University of Southern California Dissertations and Theses

Towards health-conscious spaces: building for human well-being and performance

(USC Thesis Other)

Towards health-conscious spaces: building for human well-being and performance

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Towards Health-Conscious Spaces: Building for Human Well-Being and Performance
by
Mohamad Awada
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(CIVIL ENGINEERING)
August 2024
Copyright 2024 Mohamad Awada

ii
Dedication
I am immensely grateful to Dr. Burcin Becerik-Gerber for her exceptional supervision and invaluable
advice throughout my academic journey. Her enthusiasm and inspiration have been pivotal to my growth
and achievements. It is her dedication and efforts that have guided me to where I stand today. I attribute
much of my success to her unwavering support and insightful guidance I would also like to extend my
gratitude to two remarkable researchers and collaborators, Dr. Gale Lucas and Dr. Shawn Roll, for their
insightful feedback. Their contributions have significantly enriched the work presented in this dissertation.
Lastly, my sincere appreciation extends to Dr. Lucio Soibelman, whose expertise and support as a part of
my thesis committee have been invaluable to the conclusion of this work.
This work is a tribute to my family who have stood by me throughout this journey. A special shoutout to
my parents, Ali Awada and Nisreen Tarraf, for laying the foundation that enabled me to soar to the highest
echelons of education. Your constant belief and support have been the bedrock of all my achievements.
And to my wife, Hala, you are more than just my partner and best friend. You have been a source of
unwavering love, support, and inspiration. Together, we have navigated this path, and I am grateful for
every step we have shared.
I am deeply grateful for the friendship and support I have received from my colleagues at iLAB and the
Sonny Astani Department of Civil and Environmental Engineering at USC. Additionally, I wish to express
my gratitude to the students I have had the privilege of mentoring, for their assistance at various phases of
my PhD journey and for enabling me to develop as a mentor.

iii
Acknowledgements
I am grateful for the partial support from the National Science Foundation under grant #1931226, #1763134,
#2204942, the National Institute for Occupational Safety and Health under the seed fund grant T42
OH008412, and the Viterbi Fellowship from the University of Southern California. Any opinions, findings,
conclusions, or recommendations expressed in this material are those of the authors and do not necessarily
reflect the views of the National Science Foundation, the National Institute for Occupational Safety and
Health, or USC.

iv
Table of Contents
Dedication.....................................................................................................................................................ii
Acknowledgements......................................................................................................................................iii
List of Tables..............................................................................................................................................viii
List of Figures..............................................................................................................................................ix
Abstract.........................................................................................................................................................x
Chapter 1: Introduction and Motivation........................................................................................................1
Chapter 2: Background and Literature Review.............................................................................................3
2.1. Overview of Healthy Buildings....................................................................................................3
2.2. The Work from Home Experience ................................................................................................4
2.2.1. Worker Characteristics .............................................................................................................4
2.2.2. Workspace Context....................................................................................................................5
2.2.3. Work Context.............................................................................................................................6
2.3. Indoor Environmental Quality Factors and Health in Workplace.................................................6
2.4. Overview of Stress........................................................................................................................9
2.4.1. The General Notion of Stress....................................................................................................9
2.4.2. Definition of Work Stress.........................................................................................................10
2.4.3. Eustress Vs. Distress...............................................................................................................10
2.4.4. Environmental Stress............................................................................................................... 11
2.5. Stress and Attention Recovery ....................................................................................................12
Chapter 3: Research Objectives and Questions ..........................................................................................13
3.1 Research Objective 1 ..................................................................................................................13
3.2 Research Objective 2 ..................................................................................................................13
3.3 Research Objective 3 ..................................................................................................................13
3.4 Research Objective 4 ..................................................................................................................13
3.5 Research Objective 5 ..................................................................................................................14
Chapter 4: Working from Home During the COVID-19 Pandemic: Impact on Office Worker Productivity
and Work Experience ..................................................................................................................................15
4.1. Methods.......................................................................................................................................15
4.1.1. Respondents ............................................................................................................................15
4.1.2. Procedure................................................................................................................................15
4.1.3. Measures.................................................................................................................................16
4.1.4. Data Analysis..........................................................................................................................17
4.2. Results.........................................................................................................................................18
4.2.1. Worker Characteristics ...........................................................................................................18

v
4.2.2. Workspace Context..................................................................................................................20
4.2.3. Work Context...........................................................................................................................23
4.3. Discussion...................................................................................................................................26
4.3.1. Theoretical Implications .........................................................................................................26
4.3.2. Practical Implications.............................................................................................................28
4.3.3. Limitations ..............................................................................................................................29
4.3.4. Future Research Directions ....................................................................................................29
4.4. Conclusion ..................................................................................................................................29
Chapter 5: Associations Among Home Indoor Environmental Quality Factors and Worker Health while
Working from Home during COVID-19 Pandemic ....................................................................................30
5.1. Methodology...............................................................................................................................31
5.1.1. Procedure................................................................................................................................31
5.1.2. Participants Characteristics...................................................................................................31
5.1.3. Measures.................................................................................................................................32
5.1.4. Data Analysis..........................................................................................................................32
5.2. Results.........................................................................................................................................32
5.2.1. Descriptive Analysis................................................................................................................32
5.2.2. Associations among Demographics and IEQ Satisfaction and Health...................................34
5.2.3. Associations among Satisfaction with IEQ Factors and Physical Health Symptoms.............35
5.2.4. Associations among Satisfaction with IEQ Factors and Mental Health Symptoms................36
5.3. Discussion & Recommendations for Future Work .....................................................................38
5.3.1. Associations among Demographics and IEQ Satisfaction and Health...................................38
5.3.2. Associations among Satisfaction with IEQ Factors and Physical Health Symptoms.............38
5.3.3. Associations among Satisfaction with IEQ Factors and Mental Health Symptoms................39
5.3.4. Implications for research and applications in future home office environments ....................39
5.3.5. Limitations and future research directions..............................................................................40
5.4. Conclusions.................................................................................................................................41
Chapter 6: Stress Appraisal in the Workplace and its Associations with Productivity and Mood: A
Multimodal Machine Learning Analysis.....................................................................................................42
6.1. Methodology...............................................................................................................................42
6.1.1. Participants.............................................................................................................................42
6.1.2. Physiological, behavioral, and human-computer interaction data.........................................42
6.1.3. Experimental protocol.............................................................................................................43
6.1.4. Feature Extraction and Data Processing................................................................................44
6.1.5. Outcome formulation ..............................................................................................................45

vi
6.1.6. Prediction assessment .............................................................................................................46
6.2. Results and discussion ................................................................................................................46
6.2.1. Perceived stress, mood, and productivity levels across stress appraisal states ......................46
6.2.2. Comparison between different ML models for stress appraisal prediction.............................48
6.2.3. Comparison between different modalities for stress appraisal prediction..............................49
6.2.4. Variation in physiological and behavioral signals across stress appraisal states..................52
6.3. Conclusions.................................................................................................................................54
6.3.1. Practical implications.............................................................................................................55
6.3.2. Limitations and future research directions..............................................................................56
Chapter 7: Predicting Office Workers Productivity: A Machine Learning Approach Integrating
Physiological, Behavioral, and Psychological Indicators...........................................................................57
7.1. Methodology...............................................................................................................................57
7.1.1. Participants.............................................................................................................................57
7.1.2. Data Collection.......................................................................................................................57
7.1.3. Data Processing......................................................................................................................59
7.1.4. Analysis Plan ..........................................................................................................................61
7.2. Results and Discussion ...............................................................................................................61
7.2.1. Predicting mood, stress, eustress, and distress .......................................................................61
7.2.2. Baseline vs. extended productivity models..............................................................................63
7.2.3. Analyzing features importance................................................................................................64
7.2.4. Comparing between different modalities ................................................................................68
7.3. Limitations and Future Work ......................................................................................................69
7.4. Conclusions.................................................................................................................................70
Chapter 8: Cognitive Performance, Creativity and Stress Levels of Neurotypical Young Adults Under
Different White Noise Levels .....................................................................................................................72
8.1. Methodology...............................................................................................................................72
8.1.1. Participants.............................................................................................................................72
8.1.2. Auditory conditions.................................................................................................................72
8.1.3. Test battery..............................................................................................................................73
8.1.4. Electrodermal activity.............................................................................................................74
8.1.5. Procedure and Experimental Design ......................................................................................75
8.1.6. Data Analysis..........................................................................................................................75
8.2. Results.........................................................................................................................................75
8.2.1. Sustained Attention – Continuous Performance Test..............................................................75
8.2.2. Selective Attention and Inhibition – Stroop Test .....................................................................76

vii
8.2.3. Working Memory – Two-Back Test..........................................................................................76
8.2.4. Creativity – Remote Associate Test .........................................................................................76
8.2.5. Performance – Typing Performance Test................................................................................76
8.2.6. Stress – Change in Mean Tonic Activity..................................................................................76
8.3. Discussion...................................................................................................................................77
8.4. Conclusion ..................................................................................................................................79
Chapter 9: The Impact of Light’s Color Correlated Temperature and Illuminance Levels on Stress and
Cognitive Functions Restoration.................................................................................................................80
9.1. Methodology...............................................................................................................................80
9.1.1. Experimental Design...............................................................................................................80
9.1.2. Research Settings....................................................................................................................80
9.1.3. Participants.............................................................................................................................80
9.1.4. Experimental Procedure..........................................................................................................81
9.1.5. Lighting Conditions.................................................................................................................82
9.1.6. Measures.................................................................................................................................82
9.2. Results & Discussion ..................................................................................................................84
9.2.1. Evaluating the effectiveness of inducing stress and mental fatigue ........................................84
9.2.2. Impact of lighting interventions on stress responses...............................................................85
9.2.3. Impact of lighting interventions on cognitive performance ....................................................87
9.3. Discussion...................................................................................................................................89
9.3.1. Impact of lighting interventions on stress responses...............................................................89
9.3.2. Impact of lighting interventions on cognitive performance ....................................................89
9.3.3. Practical applications in office environments.........................................................................90
9.3.4. Limitations and future research directions..............................................................................91
9.4. Conclusions.................................................................................................................................91
Chapter 10: Conclusion...............................................................................................................................93
References...................................................................................................................................................95

viii
List of Tables
Table 1. Relative productivity and change in time spent at workstation compared to pre-pandemic ........19
Table 2. Effects of worker characteristics on relative productivity and time spent at a workstation .........19
Table 3. Workspace context variables: Statistical Overview......................................................................20
Table 4. Work traits linked to relative productivity and time spent at workstation....................................21
Table 5. Worker & workspace context linked to productivity and workstation time changes ...................22
Table 6. Work context variables: Statistical overview................................................................................23
Table 7. Productivity and workstation time changes since pre-pandemic based on workspace context....24
Table 8. Regression: Productivity and workstation time changes against work and worker traits.............25
Table 9. Descriptive statistics of IEQ satisfaction and overall mental and physical health .......................33
Table 10. Number of responses related to the physical and mental health symptoms ...............................33
Table 11. Correlation: Satisfaction with IEQ factors, overall physical and mental health, and age...........35
Table 12. Logistic regression: Predicting physical health symptoms from IEQ satisfaction .....................37
Table 13. Logistic regression: Predicting mental health symptoms from IEQ satisfaction........................37
Table 14. Features dataset...........................................................................................................................44
Table 15. Stress appraisal formulation and data distribution across four stress appraisal states................46
Table 16. Comparison of ML model accuracy between different classifiers .............................................48
Table 17. Comparison different features and data collection tools in the stress appraisal prediction ........49
Table 18. Physiological and behavioral data changes across the stress appraisal states ............................52
Table 19. Feature Dataset ...........................................................................................................................60
Table 20. Summary of mood, stress, eustress, and distress prediction models performance .....................62
Table 21. Performance comparison between the baseline and extended productivity models...................63
Table 22. Correlation between predicted productivity and physiological, behavioral features..................66
Table 23. Comparative analysis between wearable devices and workstation addons................................68
Table 24. Statistical analysis of study measures under different noise conditions.....................................77
Table 25. Post-hoc analysis summary ........................................................................................................77

