Flow Neurophysiology in Knowledge Work: Electroencephalographic

Observations from Two Cognitive Tasks

Michael T. Knierim

, Mario Nadj

, Anuja Hariharan

1,2

and Christof Weinhardt

Institute of Information Systems & Marketing, Karlsruhe Institute of Technology, Fritz-Erler-Str. 23, Karlsruhe, Germany

Opitz Consulting, Leitzstr. 45, Stuttgart, Germany

Keywords: Flow, Neurophysiology, Knowledge Work, Physiological Computing, Bio-Adaptive Systems.

Abstract: In an effort to study flow experiences in the context of less structured knowledge work (KW), we explored a

paradigm we call controlled experience sampling (cESM). Participants worked on a naturalistic, cognitive

task (a personal scientific thesis), and a difficulty-manipulated math task. Results show that the cESM

approach elicits a consistent flow experience with intensities as least as high as in the math task flow condition.

An interesting finding is that given similar flow intensities, different perceptions of stress arise between the

two paradigms. EEG results from both tasks suggest increased frontal upper alpha band (10-12Hz) activity

with increased task attention, that has higher temporal stability in flow than in a boredom condition, and that

is laterally indifferent. Integrating with the presently available literature, the results further consolidate an

understanding of flow as a state of fronto-lateral activation.

1 INTRODUCTION

The experience of flow, where the individual is

completely involved in a challenging task

(Csikszentmihalyi, 1996), is deemed a beneficial state

in the work environment due to its links to improved

performance and well-being (Spurlin and

Csikszentmihalyi, 2017). As the requirements for

flow are complex (e.g. absence of distractions,

structure of the task, state of the individual, etc.) (Ceja

and Navarro, 2012), flow facilitation at work is still a

central challenge (Spurlin and Csikszentmihalyi,

2017). However, the recent advancements on the

biological basis of flow (Harris et al., 2017; Knierim

et al., 2017) highlight promising avenues for

supportive bio-adaptive systems (Rissler et al., 2018).

Within the emerging research a central focus lies on

highly controlled game tasks (Moller et al., 2010),

leaving gaps to understand flow neurophysiology in

more unstructured tasks typical to knowledge work

(KW). Furthermore, the focus on artificial laboratory

tasks has been argued to be a central limitation in

studying the (flow) experience of effortless attention

(Hommel, 2010). Therefore, in an attempt to increase

external validity and naturalistic character of flow

laboratory research we propose the adaption of the

experience sampling method (ESM)

(Csikszentmihalyi and Hunter, 2003) to the

laboratory setting. This adaption signifies a controlled

approach (cESM) prompting individuals to work on a

personalized KW task during observation with

neurophysiological sensors and through repeated

interruption in order to “catch flow in the act”. By

comparing observations to a validated flow induction

paradigm, the main research question of how well the

cESM approach can elicit flow is to be answered.

Furthermore, while there have been serious

advancements in the field of brain-computer

interfaces that keep extending the applicability of

real-time neuroimaging to in situ phenomena like

attention, operator workload and engagement

(Blankertz et al., 2016; Ewing et al., 2016; Kosti et

al., 2018), the study of neural correlates of flow in

more externally valid scenarios is still sparse

(Katahira et al., 2018). In general, the knowledge of

how flow can be described using neural measures still

lacks of repeated insights, which is why this work fills

several important gaps. Overall, our work contributes

to flow research by: (1) advancing the understanding

of flow elicitation in laboratory settings in the context

of KW, and by (2) extending flow neurophysiology

knowledge by consolidation of related work, across

task analysis and study of high interest brain regions.

Knierim, M., Nadj, M., Hariharan, A. and Weinhardt, C.

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks.

DOI: 10.5220/0006926700420053

In Proceedings of the 5th International Conference on Physiological Computing Systems (PhyCS 2018), pages 42-53

ISBN: 978-989-758-329-2

2 BACKGROUND

2.1 Flow Theory

Flow research spans contexts like arts (de Manzano et

al., 2010), gaming (Moller et al., 2010; Harmat et al.,

2015), or writing (Csikszentmihalyi, 1996; Erhard et

al., 2014) and has found the state to occur remarkably

similar across contexts. The experience is described

in nine dimensions, that are classified temporally (see

Table 1).

Table 1: Flow Experience Components (cf. Nakamura and

Csikszentmihalyi, 2009).

Component Class

1) Challenge-skill balance Antecedents

2) Clear Goals

3) Unambiguous Feedback

4) Action-Awareness Merging During

Experience

5) Sense of Control

6) Loss of Self-Consciousness

7) Transformation of Time

8) High Concentration

9) Autotelic experience Consequences

2.2 Research Paradigms

In the past, primarily self-reports have been used to

develop flow descriptions (interviews & surveys)

(Moneta, 2012). The ESM was specifically designed

for this purpose, in order to surpass interview

limitations (e.g. recall bias) and to study flow close to

its occurrence through repeated interruption

(Csikszentmihalyi and Hunter, 2003). The naissance

of experimental flow induction has only later and

recently occurred, focusing on the paradigm of

difficulty manipulation (DM) (Moneta, 2012). The

manipulation of difficulty is used to elicit experiences

of boredom, flow, and anxiety (through

low/balanced/high difficulty). It has been criticized

whether or not the approach can elicit real flow

experiences, given the reduced motivation and

involvement common in laboratory tasks (Moller et

al., 2010), and given the elicitation of effortful

attention from novel and artificial tasks (Hommel,

2010). Other approaches have focused on

engagement (E) paradigms where participants are for

example asked to play a game and report their

experience afterwards (e.g. Labonté-Lemoyne et al.,

2016). Flow physiology research has extensively used

the DM paradigm. Based on a previous survey of 20

studies on the peripheral nervous system (PNS)

(Knierim et al., 2017) and three studies published

since then (Klarkowski, 2017; Tian et al., 2017;

Barros et al., 2018), we found that 13 of 23 studies

used this paradigm. Furthermore, 17 of 23 studies

used game tasks, a pattern similarly visible in flow

neuroimaging research (see next section). This shows

a focus with low external validity, that has led to calls

for creative laboratory research (Harris et al., 2017).

