formance. Acta Psychologica, 139:448–471.

Taniguchi, K., Nishikawa, A., Sugino, T., Aoyagi, S., Seki-

moto, M., Takiguchi, S., Okada, K., Monden, M., and

Miyazaki, F. (2009). Method for Objectively Eval-

uating Psychological Stress Resulting when Humans

Interact with Robots. In Kulyukin, V. A., editor, Ad-

vances in Human-Robot Interaction. InTech China.

Tops, M., van Peer, J. M., Wester, A. E., Wijers, A. A., and

Korf, J. (2006). State-dependent regulation of corti-

cal activity by cortisol: An EEG study. Neuroscience

Letters, 404:34–39.

van den Broek, E. L. and Westerink, J. H. D. M. (2012).

Biofeedback Systems for Stress Reduction - Towards

a Bright Future for a Revitalized Field. In Interna-

tional Conference on Health Informatics (HEALTH-

INF 2012), pages 499–504.

PhyCS2014-InternationalConferenceonPhysiologicalComputingSystems

106