ix
List of Figures
Figure 1. Plots of the sample distribution in ratings of satisfaction with humidity, air quality, indoor
temperature, and overall mental health across income categories..............................................................34
Figure 2. Boxplots of perceived productivity, mood, and stress across stress appraisal states..................47
Figure 3. Feature importance for the stress appraisal prediction model.....................................................51
Figure 4. Participant taking the experiment showing the used sensors and the data collection platform ..58
Figure 5. Overview of analysis ..................................................................................................................61
Figure 6. Feature importance for the extended productivity prediction model..........................................65
Figure 7. Experimental Procedure..............................................................................................................82
Figure 8. Comparison of stress responses metrics pre- and post-restoration .............................................86
Figure 9. Comparison of cognitive performance metrics pre- and post-restoration...................................88

x
Abstract
The topic of occupant health in buildings is an emerging area for both academic research and industry
practices. Despite the importance of healthy buildings, we do not have a clear and commonly accepted
definition of what a “healthy building” means to buildings’ professionals and occupants. Moreover,
designers do not have a systematic process to incorporate the fundamental definitions of health in buildings.
As people spend the majority of their times in indoor environments [1], understanding how buildings affect
the health and well-being of occupants has become an increasingly important issue. Recently, the world
witnessed the spread of the novel SARS-CoV-2 virus. This channeled much-needed attention on the quality
of indoor life and its consequences on occupant health.
Accounting for 18.5 million people [2], office work is the most common type of work across the U.S.
workforce, therefore researchers have allocated substantial efforts in studying the impact of healthy
workplaces on the well-being, productivity and performance of office workers. Office work is often
associated with knowledge-based work [3]. Such work flourishes in workspace environments that promote
optimal conditions for better cognitive performance [4], [5]. Cognitive performance refers to various mental
processes including learning, thinking, reasoning, remembering, attention, perception, and executive
function [5], [6]. In any professional area, cognitive skills are a necessity to overcome work-related
challenges and create high-quality work. Thus, previous research has extensively examined how different
office Indoor Environmental Quality (IEQ) parameters can influence the cognitive performance of office
workers [7]–[10]. IEQ refers to the quality of a building's environment in relation to the health and wellbeing of those who occupy space within it. [6]. IEQ is determined by many factors, including indoor air,
thermal comfort, lighting, acoustics, water quality, interior design, spatial organization, and their
psychological impacts individually and collectively. A growing body of the literature shows that
improvement in IEQ conditions can boost the health and well-being of office workers, enhance their
performance, improve their concentration levels, and reduce their mental fatigue keeping them focused on
work [11], [12]. On the other hand, degradation of these conditions leads to quantifiable losses. For instance,
it was estimated that productivity losses from building-related illnesses range between $10 billion and $70
billion in the United States only [13].
The outbreak of SARS-CoV-2 virus forced office workers to conduct their daily work activities from home
over an extended period. Companies and organizations now aim to engage the work, workforce, and
workplace in a new system that identifies the work as a set of tasks to be completed, rather than linking it
with a specific location. Thus, the Work From Home (WFH) situation is likely to become part of future
office work and as such, it is important to understand what effect WFH have on workers, particularly their
health, productivity, and work experience [14], [15]. Furthermore, given this unique situation, an
opportunity emerged to study the satisfaction of office workers with IEQ factors of their houses where work
activities took place and associate these factors with mental and physical health.
Among the many consequences of the recent pandemic, the severity and prevalence of symptoms of
psychological distress have increased considerably in the United States and around the globe [16]. Stress is
recognized by the World Health Organization as the epidemic of the 21st century [17]. While there are many
reasons for stress, the American Psychological Association specifies that job pressure remains a major
stressor for most Americans [18]. Office work is a significant driver of work-related stress among
Americans, due to long hours of work, heavy workload, job insecurity, conflicts with co-workers or bosses,
and inadequate assignments of tasks. However, the literature distinguishes between two types of stress:
distress and eustress [19], [20]. Distress is what most people refer to when they feel “stressed out”. It usually

xi
results in people feeling overwhelmed when the cause of stress is not within their control. Distress can have
detrimental psychological effects on workers: loss of concentration, impaired performance, feeling of
insecurity as well as physical consequences: tension, insomnia, headaches, etc. [21], [22] On the other hand,
eustress motivates individuals to reach their goals, face challenges and achieve success and fulfillment [23].
Contrary to distress, when people feel confident about their ability to handle a stressor, their stress reaction
tends to be positive [22]. This type of stress is usually associated with higher concentration, productive
energy, motivation, confidence, and excitement [24]. Differentiating between a positive and negative
appraisal of stress is important, as eustress may be one of the most powerful resources to prevent or reduce
distress at work.
While there are many stressors related to work and the workplace, one of the sources most under the control
of the organization is environmental stress [25]–[28]. Environmental stress is defined as the physiological
and psychological responses to uncomfortable or unhealthy indoor conditions [29]. However,
environmental stress is not the only kind of stress experienced in the workplace, work stress -from the
demands of the work tasks outstripping workers capabilities and/or resources- also occurs. Thereby, even
if stress-oriented office design and operation guidelines could mitigate all the environmental stress, workers
would still experience work stress in the office. Accordingly, efforts need to also focus on recovering from
work stress and not only prevent environmental stress. The literature suggests that nature-based
interventions through visual, auditory, or olfactory stimuli can help with stress restoration [30]. More
recently, researchers became more interested in studying how appropriate indoor environmental
interventions can bring the worker to a relaxed state following a work stressor and restore the cognitive
functions of mentally fatigued office workers. Nevertheless, such studies are limited, and therefore more
research is needed in this area.
Based on this background, the research objectives of this proposal are defined as follows: (1) To understand
the worker-, workspace-, and work-related factors that affected office workers’ work experience on a typical
WFH day during the pandemic, (2) To study the associations between satisfaction with IEQ factors and
workers’ physical and mental health while working from home, (3) To establish an automated, multimodal
approach to monitor stress and differentiate between eustress and distress among office workers, (4) To
establish an automated, multimodal approach to monitor productivity of office workers, and (5) To
determine how different IEQ-related interventions can affect office workers’ cognitive performance and
psychophysiological states. The structure of this proposal is organized as follows: Chapter 1 provides the
overall background and motivation for the research efforts. In Chapter 2, a thorough literature review of
the proposal scope is presented, and research gaps are identified. Research objectives and questions are
described in Chapter 3.
To address research objective 1, Chapter 4 present a questionnaire-based study. Following the COVID-19
pandemic, office workers were forced to WFH. Chapter 4 answers three research questions: How do
workers’ demographics and their physical and mental health statuses affect workers’ productivity and the
time spent at the workstation when work is performed from home? How do workspace characteristics affect
workers’ productivity and the time spent at the workstation when work is performed from home? and How
do work conditions affect workers’ productivity and the time spent at the workstation when work is
performed from home? Thus, we designed and administered a questionnaire that was open for 45 days
during the COVID-19 pandemic and received valid data from 988 respondents. Results show that the
overall perception of productivity level among workers did not change relative to their in-office productivity
before the pandemic. Female, older, and high-income workers were likely to report increased productivity.
Productivity was positively influenced by better mental and physical health statuses, having a teenager,

xii
increased communication with coworkers, and having a dedicated room for work. The number of hours
spent at a workstation increased by approximately 1.5 hours during a typical WFH day. Longer hours were
reported by individuals who had school-age children, owned an office desk or an adjustable chair, and had
adjusted their work hours.
The second study of this dissertation, which used the same questionnaire presented in Chapter 4, is
described in Chapter 5 and addresses research objective 2. Chapter 5 answers three research questions:
How did satisfaction with the IEQ factors relate to the prevalence of physical and mental health symptoms
while working from home? What worker demographics were associated with satisfaction with IEQ and
overall mental and physical health while working from home? and What insights regarding the impact of
IEQ factors on health can be concluded based on the transition from traditional office environments to
home office environments? The results show that low satisfaction with natural lighting, glare, and humidity
predicted eye-related symptoms, while low satisfaction with noise was a strong predictor of fatigue or
tiredness, headaches or migraines, anxiety, and depression or sadness. Nose- and throat-related symptoms
and skin-related symptoms were only uniquely predicted by low satisfaction with humidity. Low
satisfaction with glare uniquely predicted an increase in musculoskeletal discomfort. Symptoms related to
mental stress, rumination, or worry were predicted by low satisfaction with air quality and noise. Finally,
low satisfaction with noise and indoor temperature predicted the prevalence of symptoms related to trouble
concentrating, maintaining attention, or focus. Workers with higher income were more satisfied with
humidity, air quality, and indoor temperature and had better overall mental health. Older individuals had
increased satisfaction with natural lighting, humidity, air quality, noise, and indoor temperature.
Through a controlled experimental procedure that involved 48 participants, physiological and behavioral
data were collected to establish automated frameworks for assessing stress appraisal and monitoring
productivity. This experiment serves as the foundation for both Chapters 6 and 7. Chapter 6 tackles
research objective 3, while Chapter 7 focuses on objective 4. The research questions answered in these
chapters are: How can we differentiate between eustress and distress among office workers using an
automated and multimodal approach? How does the inclusion of predictions about the psychological state
of office workers, when combined with physiological and behavioral features, enhance the precision of
productivity predictions? and Which data modalities are most effective for predicting stress appraisal and
productivity in individuals, and how do these modalities interact to enhance predictive accuracy? The
results from Chapter 6 show that moderate stress levels, as per the Yerkes-Dodson law, are linked to
increased productivity and positive mood, while both low and high stress levels are associated with
decreased productivity and negative mood, especially when distress overshadows eustress. An XGBOOST
model demonstrated the highest prediction accuracy for stress appraisal, highlighting the significance of
physiological data, particularly electrodermal activity, skin temperature, and blood volume pulse. Chapter
7 advances the research by employing a machine learning framework to predict office workers' perceived
productivity, integrating physiological, behavioral, and psychological features. The extended model,
incorporating psychological states, outperformed the baseline physiological and behavioral model, with
mood and eustress identified as critical productivity predictors. Wearable devices showed superior
performance over workstation addons in productivity prediction. Collectively, these findings advocate for
the integration of the proposed models within smart workstations, enabling adaptable environments that
enhance health, amplify eustress, and boost productivity and overall well-being among office workers.
Of all IEQ factors, environmental noise is a primary reason for dissatisfaction among office employees,
especially those in open-plan offices [31]. This was also supported by our results from Chapter 5, where
satisfaction with noise was among the lowest in comparison to the satisfaction with other IEQ parameters.

xiii
To solve this problem, office managers rely on white noise masking solutions [32]. However, studies about
the effect of white noise on environmental stress and the performance of office workers are limited. Thus,
in Chapter 8, we answer the first research question of objective 5: How does exposure to different white
noise levels affect the physiological responses of office workers? Our findings showed that white noise level
at 45 dB resulted in better cognitive performance in terms of sustained attention, accuracy, and speed of
performance as well as enhanced creativity and lower stress levels. On the other hand, the 65 dB white
noise condition led to improved working memory, which leads to the conclusion that different tasks might
require different noise levels for optimal performance. These results lay the foundation for the integration
of white noise as a tool to enhance the work of office workers.
Chapter 9 addresses the second research question of objective 5: What are the stress and attention
restorative effects of different light illuminations and color temperatures? In this study, we recruited 100
participants to examine how different lighting conditions affect stress reduction and the restoration of
cognitive functions following mentally taxing office tasks. This investigation revealed that certain lighting
conditions have a marked influence on cognitive performance and stress. Specifically, lighting with a
7000K color temperature significantly improved cognitive functions, likely due to its stimulating effects
similar to natural daylight, as demonstrated by enhanced scores on the Continuous Performance Test (CPT)
and the Visual Backward Digit Span (VBDS). In contrast, a 3000K color temperature at 100 lux was shown
to significantly lower physiological stress markers, suggesting its calming effects conducive to relaxation
and sleep readiness. These outcomes highlight the nuanced effects of lighting conditions on both cognitive
and emotional well-being in work environments.