2.3 Flow Neurophysiology

Given the youth of experimental paradigms, flow

research has only recently produced increased

amounts of theoretic and empiric research on

underlying neurophysiological processes (Peifer,

2012; Harris et al., 2017). One of the first

propositions of flow neurophysiology is the reduction

of prefrontal cortex activity during flow in favour of

more implicit, automated processing of learned

behaviour (Transient Hypofrontality = TH) (Dietrich,

2004). Extending this proposition, linear reductions

of default mode network activity with flow

experience have been put forward that would

alternatively explain the experience of automaticity

and the absence of self-referential processing (Peifer,

2012; Harris et al., 2017). Furthermore, the

proposition of flow as an emergent property of highly

synchronized activity in attention and reward

networks of the brain has been highlighted

(Synchronization Theory = ST) (Weber et al., 2009;

Harris et al., 2017).

Extending the aforementioned literature review

corpus, several peer-reviewed studies of flow

neurophysiology were identified. Much of this

research has focused on hemodynamic imaging (e.g.

Ulrich et al., 2014; Harmat et al., 2015; Barros et al.,

2018). Also, there has of late been an increase in

electroencephalographic (EEG) flow research (Wolf

et al., 2015; Ewing et al., 2016; Katahira et al., 2018).

Yet there has been little consolidation of these lines

of work. For this report we decided to focus on results

on frontal brain regions, as the study of frontal

regions has been preferred often based on the early

TH account. (Dietrich, 2004). So far, for TH’s main

hypothesis of overall frontal activity reduction during

flow, little support has been found in fMRI (Ulrich et

al., 2014) and fNIRS (Harmat et al., 2015; Barros et

al., 2018) imaging studies. Instead, it appears parts of

the prefrontal cortex (PFC), specifically lateral parts,

are highly active during flow, yet the medial PFC

shows activity decreases during flow, and boredom

conditions show a more general PFC reduction

(Harris et al., 2017; Barros et al., 2018).

Frontal activity has also been reported on in most

of the related flow EEG studies, with repeated results

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks

supporting the region as a location of interest. While

two of these studies (Chanel et al., 2011; Berta et al.,

2013) report on the relevance of frontal activity for

the machine-learning (ML) based classification of

flow states, six other studies describe activity in more

detail (see Table 2). Aggregating the results of these

studies that primarily focused on frequency band

activity across difficulty-manipulated conditions, we

conclude several results that have not been integrated:

(1) One of the more clear findings is a difference

in frontal theta band activity in flow conditions, with

increases compared to boredom task conditions, and

either similarity between flow and overload

conditions (Soltész et al., 2014; Katahira et al., 2018)

or decreases from flow to overload conditions

indicating an inverted U-shaped relationship between

frontal theta activity and task demands (Ewing et al.,

2016). Support for theta band differentiation has also

been noted in ML research on flow classification

(Chanel et al., 2011).

(2) Repeated observations have been made for

frontal alpha activity. While some find increased

alpha power with higher flow self-reports (Léger et

al., 2014; Labonté-Lemoyne et al., 2016), within the

DM group comparison studies, findings point more to

decreases in alpha activity with increasing task

difficulty (Ewing, Fairclough and Gilleade, 2016;

Katahira et al., 2018 report the inverse relationship,

but use amplitudes as unit of analysis). ML research

also finds frontal alpha activity to be a differentiating

feature (Berta et al., 2013).

(3) Lastly, a few observations have also been

made regarding frontal beta band activity in flow,

with ML reports demonstrating differentiation

potential alone (Chanel et al., 2011; Berta et al.,

2013), left frontal beta band reductions correlated

with higher flow self-reports (Léger et al., 2014), but

also right frontal beta band increases with task

difficulty (Klarkowski, 2017).

Within this body of research, additional interesting

EEG observations have been mentioned that are lateral

differences, specifically frontal alpha asymmetry

(FAA) (Wolf et al., 2015; Labonté-Lemoyne et al.,

2016), frequency band separation, (e.g. individualized

theta, alpha band and beta band splits) (Berta et al.,

2013; Soltész et al., 2014; Ewing et al., 2016), and also

temporal differences of frequency band activity within

task conditions (Soltész et al., 2014). In this study, we

followed up on several of these in favor of an in-depth

analysis of frontal activity patterns.

3 METHOD

3.1 Materials & Procedure

Overall, 12 students (3 female) ages 21-30

participated voluntarily in our laboratory study. Each

participant worked on (1) writing for a scientific

thesis, and on (2) solving math equations in

manipulated difficulties. Scientific writing was

chosen for its challenging and frequent task nature for

scholars and students (exemplary KW). Also, writing

(scientific or literary) has previously been related to

engaging experiences in general and flow in

particular (Csikszentmihalyi, 1996; Erhard et al.,

2014; Galluch et al., 2015).

Table 2: Frontal EEG results in related work (Legend: Par. = Paradigm, Anal. = Type of analysis, (Q-)Com. = (Quasi-)

Condition comparison, Regr. = Regression EEG & self-reports, Un. = Unit of analysis, µ = Frequency amplitude, µ

Frequency power. Exemplary explanation of symbols: = No significant differences, = Boredom condition significantly

different from flow & overload condition, = Positive, linear relation of frequency band and self-report).

Reference Par. Anal. Un. Frontal Electrodes Bands & Ranges & Findings (Frontal sites only)

Left Mid Right

lo-α hi-α

lo-

hi-

Katahira et al.,

2018

DM Com. µ

(AF3,F3,

F7,FC3)

(Afz,F

z,FCz)

(AF4,F4,

F8,FC4)

4-7

10-13

14-30

Ewing et al., 2016 DM Com. µ

4-7 7,5-10

10,5-13

Klarkowski, 2017 DM Com. µ

AF4

4-8

8-13

13-30

Soltész et al., 2014 DM Com. µ

(Fp1,Fp2,F3,F4,F7,F8,Fz)

4-8 8-11 11-13

13-25

25-35

Labonté-Lemoyne

et al., 2016

Q-

Com.

(F4,F8)

4-7

8-12

13-30

Léger et al., 2014 DM Regr. µ

8-12

12-22

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

Participants brought their own, active thesis project

(bachelor or master level) to work on for a session of

20-25 minutes. Initially they were given time to

inspect the state of their document and to define a

challenging, yet achievable goal for a writing session.

To standardize the goal setting approach the SMART

mnemonic was used (Doran, 1981). This approach

was also considered to facilitate the flow experience.