1
Chapter 1: Introduction and Motivation
Even before the COVID-19 pandemic, people spent on average around 90% of their time indoors [33]. Now
more than ever, it is crucial that we radically rethink the design and operation of buildings. IEQ directly
affects the comfort and well-being of occupants. When IEQ is compromised, occupants are at increased
risk for many diseases that are exacerbated by both social and economic forces. In the U.S. alone, the annual
cost attributed to building-related illnesses in commercial workplaces is estimated to be between $10 billion
to $70 billion [13]. It is imperative to understand how parameters that drive IEQ can be designed properly
and how buildings can be operated to provide an ideal IEQ to safeguard health. While IEQ is a fertile area
of scholarship, there is a pressing need for a systematic understanding of how IEQ factors impact occupant
health. If the impact of buildings on occupant health is not well understood and the benefits of healthy
buildings are not clearly enumerated then integration of health objectives into the design, construction, and
operation of healthy buildings is not formalized. Health objectives are therefore not widely adopted by
building practitioners.
With the spread of the novel SARS-CoV2 virus, most office workers were obliged to shift to remote
working almost overnight in mid-March 2020, and the adoption of WFH strategies is likely to persist
beyond the pandemic. More specifically, companies and organizations are trying to incorporate the work,
workforce, and workplace in a new structure that recognizes the work as a set of duties to be completed
irrespective of the work location. Thus, it is important to understand the effects of the worker, workspace,
and work’s characteristics relative to the WFH experience [14], [15]. Studies have been conducted to
demonstrate the links between IEQ factors (lighting, glare, temperature, humidity, air quality, noise, etc.)
on occupant health and well-being at work in traditional office environments [34], [26], [35], [36]. Despite
the obligatory full-time WFH due to the COVID-19 pandemic, there has been limited examination of
occupants’ satisfaction with IEQ factors in home environments and their relationship with individual health
issues while working from home. This is because previous studies have examined the IEQ of households
as places to live but not to work [37], [38]. Given this unique situation, an opportunity emerged to study
the satisfaction of office workers with IEQ factors of their houses where work activities took place and
associate these factors with mental and physical health.
The results of Chapter 4 showed that at least 1 in 5 of office workers working from home reported an
increase in all mental health symptoms, with 33% showing symptoms of mental stress, rumination, or
worry. In addition, we found that the prevalence of mental stress symptoms was predicted by low
satisfaction levels with air quality and noise. Furthermore, results from [12] indicated that buildings’
professionals considered stress and anxiety to be the most important mental well-being issues that need to
be the focus of design, construction, and operation of buildings to support and promote occupant health.
With the outbreak of the SARS-CoV-2 virus, stress levels have increased across the globe [16]. In fact, we
regularly face stress during our everyday activities, to the extent that stress is recognized by the World
Health Organization as the epidemic of the 21st century [17].
There are many facets of life that can be sources of stress – people can experience stress based on various
stressors such as the outbreak of the SARS-CoV-2 virus, financial challenges, marital or family difficulties,
as well as a variety of stressors experienced at work. However, sources of stress are out of the direct control
of individuals and organizations, such as “the outbreak of the SARS-CoV-2 virus” and “the future of our
nation.” [18] In contrast, one of the top sources of stress is more in the direct control of individuals, and
especially organizations: work pressure [18]. Work-related stress negatively influences workers’

2
physiological, psychological, and cognitive functions [39], [40], thus leading to lower productivity and
increased absenteeism [41].
Stress is how humans respond physically and psychologically to adjustments, experiences, conditions, and
circumstances in their lives. When discussing stress, our minds immediately relate to the overwhelming
feeling we acquire when we are “stressed out”. However, stress is defined as the arousal witnessed by an
individual when facing a stressor, but the way people perceive that stressor determines whether the stress
reaction is considered as positive (eustress) or negative stress experience (distress) [42].
While distress can have detrimental psychological effects on workers: loss of concentration, impaired
performance, feeling of insecurity as well as physical consequences: tension, insomnia, headaches, etc.
[43], eustress motivates individuals to reach their goals, face challenges and achieve success and fulfillment
[24]. With the major breakthroughs in technology over the last two decades, harvesting the power of
machine learning creates a major opportunity to examine the psychological reaction of workers and their
appraisals of job stressors. Differentiating between a positive and negative appraisal of stress is important,
as eustress may be one of the most powerful resources to prevent or reduce distress at work [44]. The
literature suggests that researchers have a good understanding of the physiological, behavioral, and
psychological consequences of distress, but this knowledge does not provide a complete picture of the stress
research. A holistic understanding of stress requires a similar analysis but in terms of eustress. To the best
of our knowledge, there was no research study that tried to predict the appraisal of stress among office
workers.
Given that work can be a major source of stress, workers are interested in optimizing their office workspaces
to reduce experiences of stress at work [12]. This is well founded because the built office environment can
indeed induce stress. Specifically, office environmental conditions (e.g., thermal, indoor air conditions,
lighting, and noise) and interior design parameters (e.g., space, colors, furniture, access to views, and
biophilic design) have been found to affect office workers' stress levels. Theories such as Stress Reduction
Theory [45] and the Biopsychosocial Model of Stress [46] offer insights for understanding stress – and then
developing built office environments that limit environmental stress and help reduce the negative impacts
of stress.
Of all IEQ factors, environmental noise is a primary reason for dissatisfaction, reduced performance and
stress among employees, especially those in open-plan offices [31]. To solve this problem, office managers
rely on white noise masking solutions. Masking is the practice of using sound to cover and interfere with
disruptive noise in an office environment, by adding a layer of barely noticeable, specially-tuned, and
electronically generated sound (e.g., white noise), distracting conversations and sudden noises are
minimized [47]. Research shows that environmental noise induces stress resulting in physiological arousal
and reduced performance [48], however, there is limited research about the impact of masking white noise
on stress and performance. Lighting, another indoor environmental parameter, has a major impact on the
environmental stress and performance of office workers. This is because lighting has several components
(e.g., color correlated temperature, illuminance) that each can influence the physiological and psychological
states differently [49], [50]. While the impact of lighting on environmental stress as well as on performance
has been extensively studied, to the best of our knowledge, no study examined the possibility of using
lighting interventions for stress and performance restoration.

3
Chapter 2: Background and Literature Review
2.1. Overview of Healthy Buildings
The World Health Organization (WHO) defines health as “a state of complete physical, mental and social
well-being” [51]. Physical well-being is defined as the ability of our bodies to function appropriately and
resist illness [52]. The modern definition of mental well-being transcends the traditional definition of
“absence of mental illness” and is better defined as an individual’s ability to realize his or her abilities, and
be productive while coping with daily stresses of life [53]. Social well-being refers to a person's level of
social engagement and sense of belonging [54]. According to the WHO, the concept of health is affected
by the economic, social and personal factors as well as the physical built environment [55]. As such, Samet
and Spengler stated that indoor environments should be designed with the aim of enhancing the
physiological, psychological, and sociological functioning of occupants [56]. The literature thoroughly
explains how different Indoor Environmental Quality (IEQ) factors such as lighting, acoustic and thermal
conditions, indoor air quality, ventilation, humidity, spatial organization, ergonomics, and aesthetics can
trigger various physical, mental and social responses among occupants [36], [57]–[60]. Despite clear
evidences showing the effect of IEQ on health [61], to date, other building-related areas of research such
as energy efficiency and occupant comfort have received more attention [62].
Beyond the cause-effect relationship, the study of IEQ’s effect on occupant health is complex and multilayered. To assess health, researchers mainly rely on two major assessment schemes: subjective assessments
through surveys [63] or interviews [64] and objective assessments through physiological measurements
using sensors [65] or psychometric tests [66]. Other methods have also been employed in this area, such as
conducting medical tests and examining sick leave reports [67], [68], but these methods are less popular.
Occupant health in buildings depends on the type of building under study. For example, residential cooking
is considered one of the most substantial sources of indoor air pollutants in households, exposing occupants
to fine air particles that can lead to respiratory problems [69]. In office spaces, the goal remains to establish
a more comfortable, productive, and healthier work environment for office workers who spend most of
their time sitting. This makes the study of ergonomics and its effect on musculoskeletal disorders one of the
most important topics in the context of healthy office spaces [70]. Additionally, in offices as well as in
educational buildings, researchers have examined the relation between IEQ and fatigue, tiredness,
headaches, attention and focus, to address student learning and worker productivity [71], [72].
The objective of creating healthy buildings spans over the different lifecycle phases of a building: design,
construction, and operation. For example, in the early design phases, building designers should consider
the building orientation (whenever possible) to maximize access to daylight [73], create an interior design
that reduces noise transmission [74] (especially in offices), consider natural ventilation when possible, and
so on. Similarly, in the construction phase, contractors should avoid using materials with chemicals that
can leach into the indoor environment and flush the building before occupancy to eliminate indoor air
pollutants from newly installed systems [75]. During the operational phase, building owners and facility

4
managers should commit to an occupant-centered approach that prioritizes health. Facility managers should
continuously monitor the indoor air quality, thermal, acoustic, and luminous conditions and solicit occupant
feedback since they are the end users [76]. Given the stakes and the different stakeholders involved in
creating healthy buildings, it is necessary to engage both building researchers and building practitioners
and to understand their perspectives about the challenges to the healthy building movement [77].
2.2. The Work from Home Experience
Following the COVID-19 pandemic, companies and organizations now aim to engage the work, workforce,
and workplace in a new system that identifies the work as a set of tasks to be completed, rather than linking
it with a specific location. The WFH situation is likely to become part of future office work and as such, it
is important to understand the effects of the three aspects of future work -the worker, workspace and work
[14], [15] - relative to the WFH experience.
2.2.1. Worker Characteristics
The WFH experience is inevitably influenced by worker demographics, including gender, age, and income
[78]. Even in 2021, gender gaps in the workplace and at home persist [79], which might lead to the
assumption that women might have lower productivity than men because women spend more time on
household chores and child caregiving [80], [81]. However, research prior to the pandemic has shown that
women’s productivity is similar –if not higher— compared to men’s productivity [82]. Income and age have
also been well-studied relative to work-related productivity, noting that higher-waged, middle-aged workers
are more productive than lower-waged, younger counterparts [83], [84].
The impact of these demographic factors on productivity in remote work is less clear. Applying a traditional
perspective of women’s engagement in household tasks would suggest potential negative impacts on
productivity, while an equity approach that considers increased involvement of men in household duties
could level any gender gaps in effects on productivity during remote work [81]. Similarly, other factors at
home, such as the presence of family members due to the pandemic, could alter the conventional ageincome-productivity association. For example, middle-aged workers who are working alongside their
children, could become overwhelmed due to parenting demands at the expense of their work engagement
and this can negatively impact their productivity [85].
In addition to the demographic factors, WFH may create different challenges for workers with different
occupational backgrounds. Prior studies have investigated the impact of WFH on productivity within
specific group of workers (e.g., workers of a Chinese travel agency [86], workers of U.S. Patent and
Trademark Office [87]). However, there has not been a study that investigated effects of WFH on
productivity across different occupational groups. A transition to remote work would likely have low risk
of loss in efficiency for workers who primarily engage directly with computer workstations throughout their
day (e.g., programmers) as opposed to individuals working in jobs that require mixed tasks in an
interpersonal environment (e.g., health care office workers). In addition to potential negative impact on
efficiency in job performance, some of these occupational groups that do not typically spend their entire
day at a computer would likely experience a dramatic shift in the amount of sedentary time at their
workstations. For example, WFH during the pandemic has led lawyers and judges to spend more time at
their workstations to virtually attend court trials [88].