For example, setting a goal that is specific (S) (i.e.

less abstract) has been found to facilitate high quality

writing outcomes (Flower and Hayes, 1981), and

should further provide on one of the flow pre-

requisites of having a clear goal. Deriving a goal

attainment measure (M), was considered to be helpful

in fulfilling the second flow pre-requisite of

unambiguous feedback. Lastly, the focus on a

relevant (R) and achievable (A) goal, was considered

to further enhance the optimality of a task challenge.

The thesis writing software was standardized to

Microsoft Word in full-screen mode.

The math task was chosen as reference to a

validated DM task (Ulrich et al., 2014; Katahira et al.,

2018). Replicating the design by Ulrich et al., (2014),

participants sum two or more numbers. Two

adjustments were made to the design as task difficulty

was found to be too high in previous tests. The

boredom condition was adjusted so that, subjects

solved randomly drawn equations in one of three

forms (101 + 1, + 2, or + 3). The flow condition was

altered so that, difficulty was increased/decreased

when three sequential responses were

correct/incorrect. There was a constant waiting period

between trials of four seconds. The math and writing

task order was randomized. Also, the three math task

conditions were ordered randomly, which resulted in

a total count of 12 procedure variations (2 * 3!

combinations). All variations were executed once. At

the start of the experiment, participants completed

eyes-open and eyes-closed baseline phases in which

they were asked to “let their mind wander to wherever

it takes them”, to keep their eyes focused on a black

fixation cross on a white screen (in the eyes-open

phase), and to avoid unnecessary movements. The

same message and fixation cross were shown for the

washout screens prior to each math task condition and

between math and writing task. The complete

procedure is outlined in Figure 1. In the recruitment

survey participants reported mean thesis challenge

levels of 4.3 (SD: 0.98) and Wilcoxon comparisons

showed no difference in preference for writing or

math tasks (measured using three questions from

Ulrich et al., 2014).

3.2 Measures

Round questionnaires contained scales on flow and

task demand (ten item Flow Short Scale (FKS) and

one additional task demand question all by Engeser

and Rheinberg, 2008), stress (five item construct by

Tams et al,. 2014), and affect (single question arousal

SAM scale by Bradley and Lang, 1994), amongst

others. Between-task surveys included scales on task

importance (Engeser and Rheinberg, 2008). Almost

all questions used 7-p Likert scales (SAM arousal

used 9-p). Additionally, ECG data was collected in

Lead II configuration using gelled electrodes. EDA

data was collected using gold cup electrodes on the

left foot. However, we focus in this report on the

analysis of EEG data only. EEG data was collected

with an Emotiv Epoc+ headset. This 14-channel

wireless headset uses saline-based electrodes,

collecting data at a sampling rate of 256Hz. Electrode

sites are: AF3, F3, F7, FC5, T7, P7, O1, O2, P8, T8,

FC6, F8, F4, AF4 (10-20 system). Two reference

electrodes, the ‘‘common mode sense’’ (CMS) and

‘‘driven right leg’’ (DRL) are placed on the left and

right mastoids. While the headset comes with

downside regarding data quality, it has been found to

deliver adequate data for our type of study (Barham

et al., 2017) and has been used in previous studies

related to the KW context (Kosti et al., 2018), and to

Figure 1: Experiment procedure.

Preparation

Conclusion

Rest Eyes Open (5min)

Rest Eyes Closed (1min)

Introduction, Sensor

Attachment, & Start Survey

Exit Survey, Sensor

Removal, & Debriefing

Math  OR (Randomized)  Writing

Orientation (2min)

Goal Setting (Open)

Writing Round (3x)

Round Survey

Task Survey

Round Survey

Boredom/Flow/Overload

Randomized (5min)

Washout Screen (60s)

Orientation (3min)

Calibration (4min)

Task Survey

Math Round (3x)

Writing (7min)

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks

flow experiences (Klarkowski, 2017). Prior to

application of the headset, the felt-pad electrodes

were moistened with a standard 0,9%-NaCl saline

solution. After application, acceptable contact quality

was controlled for all electrode sites using the

proprietary impedance information supplied by the

manufacturer’s API.

3.3 Psychometric Data Preparation

Outliers were removed for all psychometric variables

(>2 SD from the construct mean). Distribution

normality (Shapiro-Wilk test) and homogeneity of

variances (Fligner-Killen test) did not hold for many

samples (p < 0.05), prompting non-parametric test

use. Removing one item from the stress construct

improved Cronbach’s Alpha in the math boredom

condition. Further item removal did not improve

Alpha values, which left the stress construct Alpha

level bordering a critical value of 0.6 that is deemed

acceptable in some cases (Hair et al., 2011). Given the

high internal consistency across phases (see Table 3),

and corroborating results from the arousal item, the

stress construct was retained.

Table 3: Cronbach’s Alpha levels per experiment phase (m-

B/F/O/All = Math Boredom/Flow/Overload/All conditions,

w1-3/All = Writing round 1-3/All rounds).

mB mF mO mAll w1 w2 w3 wAll

Flow 0.68 0.88 0.89 0.80 0.77 0.98 0.93 0.95

Stress 0.84 0.58 0.78 0.84 0.84 0.78 0.90 0.82

3.4 EEG Data Pre-Processing

EEG data was processed along the guidelines of Cohen

(2014) and Picton et al. (2000). Data was processed

only for a homogenized sub-sample (three female

participants were excluded) (Picton et al., 2000). Also,

two data sets had to be excluded due to recording

failure. The retained sample for EEG analysis

comprised 7 right-handed males. Data preparation,

feature extraction, and analysis were conducted in R,

signal processing and artefact removal in EEGLab

(Ver. 14.1.1). Initially, experiment phases of interest

were extracted for each participant (eyes open baseline,

all three math task conditions, all three writing task

rounds). Channels were first centred through mean

subtraction. Afterwards, the extracted data was loaded

into EEGLab where a 0.5-45Hz bandpass, and a 50Hz

notch filter were applied. Signal data was then visually

inspected for artefact removal. First, channels that had

failed to collect data were removed. Then, paroxysmal

artefacts were removed manually. Afterwards, using

the infomax algorithm, an independent component

analysis (ICA) was performed to identify and remove

components of the data related to eye blinks and

sideway saccades (EOG artefacts). Next, data was re-

imported in R in order to extract frequency band

information for the frontal electrodes (AF3, F3, F7,

FC5, FC6, F8, F4, AF4) similar to (Ewing et al., 2016)

on the basis of 2s long epochs with 50% overlap and

tapered using a Hann windowing function. Average

band power (µV

) was extracted using the Fast Fourier

Transformation (FFT). Only artefact-free and

complete epochs were used for feature extraction

(epochs containing more than 95% of required

samples, i.e. > 2s * 256Hz = 512 samples). Extracted

frequency bands are: Theta (4-8Hz), Alpha (8-12Hz),

and Beta (12-30Hz). Also, for the Alpha and Beta band

additional sub-segments were extracted that are low

Alpha (8-10Hz), high Alpha (10-12Hz), low Beta (12-

15Hz), mid Beta (15-20Hz), and high Beta (20-30Hz).