5
Increased sedentary work and other potential aspects of personal health and well-being are important
considerations in the discussion of WFH. Importantly, worker health has been consistently associated with
productivity, such that the more health issues workers report the worst their productivity levels [89]. A
variety of physical health issues such as eye strain, nose related symptoms, fatigue, and headache, as well
as mental health issues such as anxiety, depression, stress, and insomnia can all have a negative impact on
productivity. Specifically, there is a growing body of evidence related to the concept of worker presenteeism
that demonstrates degraded work performance due to the existence of physical and mental health issues
[90]. Although it may be more easily concluded that these relationships would exist similarly in remote
contexts, there are limited studies that examine the impact of worker health on productivity while WFH.
2.2.2. Workspace Context
Workspace context plays a major role in shaping the work experience. Satisfaction with one’s workspace,
privacy, and ability to personalize workspace are predictors for worker productivity [91], [92]. The shift
from working in a well-established office space to work from home can be challenging for many office
workers. Such challenges can be stressful and might negatively affect a worker’s desire to work and thus
reduce their productivity. Having the optimal physical setup, proper ergonomics and the necessary
equipment is crucial to create an effective workspace that boosts productivity and increases the workers’
engagement with their workstation. In their analysis of the workforce shift to the WFH, Moretti et al. [93]
explained that workers are expected to engage extensively with their workstations while working from
home, and therefore presented their suggestions for a comfortable workstation (i.e., an adjustable desk and
chair to prevent back and joints pain, along with a footrest, and an adjustable monitor screen).
Research also shows that separating the workspaces from living spaces is an important factor when working
remotely. It is recommended to have a dedicated workspace to create physical boundaries, help workers
establish a productive work atmosphere, increase worker desire to stay longer hours at their workstation
and signal to other household members that they do not want to be distracted [94]. Yet, when the space at
home is limited and several members of the household need a space to work, sharing the same workspace
might become inevitable. In fact, in a survey conducted by Suart et al. [95], it was found that only 48.6%
of the respondents had a dedicated workspace, 31% were sharing their workspace with others and the
remaining 20.4% were working in a variety of places in their homes. However, research shows that
productivity decreases with lack of ability to adjust/personalize workspace as well as lack of storage space
[96]. On the other hand, Rudnicka et al. [97] conducted a survey about the WFH during the COVID-19 era
and found that some respondents felt that constantly changing the workspace helped them focus and
enhanced their work performance.
In addition to the workspace, indoor environmental quality (IEQ) (e.g., lighting, temperature, ventilation,
air quality, noise) also plays a major role in creating a comfortable work experience [98]. Research studies
that investigated the effect of IEQ parameters on worker performance suggest that the more satisfied the
workers are with the IEQ, the more productive they are [99], [100], an effect that is stronger in private
offices as opposed to shared offices [101]. In fact, increasing daylight illumination in office spaces can
increase workers’ performance by 13% while also reducing fatigue [102]. Beyond lighting, improvement
in indoor air quality and thermal conditions have also been linked to enhanced productivity, as well as
increased attention and concentration [103]. Research supports that the more control individuals have over
their environments, the more satisfied they are with it [104], thus having access to environmental controls
might also improve worker productivity. While one might expect to find a similar relation between IEQ
satisfaction and productivity during WFH, it is important to investigate this relation for home offices.

6
2.2.3. Work Context
Another important consideration in worker productivity and work experience when WFH is the ability of
workers to set and maintain appropriate boundaries between work duties and house responsibilities. With
role conflicts, workers used to find it challenging to manage work and family/life commitments, even before
the pandemic era [105], [106]. With workers shifting to WFH abruptly with the pandemic, new forms of
conflicts between work and life occurred. When working and living in the same space, setting boundaries
between the work and life becomes more challenging. For example, the sense of time might fade in the
homogeneous work-home environment and workers might elongate their working hours, start working
earlier, later or on the weekends [107]. Some workers might embrace the flexibility in their work hours
while other workers might have no choice but to schedule their working hours around their household
members or responsibilities [108].
At the same time, the unclear boundaries between home and office might have increased work expectations
[109]. For example, Peasley et al. [110] found that sales personnel felt burned out when trying to meet the
management’s expectations and they believed that job expectations became higher as soon as they started
working from home during the COVID-19 pandemic. With this increase in expectations, workers might be
tasked with more duties and expected to deliver additional work, increasing working hours and requiring
them to spend additional time at their workstations [111]. Meanwhile, communication among different
parties (e.g., coworkers, supervisors, employers), which is crucial to keep the remote workers productive,
might be impaired as communication during the pandemic usually comes in one form: virtual and is limited
by personal, organizational and technological means. Yet, communication provides work resilience by
sustaining the usual work operations and is key to mitigate the undesired effects of the sudden shift to WFH
[112]. Thus, the impact of work adjustments and work expectations on workers’ productivity and work
engagement needs to be further investigated.
2.3. Indoor Environmental Quality Factors and Health in Workplace
Indoor Environmental Quality (IEQ) is defined as the conditions of built environment in relation to the
comfort, health, and well-being of occupants [113]. Different IEQ parameters can trigger various physical
and psychological responses among people that depend on the intensity and duration of exposure in the
indoor environment. Because we spend a significant portion of our lives at work and much of this work
occurs indoors, it is important to understand the relationships among IEQ parameters and health,
particularly within the workplace environment. Key IEQ factors related to office workplaces include noise,
lighting, temperature, humidity, and air quality, each of which can impact physical and mental health.
Of all IEQ factors, noise is considered to be a primary reason for dissatisfaction for employees, especially
those in open plan offices [31]. Office noise can be generated by building systems, outdoor traffic, electronic
devices, drawers and doors being opened and closed, and conversations among coworkers, both
comprehensible and incomprehensible [114]. Evans argues that noise, an environmental stressor, has
profound psychological, behavioral, and cognitive consequences on workers [115]. On the psychological
aspect, office employees subjected to lower levels of noise experience less cognitive stress and hypertension
[116]. Also, Lamb and Kwob (2016) found that noise results in deteriorated mood and increased risk of
headaches. Behaviorally, a study showed that when workers are exposed to prolonged durations of noise,
they were less likely to make postural adjustments which could increase their risk to be affected by
musculoskeletal disorders (Evans and Johnson, 2000). Regarding the cognitive effects, an experiment

7
conducted by Jahncke et al. (2011) showed that the exposure to noise reduces the motivation to work and
reduces the memory span.
In addition to noise, the luminous environment plays an important role in supporting healthy indoor working
conditions. Access to natural lighting has been associated with long-lasting effects on the physical and
mental well-being of occupants [119], such as improved mood, better sleep quality [120], and reduction in
eye strain [121]. On the other hand, poor access to daylight, can disturb the human circadian rhythm [122].
The circadian system is the biological clock of the human body, which allows humans to stay synchronized
with 24 hour day cycle [123]. Such disturbance of the circadian rhythm affects sleep, alertness, and the
physiological and psychological body functions [124].
The color temperature and intensity of electric lighting also affects the psychological and mental states of
occupants. For example, in a laboratory experiment, Lan et al. (2021) found that bright and cool color
lighting induced positive effects and improved the mood of office workers. Viola et al. (2008) exposed 94
office workers to two lighting conditions: blue enriched-white light (17000K) and white light (4000K).
Workers presented improved alertness, mood, performance, concentration, less fatigue, and eye discomfort
under the blue enriched-white light condition in comparison to the white light.
Another factor related to visual conditions in built environments is glare, which is a visual sensation caused
by poor light distribution and excessive brightness in the field of view. Glare limits people’s ability to see
clearly and creates a feeling of annoyance and discomfort that can lead to a loss of concentration and
attention [126]. Sustained exposure to glare can result in eyestrain and eye fatigue that can lead to impaired
vision and, in extreme cases, eye injuries [127].
Inadequate indoor thermal conditions represent another factor that can degrade occupants’ well-being and
health [128]. Witterseh, et al. (2004b) found that office workers who were uncomfortable with typical
thermal conditions in their workspace showed higher prevalence of headache, throat, and eye irritation. In
addition, Wyon and Wargocki (2006) suggested that rapid temperature swings aggravate sick building
syndrome symptoms and have detrimental effects on cognitive performance. Furthermore, extreme thermal
events can result in conditions such as hypothermia or heat stroke and can increase cardiovascular mortality,
especially among children and the elderly [130].
Humidity, or the relative amount of moisture in the air, can also affect the health of occupants in indoor
environments. Low humidity levels can stimulate the evaporation of the tear film leading to a dryness
sensation of the eye, which results in increased irritation and eyestrain [131]. Also, low humidity levels can
cause the skin and nose to dry out and lead to itching, chapped lips, and skin and nose irritation [132]. On
the other hand, high humidity levels accelerate the growth of mold which can reduce overall air quality and
aggravate allergies, asthma, and cause other breathing problems[133].
Over the last two decades, increasing attention has been given to the effect of indoor air quality (IAQ) on
occupants’ health in buildings, and this focus has intensified with the spread of the SARS-CoV-2 virus. The
indoor air in buildings is a mix of outdoor air contaminants brought into the building through the mechanical
or natural ventilation systems, and indoor air contaminants associated with building materials, tap water,
appliances, excessive moisture, pets, and human behaviors [134]. Indoor gases like radon, carbon
monoxide, ozone and oxides of nitrogen, volatile organic compounds and particulate matters can cause

8
short term health issues such as eye, nose, and throat irritation, headaches, and vomiting. They can also
cause long-term health problems associated with cancer, and damage to the liver, kidney, and central
nervous system, asthma and chronic obstructive pulmonary diseases [135]–[137]. In addition to the direct
effects the indoor air pollutants have on the physical health of occupants, several research studies
demonstrated the relationship between increased indoor air pollution and mental health issues and disorders.
Both observational and experimental procedures have proven that indoor air contaminants are linked with
deteriorated mood, amplified aggressive behaviors, degraded attention, mental fatigue and higher
depression and stress rates (Evans 2003, H. Shin et al. 2018). Furthermore, several studies examined the
detrimental effect of degraded IAQ on office workers’ cognitive performance. For instance, Allen et al.
(2016) concluded that office workers showed higher cognitive function scores when carbon dioxide and
total volatile organic compounds concentrations were minimal. Similarly, other studies found that better
indoor air quality in green buildings is associated with fewer sick building syndrome symptoms, higher
sleep quality, and higher cognitive test scores [141], [142]. Zhang et al. (2017) found that degraded air
quality lessens the ability to think clearly, while decreasing the answering speed, response time and number
of correct answers in several cognitive tests.
Prior to the pandemic, office workers with different income levels shared the same office environment,
experiencing similar IEQ conditions. Thus, within the same office environment, office workers’ income had
less effect on workspace IEQ conditions, satisfaction with IEQ and their associations with health. When
workers shifted to working from home, their income level, which is a direct indicator of housing quality,
might have influenced the IEQ conditions. Several studies found that low income families, in comparison
to high income families, were exposed to higher noise and indoor air pollution levels [144], and had more
crowded homes with lower structural quality [145]. In another study, Hong et al. (2009) found that the
renovation of low-income family houses increased overall satisfaction with thermal environment from
36.4% to 78.7%.
People with low income are more likely to live in substandard housing conditions, which inevitably creates
major health disparities between high and low-income house owners [147]. Substandard houses are
characterized with low structural quality, water leakages, lack of proper ventilation, degraded thermal
conditions and insufficient lighting leading to poor physical and mental health and well-being among
building occupants [148]. Krieger and Higgins (2002) proved that substandard houses increase the risk of
asthma, and physical injuries and aggravate mental health problems. To that end, Evans et al. (2000) showed
that when housing quality is better, occupants’ psychological distress drop, while Suglia et al. (2011)
postulated that living in degraded housing conditions is generally associated with intense depressive
symptoms. Additionally, children’s emotional health was found to be worse in households with deteriorated
interior and exterior physical conditions [152].
IEQ conditions vary not only with income but also with age and gender. A survey study by Bae et al.(2020)
found that people in the 35-54 years age group were less satisfied with IEQ than both younger and older
groups. Furthermore, it was found that workers between 46 and 55 years old were more satisfied with noise
in comparison to others, and the youngest and oldest groups showed higher satisfaction with the air quality
compared to workers between 46 and 66 years old. [154]. Regarding gender, male workers were found to
have higher satisfaction with the indoor environment, including electric lighting, noise and thermal
conditions, when compared to female workers[155]. Finally, women were found to be more comfortable