Afterwards, frequency band data was normalized (Ln

transformation). Electrodes were pooled by computing

the mean for three regions of interest that are all frontal

sites (AF3, F3, F7, FC5, FC6, F8, F4, AF4), left frontal

sites (AF3, F3, F7, FC5), and right frontal sites (FC6,

F8, F4, AF4) for each epoch. Next, feature epochs were

aggregated temporally by computing the median over

each experiment phase. Median use was preferred as a

way of conservative data interpretation, taking care of

potential outliers. Finally, to facilitate comparisons

between experiment phases, change scores were

computed by subtracting the eyes open baseline phase

mean from each experiment phase (e.g.

∆theta

theta

Task

– theta

Baseline

). For an additional analysis of

temporal segments of each experiment phase, the same

procedure outlined above was repeated on 30s-long

epochs within each phase. The window length of 30s

was chosen based on the report by Soltész et al. (2014)

who argue that at the start of phases temporal

differences could occur in this interval already.

Distribution tests indicated that assumption of

normality was violated for many groups (Shapiro-Wilk

on the temporally and spatially aggregated data sets for

each condition and frequency band, p < 0.05), which is

why non-parametric tests were used afterwards.

Fligner-Killeen tests showed no violation of variance

homogeneity assumptions.

4 RESULTS

We report on four psychometric (flow, task demand,

stress, arousal) measures together with multiple EEG

features compared across six experiment phases

(three math conditions, three writing rounds). Beyond

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

statistical comparisons, the reduced number of

samples in the EEG data prompted us to include

additional descriptive analyses. We believe this

approach also has merit in light of the young age of

the research on flow neurophysiology. The

descriptive approach has more of a case study

character, a format that has previously been employed

for flow PNS measures (Harmat et al., 2011).

4.1 Psychometric Data

Friedman tests indicated the presence of main effects

in psychometric variables at significant levels (p <

0.01). Variable means and standard deviations are

shown in Table 4, and post-hoc pairwise Wilcoxon

comparisons of experiment phases in Table 5.

The task demand variable was inspected for a

manipulation check (cf. Keller et al., 2011; Tozman

et al., 2015). Between all math conditions significant

differences were found, displaying increasing task

demand from boredom to overload conditions. DM

success was thereby confirmed. Within the writing

samples task demand levels lay continuously between

the math boredom and overload condition. Possibly

task demand in writing was also lower than in the

math flow condition (trend level indication). No

differences were found within the writing task for

task demand, with the exception of one trend level

difference between writing round 1 and 3. Also, no

differences were found in all other psychometric

variables across writing rounds and are therefore not

reported further.

Within the math task, flow report (FKS)

comparisons show significant differences between

the math flow and overload condition. Also, repeated

significant differences between the math boredom

and overload conditions with the writing rounds are

found. Lastly, a trend level indication is visible for

higher reported flow in writing round 1 compared to

the math flow condition. As there are no significant

differences within the writing task, flow was reported

as high in writing as in the math flow condition in all

writing rounds. Support for this consistency is also

visible in the range of flow reports per participant

(mean range = 1.13, SD = 0.62, writing task only).

The stress report comparison showed significant

differences between all math task conditions,

increasing with difficulty at every step. In the writing

task, stress levels were consistently below the math

flow and overload conditions. Comparisons of the

arousal reports reveal a similar pattern, with

increasing arousal from math boredom to overload

conditions. Albeit only with a significant difference

for the boredom condition with the other two. Like

stress, arousal was consistently reported lower in

writing than in math flow and overload conditions.

Finally, after both tasks, participants rated the

importance of the task. No significant differences

were found (Means: math = 3.82, writing = 4).

4.2 EEG Data

4.2.1 Results Between-Phase Comparisons

Friedman tests were computed for each feature

(pooled sites) and frequency band to detect main

effects across experiment phases. A main effect was

found only for the hiAlpha band (p < 0.05). No

different effects were found for either the left side or

right side alone, which is why the analysis of

hemispheric differences was not pursued further.

Table 4: Psychometric variable means & standard deviations (in parentheses) across experiment phases.

mB mF mO w1 w2 w3

Flow 4.03 (0.80) 4.53 (1.16) 4.02 (1.21) 5.43 (0.59) 4.93 (1.55) 5.09 (1.14)

Stress 2.96 (1.32) 4.18 (0.74) 4.75 (1.18) 2.81 (1.26) 2.79 (1.02) 2.64 (1.07)

Arousal 2.73 (1.19) 6.18 (0.98) 6.27 (1.27) 3.33 (1.50) 2.82 (1.08) 3.91 (1.64)

Demand 1.45 (0.93) 5.18 (0.60) 6.09 (0.70) 4.42 (0.90) 4.17 (1.19) 3.73 (1.01)

Table 5: P-values psychometric & EEG data pairwise Wilcoxon tests across experiment phases.

Demand Flow Stress Arousal ∆hiAlpha

mB mF mO mB mF mO mB mF mO mB mF mO mB mF mO

mF <.01 >.1 <.05 <.01 <.1

mO <.01 <.05 >.1 <.05 <.01 <.05 <.01 >.1 <.05 >.1

w1 <.01 <.1 <.01 <.01 <.1 <.01 >.1 <.05 <.01 >.1 <.01 <.01 <.05 <.1 <.1

w2 <.01 <.1 <.01 >.1 >.1 <.05 >.1 <.01 <.01 >.1 <.01 <.01 <.1 >.1 >.1

w3 <.01 <.05 <.01 <.05 >.1 <.1 >.1 <.01 <.01 <.1 <.01 <.01 <.05 >.1 <.1

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks

Post-hoc pairwise Wilcoxon tests were conducted

on the remaining “all frontal” feature hiAlpha

frequency band (see Table 5). Within the math task,

the hiAlpha band shows significantly higher levels in

the boredom condition than in the overload condition,

and the flow condition indicated on trend level, and

no difference between flow and overload condition.

Across tasks, the hiAlpha band activity in the math

boredom condition is significantly higher than in

writing rounds 1 and 3, and also higher than in writing

round 2, indicated at trend level. For the hiAlpha

band, trend level differences also indicate lower

hiAlpha in writing round 1 than in the math flow

condition, and lower hiAlpha in writing rounds 1 and

3 than in the math boredom condition.

To further deepen the analysis, phase medians

were descriptively compared. Difference thresholds

were defined conservatively by ascertaining that

classified results fit the previously described

Wilcoxon tests as medium to larger differences (e.g.

the significant hiAlpha difference between the math

boredom and math flow condition has a median

difference of 0.493). Therefore, differences between

0.15 and 0.30 are considered as smaller, between 0.30

and 0.45 as moderate, and above 0.45 as larger

differences. This process lead to 33.3% of math

comparisons, 0% of writing comparisons, and 37.5%

of across task comparisons being subject of

descriptive interpretation. Within the writing task, no

differences are found, indicating a consistent

experience. Within the math task, median

comparisons show higher hiAlpha in math boredom

compared to math flow and overload conditions (mB-

mF = 0.493, mB-mO = 0.356), a contrast that is

similarly visible in the broad alpha band, although

with smaller differences (mB-mF = 0.284, mB-mO =

0.277). Furthermore, the descriptive comparison

points to higher theta in the math flow than the math

boredom condition (mB-mF = 0.171), also to higher

hiBeta in math overload compared to math boredom

(mB-mO = 0.154) and also to higher beta in the math

overload condition compared to both boredom and

flow conditions (mO-mB = 0.2, mO-mF = 0.231), all

with smaller differences. Across tasks, the descriptive

data again shows increased hiAlpha in the math

boredom condition compared to all three writing

rounds (mB-w1 = 0.564, mB-w2 = 0.488, mB-w3 =

0.548) with larger differences. The same pattern is

visible for the broad alpha band (mB-w1 = 0.356,

mB-w2 = 0.332, mB-w3 = 0.394), albeit with

moderate differences, and the loAlpha band (mB-w1

= 0.231, mB-w2 = 0.175, mB-w3 = 0.186) with

smaller differences. Furthermore, the median

differences point to lower hiAlpha in writing rounds

1 and 3 than in the math overload condition (mO-w1

= 0.208, mO-w3 = 0.192) with smaller differences.

Also, hiBeta shows higher levels in writing round 1

than in math boredom and flow conditions (mB-w1 =

0.2, mF-w1 = 0.157), as does the broad beta band

(mB-w1 = 0.233, mF-w1 = 0.265), all with smaller

differences. Both the median levels and group

significance differences are visualized in Figure 2.

4.2.2 Results within-Phase Comparisons

To analyse the potential of temporal variation in

frequency band activity during flow (Soltész et al.,

2014), within experiment phase effects were

investigated. Friedman tests on 30s-based segments

of each experiment phase were computed (10/14

segments for each math/writing task phase).

Results show main effects for the alpha and

hiAlpha band in the math boredom condition and

writing round 3 (all p < 0.05), and for the beta and

midBeta band in writing round 1 (both p < 0.01). For

writing round 3 post-hoc pairwise Wilcoxon tests

revealed only a single significant difference in the

hiAlpha band (out of 91 comparisons), which is why

this finding is considered an anomaly. For writing

round 1 on the other hand, multiple significant

differences are found for midBeta (19/25) and beta

(20/23) (p < 0.05/0.1), with the most pronounced

differences for early vs. late segments, pointing to a

beta activity increase in the first minutes of writing

round 1. For the math boredom condition, multiple

significant differences are found for alpha (5/11) and

hiAlpha (11/14) (p < 0.05/0.1) (out of 45

comparisons). This pattern is more volatile with alpha

showing a difference of the first 30s to the mid part of

the boredom task round (alpha appears to peak

slightly in the first 1-3min), but hiAlpha shows

repeated differences between segment 1 and 3-6, then

again, a difference of segment 3 to segments 7-9, and

segment 6 to 7-9, indicating an early and late peak

(and a mid-part valley). No repeated start or end

effects were found in all bands and experiment

phases. Also, besides the beta pattern in writing round

1, the phases showing higher flow reports are more

strongly marked by consistency than volatility.

Significant differences are shown in Figures 3 & 4.

5 DISCUSSION

5.1 Psychometrics Findings

Within the writing task, all variables indicate

experience consistency, despite repeated interruption.

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

This is an important finding as interruptions are often

considered a central flow hindrance (Rissler et al.,

2017), which is why we anticipated more experiential

variance. Possibly, some factors in the writing task

design (like the goal setting procedure) dampened

such interruption impacts by providing structure.

Within the math task, our results show DM

success with results comparable to previous research,

showing that flow is reported most strongly when task

demands are balanced (Keller et al., 2011; Tozman et

al., 2015; Klarkowski, 2017).

The results are taken as first support that the

cESM approach (with this scientific writing design)

can be used to elicit flow, with at least similar

intensities compared to a standard DM paradigm (the

math task). However, a clear difference between the

two paradigms appears as writing is perceived to be

less stressful and demanding than the math task in

flow and overload conditions. A key reason for the

stress difference could be that per design multiple

stress factors present in the math task and typical DM

designs (task demand overload, social-evaluative

threat, lack of control) (Tozman et al., 2017) were not

present in the writing task. In the past, these stressors

have been purposefully introduced to DM designs in

order to elicit motivated task performances (Ulrich et

al., 2014; Tozman et al., 2015, 2017). At the same

time, in these approaches repeated sightings of

psychometric reports that point to increased

stress/arousal in balance and overload conditions

compared to boredom conditions have been made,

even in contexts where threat experiences could be

less likely (e.g. in gaming) (Harmat et al., 2015;

Tozman et al., 2015, 2017; Klarkowski, 2017). Our

results indicate, that a naturally important task

lacking these stressors, results in similar reported

flow intensities without perceptions of strain. It

would appear that the critique on the applicability of

the DM paradigm to elicit real flow could therefore

receive some support (Moller et al., 2010), as could

the proposition that naturalistic tasks are perceived as

less effortful (Hommel, 2010). However, these results

could also indicate a central limitation to how flow is

collected psychometrically.

5.2 EEG Findings

Within the writing task, EEG results mostly support

the view of a consistent experience across writing

trials. The only effect that shows variation is the

initial beta increase within the first part of writing task

round 1 (temporal analysis). Given that this variation

is not apparent in later phases, we believe it to be most

likely attributable to a type of task initiation activity.