9
and satisfied with the thermal environment at 24°C, while men were comfortable at lower indoor
temperatures, i.e., 23°C [156].
During the COVID-19 pandemic, the necessity for a comfortable and healthy workplace has been pushed
into the spotlight and the conversation about IEQ has exploded. This pandemic exposed the weaknesses in
our indoor environments to protect us not only against the SARS-CoV-2 virus but also from other indoor
air contaminants. In addition to air quality, it quickly became apparent that the indoor environments within
workers’ homes were not prepared to support the sudden shift to work from home. Importantly, many
workers might not have a dedicated workspace at home, nor a comfortable or healthy workspace which
lacks proper IEQ conditions to continue working effectively and in a health-promoting manner. Even when
the pandemic is resolved, many workers may continue to work from home. To that end, it is necessary to
further understand the IEQ within workers’ homes to identify which IEQ factors are most salient to support
worker physical and mental health when working at home.
2.4. Overview of Stress
2.4.1. The General Notion of Stress
The first definition of stress was provided by Hans Selye, in the 1930s: “stress is a non-specific response
of the body to any demand” [157]. Selye established a three-stage stress model, namely the General
Adaption Syndrome. This model explains how the human body reacts and adapts to stressors [158]. The
first stage, called the alarm reaction stage, refers to the initial reactions (increased heart rate, cortisol release,
boost of adrenaline) the body undergoes when facing a stressor. Selye refers to the second stage as the stage
of resistance, during which the body tries to overcome the stress shock. If the stressful situation does not
exist anymore, the body reduces the secretion of the hormones, and stabilizes the heart rate and blood
pressure until they reach the pre-stress phase levels. However, if the stressful situation persists, the body
cannot recover and restore pre-stress functioning levels, leading to the third stage: the exhaustion stage.
Battling with stress for long periods can drain the body's energy by depleting the physical, emotional, and
cognitive power.
This calls for different definitions of stress depending on duration. Acute stress is short-term stress which
occurs due to a recent or anticipated encounter or unexpected event. During an episode of acute stress,
people experience emotional distress and irritation as well as muscle tension, headaches, and back pain
[159], [160]. Episodic stress occurs when people experience acute stress repeatedly (e.g., repeated tight
deadlines at work). Episodic Stress can lead to migraines, hypertension, and heart diseases [160]. Chronic
stress results from stressors that persist continuously over time (e.g., difficult marriage or job, health
problems, poor living conditions, etc.). Chronic stress is associated with digestive and sleeping problems,
loss of focus on daily tasks and can cause diseases, such as heart problems and type 2 diabetes [161].
Following Selye’s model, different approaches have been proposed to examine the concept of stress. For
instance, the engineering model identifies stress as what happens to an individual rather than what happens
within the individual, by arguing that stress lies in the stressor’s characteristics of a person’s environment
[162]. Finally, the transactional model considers stress as an outcome of the interaction between an
individual and the environment while taking into consideration their perception [163]. This model is
considered a mix of Selye’s, and the engineering stress models.

10
Over the years, several definitions emerged to reflect a broader and modern understanding of stress. For
instance, the Cleveland Clinic’s definition described stress as “a normal reaction (physically, mentally, or
emotionally) the body has when changes occur” [164]. The American Psychological Association defines
stress in the Diagnostic and Statistical Manual of Mental Disorders as “the pattern of specific and
nonspecific responses a person makes to stimulus events that disturb his or her equilibrium and tax or
exceed his or her ability to cope” [165]. Though there is not a unified definition for stress, most definitions
agree that stress is caused by a stressor and manifested by a bodily reaction [166].
2.4.2. Definition of Work Stress
The BioPsychoocial (BPS) model of stress explains how the manifestation in a bodily reaction depends on
how the stressor is appraised, specifically whether the stressor is seen as a challenge or a threat [46]. That
is, stressors can be appraised as either a challenge or threat and those two different appraisals result in
different bodily reactions, which have physiological markers. A stressor is perceived as a challenge when
the individual believes they have sufficient capabilities and personal resources to meet the demands of the
stressful work situation. A stressor is perceived as a threat when the individual believes they do not have
sufficient capabilities or personal resources to meet the demands of the stressful work situation. The bodily
reaction that follows from perceiving a stressor as a challenge is a more efficient cardiovascular pattern
than the one that follows from perceiving it as a threat. Specifically, perceiving it as a challenge increases
blood flow throughout the entire body (preparing the body to meet the challenge, e.g., by fighting it headon), whereas perceiving it as a threat decreases blood flow throughout the entire body (prepares the body
to merely survive the threat, e.g., by not bleeding out).
Over time, consistently perceiving stressors as a threat will result in hypertension (and other consequent
ailments described above) due to ongoing vasoconstriction; perceiving stressors as challenges do not. This
supports the aforementioned definition of work stress –when demands in a work situation exceed the
capabilities/resources of the worker– which matches the above definition of threat but suggests that
stressors perceived as challenges do not qualify as “work stress.” This can be understood by the fact that
“work stress,” as defined by NIOSH and others, is limited to distress. Indeed, Selye distinguished between
two types of stress: negative distress and positive eustress - the former being unpleasant and the latter being
healthy and desirable [19]-[20]. In this context, stress is defined as the arousal witnessed by an individual
when facing a stressor, but the way people perceive that stressor determines whether the stress reaction is
considered as eustress or distress [163].
2.4.3. Eustress Vs. Distress
One of the most influential models for understanding the impact of work stress on worker health is the
demand-control model [167]. Within this model, low control contributes the most to stress. Other
frameworks identify additional moderating factors (e.g., lower reward increases stress in high demand job
[168], lack of support induce stress [169]). However, these models focus on the detriments of stress to
workers, ignoring the fact that when proper work conditions are established, positive stress experiences can
also occur [42].
Distress (here thereafter the word distress refers to the negative stress experience) is what most people refer
to when they feel “stressed out”. It usually results in people feeling overwhelmed when the cause of stress
is not within their control [170]. Distress can have detrimental psychological effects on workers: loss of
concentration, impaired performance, feeling of insecurity as well as physical consequences: tension,

11
insomnia, headaches, etc. [43] In a survey of 17,000 American office workers, 33% missed work because
of distress [171]. American companies are estimated to lose up to $300 billion annually due to worker
distress [172], which also burdens health services [173], altogether reducing the overall workforce
productivity, and thus national gross domestic product [174]. Thus, office work distress becomes a
significant concern that requires an urgent solution to alleviate its consequences.
On the other hand, eustress (here thereafter the word eustress refers to the positive stress experience)
motivates individuals to reach their goals, face challenges and achieve success and fulfillment [24].
Contrary to distress, when people feel confident about their ability to handle a stressor, their stress reaction
tends to be positive [44]. This type of stress is usually associated with higher concentration, productive
energy, motivation, confidence, and excitement [24]. The adverse consequences of job stress have been
studied thoroughly; however limited attention has been allocated to examine the positive impact of job
stress. Most work organizations adhere to a management plan that aim at enhancing workers productivity
by limiting work distress, assuming that stress is always negative [175]. The notion of eustress is often
ignored; a successful management plan must minimize distress but promote and sustain eustress.
Differentiating between a positive and negative appraisal of stress is important, as eustress may be one of
the most powerful resources to prevent or reduce distress at work [44]. On that note, Hargove et al. [176]
suggest that organizations should identify approaches that place challenging expectations for employees,
through the optimization of work stressors to proactively minimize distress and maximize eustress.
However, the authors argue that defining eustress and distress as a function of a worker’s cognitive,
affective, and physiological state is crucial for the success of any stress-related management plan. Similarly,
Simmons and Nelson [177] posit that researchers have a good understanding of the physiological,
behavioral, and psychological consequences of distress, but this knowledge does not provide a complete
picture of the stress research. A holistic understanding of stress requires a similar analysis but in terms of
eustress.
2.4.4. Environmental Stress
In addition to the organizational, social, and economic stressors, the indoor environment of the office plays
a major role in the complex equation of work stress. More specifically, degraded indoor environmental
quality and inappropriate design have been associated with increased environmental stress. This, in turn,
leads to elevated overall work stress and as such reduced productivity [29]. In their study, Lamb and Kwok
[178] found that environmental stressors act indirectly on workers’ performance by constraining motivation
and increasing tiredness and distractibility. They argue that limiting the effect of these stressors by
enhancing the IEQ conditions will inevitably result in improved productivity. Similarly, Singh et al. [179]
conducted a case study where office workers were moved from traditional offices to LEED-rated buildings
to examine the related economic benefits resulting from improved IEQ conditions. The results indicate an
average increase in productivity of 2.86 work hours every month due to reduced environmental stress.
Accordingly, some companies are interested in optimizing their office spaces to reduce experiences of stress
at work. However, it is challenging (and expensive) to modify the nature of work and the capabilities that
workers possess. The “lower hanging fruit” that companies can more easily (and economically) control is
the office environment itself. In fact, in a survey of building professionals on the development of healthy
buildings [12], they rated stress as the most important problem for research, design, construction, and
operation of healthier buildings in terms of mental well-being. Furthermore, they indicated that offices

12
should be given high priority to promoting comfortable and more productive work conditions. This serves
as a call for companies to invest in office environments that reduce environmental stress, thus creating
healthier and more productive work environments [180].
2.5. Stress and Attention Recovery
Different IEQ and interior design conditions affect stress experienced in the workplace, in particular
environmental stress. However, environmental stress is not the only kind of stress experienced in the
workplace, work stress -from the demands of the work tasks outstripping workers capabilities and/or
resources- also occurs. Thereby, even if stress-oriented office design and operation guidelines could
mitigate all the environmental stress, workers would still experience work stress in the office. Accordingly,
efforts need to also focus on recovering from stress, and replenishing attention, returning to baseline
psychophysiological conditions after experiencing a stressor.
Stress Reduction Theory (SRT) proposed by Ulrich et al. [45] suggests that nature settings can facilitate
stress recovery. This is supported by evolutionary perspectives suggesting that our early human ancestors
have formed the predisposition towards natural contents such as plants and water to help them survive.
However, human-nature interaction lacks in the contemporary social life due to urbanization and modern
lifestyle [30]. As people spend 90% of their time indoors [1], new efforts have been made to incorporate
naturistic features into buildings. This concept, known as biophilic design, provides buildings’ occupants
with the necessary exposure to nature. Being aware of its importance, building professionals –practitioners
and researchers– have been pushing toward adopting biophilic design in office spaces [12]. The literature
provides a range of studies that discuss the positive effects of biophilic design on workers’ cognitive
performance, health, and well-being [30]. More specifically, research shows that nature contact can actively
help office workers recover from work stress [181]. Additionally, Attention Recovery Theory (ART)
underscores the significance of these natural elements in restoring attentional capacities, thereby enhancing
cognitive performance and reducing cognitive fatigue [182].
Given the impact the indoor environment has on environmental stress and the physiological state of
occupants, it is convenient to speculate whether certain indoor environmental interventions (beside those
related to nature) can help with the stress recovery process. Furthermore, it's crucial to explore how these
interventions might also facilitate attention recovery, especially considering the potential benefits of
different lighting conditions (color temperature and illumination) on this process. To the best of our
knowledge, the specific effects of lighting conditions on attention recovery in addition to stress recovery
have not been explored in prior studies.