Figure 2: Experiment phase frequency band activity over all

frontal electrodes pooled (y-axis = change scores of ln-

transformed avg. frequency band power). Median values of

each phase are listed beneath. Bars show Wilcoxon test

results with p < 0.05 (*) and p < 0.1 (t).

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks

Figure 3: Temporal variation math boredom condition.

Figure 4: Temporal variation writing round 1.

It has been reported in flow and writing research

(Flower and Hayes, 1981; Csikszentmihalyi, 1996),

that initiation of a writing session takes additional

effort to structure the task that may be required less at

later stages. Given that beta activity is often related to

increased excitatory cognitive activity, we believe

this findings shows an initially increase in cognitive

effort that dissipates after a while, and that is not

specifically related to flow experience or neural

correlates thereof, as something similar is not visible

in the math flow condition either.

Within the math task, EEG results integrate in

several ways with previous work. The finding of

frontal theta activity changes (descriptive analysis)

with difficulty increases is supported, yet only

weakly. This could be a spurious effect caused by our

small sample or point to a need to further specify theta

band activity (like Ewing et al., 2016 who select

individualized theta band activity from the 4-7Hz

range). The finding of lower hiAlpha activity with

increasing task difficulty (statistical & descriptive

analyses) is interesting in multiple ways. First, the

separation of the alpha band shows that hiAlpha is

more of a differentiating feature between math task

conditions, a finding that has not been outlined as

such in previous work, yet would explain why some

of the work that includes separation does find frontal

alpha to contribute valuable diagnostic information

between difficulty conditions (Ewing et al., 2016;

Katahira et al., 2018), while others that work with the

broad alpha band do not (Chanel et al., 2011;

Klarkowski, 2017). Whether or not the hiAlpha band

provides diagnostic potential for flow observation

beyond indication of a difference to boredom,

remains a subject of future work. Presently it appears

that flow and overload conditions show a similar level

of hiAlpha, that is lower than in the boredom phase

(thus showing a potentially reduced activity in frontal

brain regions in the boredom phase). The results of

frontal theta and alpha increases with sustained

attention and increased task difficulty are in line with

previous EEG research on mental workload (Borghini

et al., 2014). The results are also fairly similar to a

recent fNIRS-based study that finds frontal brain

activity to be reduced in easy/boredom conditions and

to increase when task difficulty increases (Barros et

al., 2018). The aforementioned authors attribute this

activity to attention on the task, which we find

plausibly transferrable given the volatile hiAlpha

signature only present in the math boredom phase

(temporal analysis), specifically as mind wandering

during this condition was noted explicitly by one

participant in the final experiment survey comment

section. However, it needs to be noted that the frontal

alpha reduction is not a unanimous finding in the

related work. While it is also inferred from the

amplitude-based results of Katahira et al. (2018), the

results by Léger et al. (2014) and Labonté-Lemoyne

et al. (2016) point in the opposite direction. Mainly,

this might stem from the difference in experimental

approaches and analyses. Labonté-Lemoyne et al.

(2016) for example observe two interacting

participants and don’t manipulate difficulty

externally. A last finding with potential implications

is the increase of beta activity in the math overload

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

condition compared to the boredom and flow

conditions (descriptive analysis). As previously

outlined, such beta band activity might be indicative

of a threshold when cognitive effort increases

strongly. Lower beta activity has been found to be

linked to higher flow experience self-reports (Léger

et al., 2014), yet again has also been found to increase

with task difficulty increases from boredom levels

(Klarkowski, 2017). It appears frontal beta can

increase at a certain level of difficulty. We would

expect this phenomenon to be visible when the

perceived stress levels increase, yet find no such

pattern. However, other studies have also found no

beta difference at all on frontal sites (Soltész et al.,

2014; Katahira et al., 2018).

Considered across tasks, beta increases would not

necessarily appear to be detrimental to flow, or

related to it for that matter, at least as the first writing

round that shows beta increases (descriptive &

temporal analyses), does not show differences in flow

experience reports (compared to the math flow

condition). Other comparisons across tasks further

support the potential relation of the hiAlpha band to

flow experience, or at least an expected corollary of it

that is attention on the task. Whether or not there is an

actually realized decrease in hiAlpha in the writing

task compared to the math overload condition

(statistical & descriptive analyses) could be an

interesting additional support of the relation of flow

experience to increased voluntary task attention.

In summary of the different frontal EEG features

investigated it can be noted, that a role of theta band

activity across tasks is in this data not supported,

pointing again to a reduced role in flow experience.

Overall, alpha band separation shows the most useful

diagnostic extension. For the beta band, this seems

less so to be the case, although a few results point to

potentially higher diagnostic properties of the

midBeta and hiBeta band. While frontal hemispheric

differences would intuitively appear to be related to

flow (e.g. as FAA is related to task approach

motivation) (Wolf et al., 2015; Labonté-Lemoyne et

al., 2016), our findings show no such pattern. Lastly,

the temporal sub-segmentation of experiment phases

indicates that at least for frontal sites, flow

experiences could be rather marked by consistency

than volatility. The findings of lower hiAlpha activity

in flow-related experiment phases point to further

support of frontal brain activity in flow experience.

This finding is in contrast to initial TH reasoning

(Dietrich, 2004), but in line with both previous EEG

work (Ewing et al., 2016), and other neuroimaging

studies indicating a more nuanced frontal activation

picture (Ulrich et al., 2014; Harmat et al., 2015;

Barros et al., 2018). Given the lack of midline frontal

electrode positions for the herein used headset and a

neglect of such dedicated differentiation of lateral and

medial frontal sites in related work (see Table 2), it

appears that the differentiating potential of frontal

EEG could be dependent on capturing more spatial

nuances (which might be difficult to attain) or have to

be accompanied by other sensors. Whether or not

frontal EEG activity alone can differentiate flow

experience from other experiential states has yet to be

explored further. Regardless of frontal activity, what

might perhaps be most interesting in the context of

this research approach, is that given the psychometric

differences in stress perceptions, it might be possible

to study a difference between the experience of flow

as a state of effortless (cESM) or effortful (DM)

attention (Hommel, 2010), if this perceptual

difference is confirmed in future work to be present

and relevant.

5.3 Study Limitations

The small sample size is a main limitation of this

study, which is why the results can only be treated as

preliminary. Through integration with related work

we have tried to somewhat overcome this limitation.