13
Chapter 3: Research Objectives and Questions
3.1 Research Objective 1
To understand the worker-, workspace-, and work-related factors that affected office workers’ work
experience on a typical WFH day during the pandemic.
• Research Question 1.1: How do workers’ demographics and their physical and mental health
statuses affect workers’ productivity and the time spent at the workstation when work is performed
from home?
• Research Question 1.2: How do workspace characteristics affect workers’ productivity and the time
spent at the workstation when work is performed from home?
• Research Question 1.3: How do work conditions affect workers’ productivity and the time spent at
the workstation when work is performed from home?
3.2 Research Objective 2
To study the associations between satisfaction with IEQ factors and workers’ physical and mental health
while working from home.
• Research Question 2.1: How did satisfaction with the IEQ factors relate to the prevalence of
physical and mental health symptoms while working from home?
• Research Question 2.2: What worker demographics were associated with satisfaction with IEQ and
overall mental and physical health while working from home?
• Research Question 2.3: What insights regarding the impact of IEQ factors on health can be
concluded based on the transition from traditional office environments to home office
environments?
3.3 Research Objective 3
To establish an automated, multimodal approach to monitor stress and differentiate between eustress and
distress among office workers.
• Research Question 3.1: How to differentiate between eustress and distress among office workers
using an automated and multimodal approach?
• Research Question 3.2: What systematic variations in physiological and behavioral data across
participants can be observed to create general eustress and distress profiles?
• Research Question 3.3: Which data modalities are most effective for predicting stress appraisal in
individuals, and how do these modalities interact to enhance predictive accuracy?
3.4 Research Objective 4
To establish an automated, multimodal approach to monitor productivity of office workers.
• Research Question 4.1: What level of prediction accuracy is attainable when focusing solely on the
physiological and behavioral features to predict the productivity of office workers?

14
• Research Question 4.2: How does the inclusion and prediction of the psychological state of office
workers, when combined with physiological and behavioral features, enhance the precision of
productivity prediction?
• Research Question 4.3: How do different modalities, specifically wearable devices for
physiological monitoring versus workstation addons for behavioral data, compare in their
effectiveness and accuracy in predicting productivity?
3.5 Research Objective 5
To determine how different IEQ-related interventions can affect office workers’ cognitive performance
and psychophysiological states.
• Research Question 5.1: How does exposure to different white noise levels affect the cognitive
performance, creativity, and physiological responses of office workers?
• Research Question 5.2: What are the stress and attention restorative effects of different light
illuminations and color temperatures?

15
Chapter 4: Working from Home During the COVID-19 Pandemic: Impact on
Office Worker Productivity and Work Experience
With the COVID-19 pandemic, organizations embraced Work From Home (WFH). An important
component of transitioning to WFH is the effect on workers, particularly related to their productivity and
work experience. The objective of this study is to examine how worker-, workspace-, and work-related
factors affected productivity and time spent at a workstation on a typical WFH day during the pandemic.
For that an online questionnaire was designed and administered to collect the necessary information. Data
from 988 respondents were included in the analyses. The following sections of this chapter are organized
as follows: Section 4.1 explains in detail the data collection procedure and describes the control, dependent
and independent variables selected based on the framework adopted in this study. Section 4.2 provides a
summary of the statistical analysis conducted with focus on relative impact of variables within each of the
three pillars. Section 4.3 offers a detailed interpretation of the study findings and identifies key
considerations for implementing WFH as an organizational strategy in future work. Finally, Section 4.4
summarizes the conclusions.
4.1.Methods
Built on the background literature that examines the relationships of the three pillars of future work, this
study aimed to understand how a wide range of worker, workspace, and work context factors affected
productivity and time spent at a workstation on a typical WFH day during the pandemic. This section
explains in detail the data collection procedure and describes the control, dependent and independent
variables selected based on the framework adopted in this study.
4.1.1. Respondents
An online questionnaire was designed and administered through Qualtrics Panel Services for 45-days from
April 27, 2020 to June 11, 2020. The study was reviewed by the Institutional Review Board of the
University of Southern California and was approved as exempt research (UP-20-00339 IRB study number).
Following initial email and public posts to social media pages, 1,409 respondents completed the survey
voluntarily without any compensation. Of the 1,409 responses, 91 responses were screened out because the
individuals were not working from home at a workstation during the COVID-19 pandemic. Furthermore,
330 responses were excluded because the respondents completed less than a quarter of the survey. Our final
sample consisted of the remaining 988 respondents (558 women, 317 men, 113 unreported) who ranged
from 18 years old to 80 years old (M = 40.9 years, SD = 13.1 years). Respondents were primarily Caucasian
(60.9%), followed by Asian (24.6%), Hispanic or Latinx (9.3%), Black (2.8%), and another race or ethnicity
(2.4%). Responses were received from 40 states, with the most respondents from California (47.3%); 6.4%
of responses were from outside of the U.S. and 10.5% of respondents did not provide their location. Level
of education was distributed with 28.6% of the respondents having a 4-years college degree or less, 37.2%
having a graduate or professional degree, and the remaining 34.2% having a doctorate degree. Most
respondents indicated being employed full-time (82.8%) as opposed to working part-time, being students,
or contractors.
4.1.2. Procedure
After consenting, respondents were asked “Does your job require you to use a workstation (e.g., desk,
computer terminal, laptop) most of the day, and are you working (or have worked) from home due to

16
COVID-19 or a stay-at-home mandate?” Only respondents who indicated “yes” to this question continued;
others were told that they were not eligible for the study and thanked. Eligible respondents continued to
complete the survey. Demographic measures included gender, age, location, and employment status and the
other demographic control variables described below as “worker characteristics.” Respondents reported
their perceived level of productivity and indicated the difference in time spent at a workstation both
compared to pre-pandemic levels. These two items served as our dependent variables of work outcome as
described below. In addition, respondents answered numerous questions related to their workspace and their
work context. All questions were optional. Upon reaching the end of the survey, respondents were thanked
for their participation and asked to share a link to the anonymous survey with others who were working
from home during the COVID-19 pandemic.
4.1.3. Measures
The survey comprised of three categories, with questions asking about the worker demographics, workspace
characteristics, work conditions and a final category investigating the work performance and time spent at
the workstation when work is performed from home during the pandemic period. First, worker
characteristics were examined and used as control variables in our analyses. Second, respondents provided
information related to the context of their physical workspace and how they conducted work within their
new WFH situations. These context items were examined as independent variables in our analyses.
Specifically, we examined individual and controlled multivariate contributions to two dependent variables
that served as indicators of work performance during the pandemic’s WFH period as compared to prepandemic work performance.
Worker characteristics: Multiple demographic characteristics were collected to describe our sample as
reported above, as well as to serve as control variables in statistical analyses. In addition to those items
already reported, respondents were also asked about their occupations. Occupational categories were
provided as answer choices based on the occupational categories of the U.S. Bureau of Labor [183]. The
full list of occupational categories was reduced to the following general groups for analysis: business and
office (receptionist, office manager, administrative assistant, etc.), engineering and architecture, education
and arts, healthcare and social services, computer sciences and mathematics, basic scientists, and services
and physical occupations. Respondents indicated their annual income from among four choices: less than
$50K, between $50K and $100K, between $100K and $150K and, more than $150K. Finally, respondents
rated their overall physical and mental health status relative to their status before WFH, using a 5-point
Likert scale with 1 being much lower and 5 being much higher.
Workspace context : Workers indicated whether they had a dedicated space to conduct work in their homes,
selecting from choices: “Yes, I have a dedicated room for work activities (e.g., home office, library, study),”
“Yes, I have a dedicated workspace with other uses (e.g., kitchen, living room, bedroom),” and “No, I work
in a variety of spaces, rooms, or locations around my home (e.g., couch, bed, dining table).” Respondents
indicated if other people (e.g., family, roommates) were usually present in the same space while working
and if they used any of the following items in their workspace: regular office desk, standing office desk,
make-shift desk (e.g., table), adjustable office chair, non-adjustable chair, laptop/tablet computer, desktop
computer, adjustable monitor, non-adjustable monitor, peripheral keyboard, peripheral mouse/trackpad,
foot rest, document holder, natural light, task light, adjustable thermostat, local temperature control. Finally,
we asked the workers to rate their level of satisfaction across multiple indoor environmental quality (IEQ)
factors using a 5-point Likert scale, with 1 being extremely dissatisfied and 5 being extremely satisfied.
Satisfaction with the visual environment was calculated as the average score of satisfaction with natural

17
lighting, electric lighting and glare. Satisfaction with the thermal environment was calculated as the average
satisfaction with the indoor air temperature and humidity. Satisfaction with air quality and noise were
individually rated.
Work context: Respondents identified if they had adjusted their work schedule due to working at home by
selecting either or both: “I now schedule my work hours around others” or “I have adjusted my work hours
(earlier/later, switched days of week, shorter/longer).” The presence (yes) or absence (no) of other
individuals or pets in the home were indicated across the following categories: independent adult (other
than respondents themselves), dependent adult (e.g., special needs, geriatric care), teenage child (13-18),
school-age child (6-12), toddler (2-5), infant (< 2 years), pets (e.g., dogs, cats). Respondents indicated ways
in which their home workspace was obtained selecting any responses that were true among the following
choices: “I purchased new items for myself,” “My employer purchased new items for me,” “I brought items
home from my office,” and “I did not get anything new.” We also asked respondents to rate their
communication levels with their coworkers, their workload expectations or requirements, and distractions
while working during the WFH period relative to their status before WFH. For these questions, a 5-point
Likert scale was used with 1 being much lower and 5 being much higher.
Work experience: We examined two variables of interest relative to understanding changes in how work is
conducted during WFH relative to pre-pandemic, in-office engagement: productivity and hours spent at a
workstation on a typical workday. To assess productivity, respondents rated their productivity relative to
the status before WFH using a 5-point Likert scale with 1 indicating much lower productivity, 3 indicating
the same as before, and 5 indicating much higher productivity. Ratings for “relative productivity” were
normally distributed around an average value of 2.90 (SD = 1.16). Respondents provided an estimate of
the duration of time they engaged at their workstation on a typical workday before and after WFH during
the pandemic. Six choices were provided starting with the first choice as “less than 2 hours” and the sixth
choice of “more than 10 hours,” with an increment of 2 hours between each choice and its successor. The
“change in time spent at a workstation” was calculated as the difference between the number of hours
reported after WFH due to the pandemic from the hours reported before WFH. On average, the time spent
at a workstation increased by 1.46 hours (SD = 3.00) during WFH. A correlation analysis was conducted to
test for any association between the two outcomes, finding very low association between relative
productivity and change in time spent at a workstation (r=0.12, N=962, p<0.001).
4.1.4. Data Analysis
We examined, in turn, the impact of the three types of predictor variables on our work performance
outcomes (i.e., relative productivity and change in time spent at a workstation): worker characteristics,
workspace context, and work context. For each type of predictors, we conducted descriptive analyses,
followed by individual tests of each predictor using t-test and Pearson correlations. Finally, linear regression
models were used to examine the individual predictors (within a type) against all other variables in terms
of predicting unique variance in the two work performance outcomes. Specifically, within each regression
model, we used: (1) the worker characteristic control variables as predictors, (2) the worker characteristic
control variables and -in a second step- the workspace context variables as predictors, or (3) the worker
characteristic control variables and -in a second step- the work context variables as predictors. Within the
regression models, dummy coding was used for all categorical variables. “Business and office” was used
as the reference category for occupation because it was the most frequently selected, and “less than $50K”
served as the reference category for income because the median income for the U.S. falls in this category
[48].

18
4.2.Results
4.2.1. Worker Characteristics
Respondents had an average age of 40.90 years (SD = 13.10 years). Female workers accounted for 56.5%
of the respondents, whereas 32.1% of respondents were male workers and the remaining 11.4% were
missing or preferred not to say. Respondents primarily worked in occupations categorized as business and
office (29.1%), engineering and architecture (24.6%), education and arts (22.1%), followed by respondents
in healthcare and social services (9.3%), computer sciences and mathematics (8.2%), basic scientists
(4.2%), and services and physical occupations (2.6%). The largest percentage of respondents reported
annual income between $50K and $100K (40.6%), with the remaining respondents almost evenly
distributed among the three remaining income levels: less than $50K (19.0%), $100K - $150K (21.7%),
and more than $150K (18.8%). The average physical health status was (M=2.84, SD=0.87) and the mental
health status was (M=2.70, SD=0.93).
We first tested for any differences in productivity levels and difference in time spent at the workstation
based on worker characteristics. There was a significant effect of gender on relative productivity (female
M=2.94, SD=1.17 vs. male M=2.78, SD=1.15; t(864)=-1.97, p=0.050) but not on the change in time spent
at the workstation (female M=1.65 hours (99 minutes), SD=2.95 vs. male M=1.37 hours (82.2 minutes),
SD=2.95; t(851)=-1.33, p=0.181). Age correlated positively and minimally (albeit significantly) with
relative productivity (r=0.12, N=746, p<0.001), while the correlation between age and change in time spent
at the workstation was not significant (r=0.06, N=736, p=0.122).
Mean and standard deviations for relative productivity and change in time at a workstation among the
occupational categories and income levels are presented in Table 1. Significant differences across
occupational categories were noted in both relative productivity (F(6, 847) = 3.28, p=0.003) and change in
time spent at the workstation (F(6, 835) = 4.90, p<0.001). Post hoc comparisons using the Tukey HSD
suggest that relative productivity was significantly lower for respondents with jobs in “engineering and
architecture” (M=2.74, SD=0.08, p=0.011), “computer sciences and mathematics” (M=2.77, SD=0.14,
p=0.032), and “healthcare and social services” (M=2.70, SD=0.13, p=0.026) as compared with increases
reported by those in the “scientists” category (M=3.50, SD=0.2). Although workers across occupational
categories all reported an increase in the time spent at a workstation after WFH, individuals with jobs in
“healthcare and social services” (M=2.63, SD=0.33) reported a change in time that was significantly higher
than those in “engineering and architecture” (M=1.25, SD=0.20, p=0.021) and “business and office”
(M=1.05, SD=0.18, p=0.012) categories. Individuals in the “business and office” category showed a
significantly lower change in time at a workstation as compared to individuals in “education and arts”
(M=2.05, SD=0.22, p=0.012).
While no difference by income category was noted in the change in time spent at the workstation (F(3, 764)
= 1.60, p=0.191), relative productivity differed across income levels (F(3,777)=3.47, p=0.021). Post hoc
analyses show that productivity was significantly lower for workers earning less than $50K (M= 2.64,
SD=0.10) than those earning $50K-$100K (M=3.00, SD=0.06, p=0.011) or $100K-$150K (M=2.98,
SD=0.09, p=0.050). Finally, while workers’ productivity significantly positively correlated with both
physical (r=0.22, N=880, p<0.001) and mental health (r=0.35, N=881, p<0.001) statuses, correlations
between the change in time spent at the workstation and physical health (r=-0.03, N=861, p=0.362) or
mental health (r=-0.06, N=862, p=0.072) were not significant.

19
Table 1. Relative productivity and change in time spent at workstation compared to pre-pandemic
Relative Productivity,
[1-5]
Change in Time at Workstation,
hours per day
Occupational Categories
Business and office 3.00 ± 0.07 1.05 ± 0.18
Engineering, architecture 2.74 ± 0.08 1.25 ± 0.20
Education, arts 2.94 ± 0.08 2.05 ± 0.22
Computer sciences and mathematics 2.77 ± 0.14 1.83 ± 0.34
Healthcare and social services 2.70 ± 0.13 2.63 ± 0.33
Service and physical occupations 2.68 ± 0.24 0.57 ± 0.63
Scientists 3.50 ± 0.20 1.88 ± 0.50
Income
Less than $50K 2.64 ± 0.10 1.28 ± 0.25
Between $50K and $100K 3.00 ± 0.06 1.41 ± 0.17
Between $100K and $150K 2.98 ± 0.09 1.92 ± 0.23
More than $150K 2.84 ± 0.10 1.70 ± 0.25
To answer our first research question, regression analysis was employed to understand the effect of workers
characteristics on the WFH experience during the pandemic. The results suggest that worker characteristics,
measured in this survey, significantly predicted respondents’ relative productivity (F(13, 720) = 10.58,
p<0.001, R2 = 0.16) and change in time spent at workstation (F(13, 705) = 2.31, p=0.005, R2 = 0.10).
Results for individual characteristics included the model are presented in Table 2. Controlling for all other
worker characteristic variables, there was a significant difference in relative productivity between workers
in the “healthcare and social services” and those in the reference of “business and office” (b=-0.38,
p=0.003). Age (b=0.01, p=0.005), physical health status, (b=0.11, p=0.029) and mental health status
(b=0.36, p<0.001) each uniquely predicted a significant increase in workers’ productivity during the WFH
period. A positive effect of working in the “healthcare and social services” also existed for change in time
spent at a workstation (b=0.92, p=0.007), along with a negative effect of mental health status (b=-0.30,
p=0.024) once controlling for all the other worker characteristic variables.
Table 2. Effects of worker characteristics on relative productivity and time spent at a workstation
Worker characteristic
variables Relative Productivity Change in Time at
Workstation
b SE p-value b SE p-value
Gender -0.17 0.09 0.059 -0.14 0.25 0.584
Age 0.01** 0.01 0.005 0.01 0.01 0.163
Engineering, architecture -0.10 0.11 0.321 -0.23 0.28 0.416
Education, arts -0.08 0.10 0.384 0.33 0.27 0.213
Computer sciences and
mathematics -0.16 0.13 0.230 0.38 0.36 0.295
Healthcare and social
services -0.38** 0.13 0.003 0.92** 0.34 0.007
Service and physical
occupations -0.27 0.18 0.125 -0.91 0.47 0.056
Scientists 0.16 0.17 0.343 0.70 0.48 0.144
Between $50K and $100K 0.07 0.06 0.299 -0.21 0.17 0.229
Between $100K and $150K 0.03 0.07 0.683 0.01 0.20 0.988
More than $150K -0.09 0.08 0.227 0.21 0.21 0.335
Physical health status 0.11* 0.05 0.029 0.03 0.14 0.847

20
Mental health status 0.36*** 0.05 <0.001 -0.30* 0.13 0.024
4.2.2. Workspace Context
Table 3 presents the mean, standard deviation (SD) of the continuous predictors, and the frequency and
percentage of the categorical ones for the workspace context related variables.
Table 3. Workspace context variables: Statistical Overview
Categorical Variables Frequency Percentage
I have a dedicated room for
work activities 317 33.0
I have a dedicated workspace
with other uses 483 50.3
I work in variety of spaces,
rooms, locations 160 16.7
Other people are present while
I’m working 447 47.6
I have a regular office desk 469 47.4
I have a standing office desk 56 5.7
I have a make-shift desk 431 43.6
I have an adjustable office chair 450 45.5
I have a non-adjustable chair 383 38.7
I have a laptop/tablet computer 860 87.0
I have desktop computer 222 22.4
I have an adjustable monitor 351 35.5
I have a non-adjustable monitor 186 18.8
I have a peripheral keyboard 419 42.4
I have a peripheral
mouse/trackpad 549 55.5
I have a footrest 104 10.5
I have a document holder 126 12.7
I have natural light (windows) 808 81.7
I have a task light 339 34.3
I have an adjustable thermostat 445 45.0
I have a local temperature
control 414 41.9
Continuous Variables Mean Standard Deviation
Satisfaction with the visual
environment 3.93 0.83
Satisfaction with the thermal
environment 4.00 1.06
Satisfaction with air quality 4.14 0.84
Satisfaction with noise 3.48 1.22
An ANOVA (Analysis Of Variance) revealed a significant effect of the type of workspace on relative
productivity (F(2, 857)=3.49, p=0.031). Tukey HSD (Honestly Significant Test) tests showed that
productivity for “I have a dedicated space for work activities” (M=3.00, SD=0.07) was significantly higher
than that of “I work in a variety of places” (M=2.69, SD=0.09, p=0.021), but the “I have a dedicated space
with other uses” (M=2.89, SD= 0.05) did not statistically differ from the other two conditions. There was

21
also a significant omnibus effect on the difference in time spent at workstation (F(2,931)=3.40, p=0.032).
Specifically, difference in time for the “I have a dedicated space for work activities” (M=1.75 hours (105
minutes), SD=0.17) was statistically higher than that of “I work in a variety of places” (M=0.98 hours (59
minutes), SD=0.24, p=0.021), but the “I have a dedicated space with other uses” (M=1.48, SD= 0.14) was
not significantly different from the two other conditions.
An independent t-test reveal that workers who do not have other people present (M=3.01, SD=1.15) were
significantly more productive than those who do have (M=2.80, SD=1.14; t(841)=2.25, p=0.011). There
was no difference in time spent at workstation based on whether respondents have people present while
working or not (t(911)=0.21, p=0.846).
Table 4. Work traits linked to relative productivity and time spent at workstation
Relative Productivity Change in time spent at workstation
Yes No t-value p-value Yes No t-value p-value
I have a regular office
desk
2.87 ±
1.14
2.91 ±
1.17 0.48 0.631 1.73 ±
2.90
1.23 ±
3.06 -2.57* 0.010
I have a standing
office desk
3.04 ±
1.05
2.88 ±
1.16 -0.95 0.340 2.18 ±
3.02
1.43 ±
2.99 -1.79 0.073
I have a make-shift
desk
2.87 ±
1.15
2.90 ±
1.15 0.32 0.749 1.34 ±
2.94
1.57 ±
3.03 1.19 0.233
I have an adjustable
office chair
2.86 ±
1.15
2.91 ±
1.15 0.68 0.498 1.72 ±
3.00
1.25 ±
2.97 -2.43* 0.015
I have a nonadjustable chair
2.89 ±
1.15
2.89 ±
1.16 -0.02 0.985 1.41 ±
2.89
1.50 ±
3.06 0.49 0.623
I have a laptop/tablet
computer
2.88 ±
1.15
2.96 ±
1.12 0.73 0.462 1.53 ±
2.99
1.01 ±
3.02 -1.76 0.079
I have desktop
computer
2.94 ±
1.15
2.87 ±
1.15 -0.73 0.466 1.57 ±
3.17
1.44 ±
2.94 -0.56 0.576
I have an adjustable
monitor
2.95 ±
1.14
2.86 ±
1.16 -1.10 0.270 1.59 ±
3.00
1.40 ±
2.96 -0.93 0.352
I have a nonadjustable monitor
2.90 ±
1.14
2.89 ±
1.16 -0.15 0.880 1.61 ±
2.80
1.43 ±
3.04 -0.72 0.471
I have a peripheral
keyboard
2.85 ±
1.18
2.92 ±
1.13 0.87 0.383 1.65 ±
3.07
1.33 ±
2.93 -1.63 0.103
I have a peripheral
mouse/trackpad
2.87 ±
1.14
2.91 ±
1.17 0.47 0.634 1.59 ±
2.89
1.31 ±
3.11 -1.44 0.149
I have a footrest 2.78 ±
1.14
2.90 ±
1.15 0.96 0.334 1.59 ±
2.57
1.45 ±
3.04 -0.45 0.653
I have a document
holder
2.98 ±
1.20
2.88 ±
1.14 -0.90 0.369 1.35 ±
2.91
1.49 ±
3.00 0.48 0.630
I have natural light
(windows)
2.90 ±
1.14
2.83 ±
1.21 -0.68 0.497 1.51 ±
2.93
1.28 ±
3.26 -0.93 0.353
I have a task light 2.93 ±
1.12
2.86 ±
1.17 -0.87 0.385 1.64 ±
3.01
1.38 ±
2.99 -1.26 0.208
I have an adjustable
thermostat
2.93 ±
1.13
2.85 ±
1.17 -1.08 0.280 1.53 ±
3.04
1.42 ±
2.96 -0.56 0.572
I have a local
temperature control
2.93 ±
1.15
2.86 ±
1.15 -0.96 0.335 1.41 ±
2.94
1.52 ±
3.03 0.55 0.581
*** p<0.001 **p<0.01 *p<0.05