Considering the experiment design, future work

should increase experiential variance in the writing

task (e.g. by including a controlled, writing boredom

phase), and employ more psychometric scales

(involvement, effort, effortlessness, etc.) to enable

more detailed insights. Similarly, the integration of a

more self-determined difficulty adjustment as in

Barros et al. (2018), could provide additional

comparability between the two paradigms.

Physiologically, the work is limited to frontal sites in

favor of a more detailed inspection. We did not take

into account that there are other topographical regions

of interest that could be providing interesting

information on what differentiates flow from other

experiences. Some research for example points to the

explicit role of central (Katahira et al., 2018),

temporal (Wolf et al., 2015), or parietal and occipital

brain regions (Chanel et al., 2011).

6 CONCLUSIONS

We took an extensive look at psychometric and

physiological data in two flow induction paradigms

and compared data to unintegrated results of related

EEG studies. The summarized contributions are:

(1) We provide evidence for the applicability and

utility of the cESM approach to study flow in more

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks

unstructured tasks in the context of KW. The writing

task design appears to elicit a constant flow

experience that is at least as high in intensity as in an

established DM paradigm. At the same time the

cESM approaches elicits lower perceptions of stress,

which makes the approach an interesting, perhaps

qualitatively different option for flow research.

(2) We provide further evidence for neural

activity in flow experience, specifically in the form of

outlining the role of frontal EEG results, first by

consolidating related work, then by analysis across

two cognitive tasks. The results point to further

refusal of the hypofrontality hypothesis and instead

point to frontal activation that is visible through split

of the alpha band over averaged frontal sites (likely

indicating increased task attention). Furthermore,

temporal physiological and experiential volatility is

in this alpha band only indicated for a boredom

condition. This could support the hypothesis that flow

is actually experienced as fairly stable (Léger et al.,

2014) at least within these short time segments (5-

7min), and that volatility might be either visible in

different brain regions or over longer periods.

In future work, frontal alpha activity together with

heart rate variability (HRV) data could be a fruitful

approach to flow detection, given the observed HRV

decreases in autonomous flow experiences (Barros et

al., 2018). HRV reductions beyond what is expected

in higher task difficulties together with stable, frontal

hiAlpha activity could be a marker of flow experience

or at least its corollary of increased task attention, that

is explained by shared regulatory mechanisms

(Peifer, 2012; Barros et al., 2018). Although this

might not characterize flow neurophysiology

uniquely, it could show sufficient diagnosticity to

infer flow (vs. boredom or overload) experiences

whilst they are occurring automatically, thus enabling

the utilization of flow-facilitating bio-adaptive

systems in KW. Further detection performance might

then be achieved by inclusion of higher spatial

resolution on frontal brain activity, as the recent

fNIRS work by Barros et al. (2018) proposes flow to

be marked by activation of lateral frontal areas and

deactivation of medial frontal areas. Whether or not

this can be achieved using EEG data could be an

interesting avenue for future work, as would be the

search for neurophysiological differences that could

explain the stress perception difference and with it the

potential difference of flow experience as a state of

effortless attention (Hommel, 2010).

REFERENCES

Barham, M. P. et al. (2017) ‘Acquiring research-grade ERPs

on a shoestring budget: A comparison of a modified

Emotiv and commercial SynAmps EEG system’,

Psychophysiology, 54(9), pp. 1393–1404.

Barros, M. F. de S. et al. (2018) ‘Flow experience and the

mobilization of attentional resources’, Cognitive,

Affective, & Behavioral Neuroscience, pp. 1–14.

Berta, R. et al. (2013) ‘Electroencephalogram and

Physiological Signal Analysis for Assessing Flow in

Games’, IEEE Transactions on Computational

Intelligence and AI in Games, 5(2), pp. 164–175.

Blankertz, B. et al. (2016) ‘The Berlin brain-computer

interface: Progress beyond communication and control’,

Frontiers in Neuroscience, 10, pp. 1–24.

Borghini, G. et al. (2014) ‘Measuring neurophysiological

signals in aircraft pilots and car drivers for the assessment

of mental workload, fatigue and drowsiness’,

Neuroscience and Biobehavioral Reviews. Elsevier Ltd,

44, pp. 58–75.

Bradley, M. M. and Lang, P. J. (1994) ‘Measuring Emotion:

The Self-Assessment Manikin and the Semantic

Differential’, Journal of Behavior Therapy and

Experimental Psychiatry, 25(1), pp. 49–59.

Ceja, L. and Navarro, J. (2012) ‘“Suddenly I get into the

zone”: Examining discontinuities and nonlinear changes

in flow experiences at work’, Human Relations, 65(9),

pp. 1101–1127.

Chanel, G. et al. (2011) ‘Emotion Assessment From

Physiological Signals for Adaptation of Game Difficult’,

IEEE Transactions on Systems, Man, and Cybernetics –

Part A, 41(6), pp. 1052–1063.

Cohen, M. X. (2014) Analyzing neural time series data:

theory and practice. MIT press.

Csikszentmihalyi, M. (1996) Flow and the psychology of

discovery and invention. 1st edn. New York: Harper

Collins.

Csikszentmihalyi, M. and Hunter, J. (2003) ‘Happiness in

everyday life: The uses of experience sampling’, Journal

of Happiness Studies, 4, pp. 185–199.

Dietrich, A. (2004) ‘Neurocognitive mechanisms underlying

the experience of flow’, Consciousness and Cognition,

13(4), pp. 746–761.

Doran, G. T. (1981) ‘There’s a S.M.A.R.T. way to write

managements’s goals and objectives.’, Management

Review, 70(11), pp. 35–36.

Engeser, S. and Rheinberg, F. (2008) ‘Flow, performance and

moderators of challenge-skill balance’, Motivation and

Emotion, 32(3), pp. 158–172.

Erhard, K. et al. (2014) ‘Professional training in creative

writing is associated with enhanced fronto-striatal

activity in a literary text continuation task’, NeuroImage,

100, pp. 15–23.

Ewing, K. C. et al. (2016) ‘Evaluation of an Adaptive Game

that Uses EEG Measures Validated during the Design

Process as Inputs to a Biocybernetic Loop’,

Frontiers in

Human Neuroscience, 10, pp. 1–13.

PhyCS 2018 - 5th International Conference on Physiological Computing Systems

Flower, L. and Hayes, J. R. (1981) ‘A Cognitive Process

Theory of Writing’, College Composition and

Communication, 32(4), pp. 365–387.

Galluch, P. S. et al. (2015) ‘Interrupting the Workplace:

Examining Stressors in an Information Technology

Context.’, Journal of the Association for Information

Systems, 16(1), pp. 1–47. Available at:

http://proxy.lnu.se/login?url=http://search.ebscohost.co

m/login.aspx?direct=true&db=buh&AN=100815876&l

ang=sv&site=eds-live&scope=site.

Hair, J. F. et al. (2011) ‘PLS-SEM: Indeed a Silver Bullet’,

The Journal of Marketing Theory and Practice, 19(2),

pp. 139–152.

Harmat, L. et al. (2011) ‘Heart Rate Variability During Piano

Playing: A Case Study of Three Professional Solo

Pianists Playing a Self-Selected and a Difficult Prima

Vista Piece’, Music and Medicine, 3(2), pp. 102–107.

Harmat, L. et al. (2015) ‘Physiological correlates of the flow

experience during computer game playing’,

International Journal of Psychophysiology, 97(1), pp. 1–

Harris, D. J. et al. (2017) ‘Neurocognitive mechanisms of the

flow state’, Progress in Brain Research, 237(1), pp. 221–

243.

Hommel, B. (2010) ‘Grounding attention in action control:

The intentional control of selection’, Effortless attention:

A new perspective in the cognitive science of attention

and action, pp. 121–140.

Katahira, K. et al. (2018) ‘EEG correlates of the flow state:

A combination of increased frontal theta and moderate

frontocentral alpha rhythm in the mental arithmetic task’,

Frontiers in Psychology, 9, pp. 1–11.

Keller, J. et al. (2011) ‘Physiological aspects of flow

experiences: Skills-demand-compatibility effects on

heart rate variability and salivary cortisol’, Journal of

Experimental Social Psychology, 47(4), pp. 849–852.

Klarkowski, M. (2017) The Psychophysiologial Evaluation

of the Player Experience. Queensland University of

Technology.

Knierim, M. T. et al. (2017) ‘The psychophysiology of flow:

A systematic review of peripheral nervous system

features’, Lecture Notes in Information Systems and

Organisation, 25, pp. 109–120.

Kosti, M. V. et al. (2018) ‘Towards an affordable brain

computer interface for the assessment of programmers’

mental workload’, International Journal of Human

Computer Studies, 115, pp. 52–66.

Labonté-Lemoyne, É. et al. (2016) ‘Are We in Flow?

Neurophysiological Correlates of Flow States in a

Collaborative Game’, in Proceedings of the 2016 CHI

Conference Extended Abstracts on Human Factors in

Computing Systems, pp. 1980–1988.

Léger, P. M. et al.

(2014) ‘Neurophysiological correlates of

cognitive absorption in an enactive training context’,

Computers in Human Behavior, 34, pp. 273–283.

de Manzano, Ö. et al. (2010) ‘The psychophysiology of flow

during piano playing’, Emotion, 10(3), pp. 301–311.

Moller, A. C. et al. (2010) ‘Developing an experimental

induction of flow: Effortless action in the lab’, in Bruya,

B. (ed.) Effortless Attention, pp. 191–204.

Moneta, G. B. (2012) ‘On the Measurement and

Conceptualization of Flow’, in Engeser, S. (ed.)

Advances in Flow Research, pp. 23–50.

Nakamura, J. and Csikszentmihalyi, M. (2009) ‘Flow theory

and research’, in Lopez, S. and Snyder, C. R. (eds)

Oxford Handbook of Positive Psychology. New York:

Oxford University Press, pp. 195–206.

Peifer, C. (2012) ‘Psychophysiological Correlates of Flow-

Experience’, in Engeser, S. (ed.) Advances in Flow

Research. New York, NY: Springer Science, pp. 139–

164.

Picton, T. W. et al. (2000) ‘Guidelines for using human

event-related potentials to study cognition : Recording

standards and publication criteria’, pp. 127–152.

Rissler, R. et al. (2017) ‘Towards an integrative theoretical

framework of IT-mediated interruptions’, in Proceedings

of the 25th European Conference on Information Systems

(ECIS), pp. 1950–1967.

Rissler, R. et al. (2018) ‘Got Flow? Using Machine Learning

on Physiological Data to Classify Flow’, in Proceedings

of the Conference of Human Factors in Computing

Systems (CHI).

Soltész, P. et al. (2014) ‘Towards understanding the

electrophysiology of flow’, in Flow, Emotional

Intelligence and Psychological Immunity: Empirical

Studies in Positive Psychological Perspective, pp. 85–92.

Spurlin, S. and Csikszentmihalyi, M. (2017) ‘Will Work Ever

be Fun Again?’, in Fullagar, C. J. and Delle Fave, A.

(eds) Flow at Work: Measurement and Implications, pp.

176–187.

Tams, S. et al. (2014) ‘NeuroIS—Alternative or Complement

to Existing Methods? Illustrating the Holistic Effects of

Neuroscience and Self-Reported Data in the Context of

Technostress Research’, Journal of the Association for

Information Systems, 15, pp. 723–753.

Tian, Y. et al. (2017) ‘Physiological signal analysis for

evaluating flow during playing of computer games of

varying difficulty’, Frontiers in Psychology, 8, pp. 1–10.

Tozman, T. et al. (2015) ‘Understanding the

psychophysiology of flow: A driving simulator

experiment to investigate the relationship between flow

and heart rate variability’, Computers in Human

Behavior, 52, pp. 408–418.

Tozman, T. et al.

(2017) ‘Inverted U-Shaped Function

Between Flow and Cortisol Release During Chess Play’,

Journal of Happiness Studies, 18(1), pp. 1–22.

Ulrich, M. et al. (2014) ‘Neural correlates of experimentally

induced flow experiences’, NeuroImage, 86, pp. 194–

202.

Weber, R. et al. (2009) ‘Theorizing flow and media

enjoyment as cognitive synchronization of attentional

and reward networks’, Communication Theory, 19(4),

pp. 397–422.

Wolf, S. et al. (2015) ‘Motor skill failure or flow-experience?

Functional brain asymmetry and brain connectivity in

elite and amateur table tennis players’, Biological

Psychology, 105, pp. 95–105.

Flow Neurophysiology in Knowledge Work: Electroencephalographic Observations from Two Cognitive Tasks