22
Differences in relative productivity and change in time spent at the workstation based on individual
workspace components are presented in Table 4. Independent t-tests reveal that no item of equipment
showed a significant relationship with worker productivity. On the other hand, having a regular desk
(t(955)=-2.57, p=0.010) or an adjustable office chair (t(955)=-2.43, p=0.015) had a significant relationship
with change in time spent at workstation. Specifically, workers who had a regular desk (M= 1.73 hours (103
minutes), SD= 2.90) or and adjustable office chair (M= 1.72 hours (102 minutes), SD= 3.00) engaged more
at their workstation compared with those who do not have such a desk (M= 1.23 hours (74 minutes), SD=
3.06) or an adjustable chair (M= 1.25 hours (75 minutes), SD= 2.97). Finally, weak correlations were noted
among satisfaction with the visual environment (r= 0.10, N=884, p=0.003), satisfaction with the thermal
environment (r=0.14, N=879, p<0.001), satisfaction with air quality (r= 0.07, N=881, p=0.034), satisfaction
with noise (r=0.09, N=877, p=0.008) and relative productivity, and no significant correlations existed
between these parameters and change in time at the workstation.
To answer our second research question, regression models examining workspace context factors as
predictors of relative productivity (F(37,687)=4.38, p<0.001, R2=0.20) and change in time spent at
workstation (F(37, 672)=1.57, p=0.021, R2=0.03) were both significant (Table 5). While controlling
worker characteristics and all other workspace variables, reporting of “I have a dedicated room for work
activities” (b=0.16, p=0.020) uniquely predicted significantly higher relative productivity than before the
WFH transition. In contrast, reporting “I work in a variety of spaces” (b=-0.16, p=0.022) uniquely predicted
significantly lower relative productivity. Despite low individual correlation to the outcome, once controlling
for all other factors, satisfaction with the thermal environment predicted greater worker productivity
(b=0.12, p=0.024). No individual workspace variables were unique predictors of the difference in time
spent at the workstation.
Table 5. Worker & workspace context linked to productivity and workstation time changes
Variable Relative Productivity Change in time
spent at workstation
b SE p-value b SE p-value
Gender -0.18 0.09 0.059 -0.29 0.25 0.256
Age 0.01* 0.01 0.039 0.01 0.01 0.503
Engineering, architecture -0.14 0.11 0.176 -0.25 0.29 0.369
Education, arts -0.06 0.10 0.550 0.34 0.27 0.214
Computer sciences and mathematics -0.14 0.14 0.305 0.41 0.37 0.265
Healthcare and social services -0.38** 0.13 0.004 0.98** 0.35 0.005
Service and physical occupations -0.21 0.18 0.246 -1.03 0.48 0.102
Scientists 0.24 0.18 0.183 0.63 0.48 0.189
Between $50K and $100K 0.08 0.07 0.216 -0.08 0.18 0.628
Between $100K and $150K 0.03 0.07 0.730 -0.07 0.20 0.719
More than $150K -0.11 0.08 0.183 0.15 0.22 0.469
Physical health status 0.13* 0.05 0.017 -0.05 0.15 0.702
Mental health status 0.35*** 0.05 <0.000 -0.21 0.14 0.125
I have a dedicated room for work activities 0.16* 0.07 0.020 0.18 0.19 0.341
I work in a variety of spaces -0.16* 0.07 0.022 -0.17 0.19 0.355
Other people are present while I’m working 0.01 0.01 0.688 0.01 0.01 0.074
I have a regular office desk -0.01 0.13 0.994 0.31 0.34 0.362
I have a standing office desk -0.01 0.18 0.941 0.61 0.50 0.222
I have a make-shift desk 0.02 0.12 0.895 0.12 0.32 0.717
I have an adjustable office chair -0.06 0.13 0.629 0.28 0.35 0.423
I have a non-adjustable chair 0.06 0.12 0.622 0.19 0.32 0.569

23
I have a laptop/tablet computer -0.13 0.14 0.364 0.48 0.39 0.215
I have desktop computer -0.07 0.12 0.554 0.10 0.33 0.762
I have an adjustable monitor 0.08 0.11 0.459 0.31 0.29 0.299
I have a non-adjustable monitor 0.04 0.12 0.697 -0.11 0.32 0.742
I have a peripheral keyboard 0.01 0.11 0.940 -0.10 0.31 0.747
I have a peripheral mouse/trackpad -0.04 0.11 0.704 0.11 0.29 0.722
I have a footrest -0.17 0.14 0.207 -0.12 0.36 0.734
I have a document holder 0.08 0.13 0.528 -0.49 0.35 0.158
I have natural light (windows) 0.09 0.12 0.433 -0.04 0.32 0.887
I have a task light 0.05 0.09 0.531 0.20 0.25 0.417
I have an adjustable thermostat -0.03 0.09 0.739 0.35 0.24 0.136
I have a local temperature control (fan, heater) 0.05 0.08 0.541 -0.27 0.23 0.244
Satisfaction with the visual environment 0.02 0.06 0.367 -0.36 0.15 0.732
Satisfaction with the thermal environment 0.12* 0.05 0.024 0.05 0.15 0.736
Satisfaction with air quality -0.05 0.07 0.457 -0.08 0.19 0.677
Satisfaction with noise -0.03 0.04 0.677 -0.04 0.10 0.121
*** p<0.001 **p<0.01 *p<0.05
4.2.3. Work Context
Table 6 presents the mean, standard deviation (SD) of the continuous predictors, and the frequency and
percentage of the categorical ones for the work context related variables, and Table 7 presents evaluation
of differences in relative productivity and change in time spent at the workstation based on work context
variables. Respondents who adjusted their work hours (earlier or later work schedule, switched days of
week for work, shorter/longer) showed a significant increase (t(955)=-3.23, p<0.001) in time spent at the
workstation (M=1.65, SD=3.01) compared to those who did not adjust their hours (M=0.94, SD=2.90).
Furthermore, respondents who purchased new items for themselves showed a significantly higher change
in time spent at the workstation (M=1.88, SD=2.96) compared with those who did not (M=1.24, SD=2.99).
Table 6. Work context variables: Statistical overview
Categorical Variables Frequency Percentage
I now schedule my work hours around others 368 36.6
I have adjusted my work hours 722 73.4
At least 1 independent adult lives with me 816 84.2
At least 1 dependent adult lives with me 65 9.3
At least 1 teenager lives with me 136 18.8
At least 1 school age child lives with me 158 21.5
At least 1 toddler lives with me 110 15.6
At least 1 infant lives with me 63 9.4
At least 1 pet lives with me 408 50.2
I purchased new items for myself 342 34.6
My employer purchased new items for me 88 8.9
I brought items home from my office 302 30.5
I did not get anything new 424 42.9
Continuous Variables Mean Standard Deviation
Communication with coworkers 2.66 1.31
Workload expectations or requirements 3.25 1.03
Distractions while working 3.35 1.33

24
Table 7. Productivity and workstation time changes since pre-pandemic based on workspace context
Relative Productivity Change in time spent at workstation
Yes No t-value p-value Yes No t value p-value
I now schedule my
work hours around
others
2.84 ±
1.27
2.93 ±
1.08 1.28 0.203 1.42 ±
3.05
1.48 ±
2.96 0.28 0.782
I have adjusted my
work hours
2.86 ±
1.20
3.00 ±
1.00 1.85 0.064 1.65 ±
3.01
0.94 ±
2.90 -3.23*** <0.001
At least 1 independent
adult lives with me
2.89 ±
1.15
2.91 ±
1.20 -0.17 0.865 1.49 ±
3.04
1.18 ±
2.75 -1.14 0.253
At least 1 dependent
adult lives with me
3.07 ±
1.14
2.87 ±
1.19 1.22 0.224 2.17 ±
2.95
1.48 ±
3.05 -1.67 0.095
At least 1 teenager
lives with me
3.10 ±
1.15
2.86 ±
1.18 2.02* 0.044 1.95 ±
2.70
1.40 ±
3.10 -1.93* 0.036
At least 1 school age
child lives with me
2.91 ±
1.16
2.88 ±
1.19 -0.20 0.838 1.97 ±
3.19
1.42 ±
2.98 -2.02* 0.044
At least 1 toddler lives
with me
2.67 ±
1.32
2.92 ±
1.17 -1.88 0.061 0.76 ±
3.48
1.58 ±
2.95 2.58** 0.010
At least 1 infant lives
with me
2.39 ±
1.37
2.92 ±
1.17 -3.16*** 0.001 0.64 ±
3.81
1.53 ±
3.00 -2.15* 0.032
At least 1 pet lives
with me
2.94 ±
1.16
2.81 ±
1.17 1.54 0.123 1.57 ±
3.14
1.35 ±
2.91 -1.07 0.224
I purchased new items
for myself
2.89 ±
1.20
2.89 ±
1.12 -0.08 0.521 1.88 ±
2.96
1.24 ±
2.99 -3.10** 0.002
My employer
purchased new items
for me
2.87 ±
1.11
2.89 ±
1.16 0.16 0.421 1.83 ±
3.02
1.42 ±
2.99 -1.21 0.242
I brought items home
from my office
2.88 ±
1.17
2.90 ±
1.14 0.56 0.562 1.44 ±
2.88
1.47 ±
3.05 0.12 0.226
I did not get anything
new
2.89 ±
1.14
2.89 ±
1.16 -0.01 0.513 1.29 ±
3.10
1.66 ±
2.89 1.90 0.462
*** p<0.001 **p<0.01 *p<0.05
Additional individual sample t-tests, presented in Table 7, revealed workers with at least one teenager living
at home reported higher relative productivity (M=3.10, SD=1.15) than those without a teenager at home
(M=2.86, SD=1.18; t(644)=2.02, p=0.044), although both means hovered around the neutral response of
3.0. Respondents with school age children at home showed a significantly larger increase (t(721)=-2.02,
p=0.044) in time spent at the workstation (M=1.97 hours (118.2 minutes), SD=3.19), in comparison to those
who do not have school age children at home (M=1.42 hours (85.2 minutes), SD=2.98). Likewise, not
having a toddler (t(692)=2.58, p=0.010) at home seemed to lessen the increase in time spent at the
workstation during WFH: respondents with a toddler at home increased their hours at the workstation less
(M=0.76 hours (45.6 minutes), SD=3.48) than those without a toddler at home (M=1.58 hours (94.8
minutes), SD=2.95). Finally, workers with an infant at home (M=2.92, SD=1.17) were significantly less
productive than those without an infant at home (M=2.39, SD=1.37; t(600)=-3.16, p=0.001), and
respondents with an infant in the home increased their hours at the workstation less (M=0.64 hours (38.4
minutes), SD=3.81) than those without an infant at home (M=1.53 hours (94.8 minutes), SD=3.00; t(661)=-
2.15, p=0.032). Additional t-tests were conducted to analyze the effect of purchasing new items on
productivity and change in time spent at the workstation. The respondents who purchased new items for
themselves show a significantly higher difference in time spent at the workstation (M=1.88, SD=2.96)
compared with those who did not (M=1.24, SD=2.99).

25
We used correlations to evaluate the relationship between our outcome variables and the work context
predictors that are continuous (i.e., communication with coworkers, workload expectations or requirements,
and distractions while working). Communication with coworkers had a moderate positive association with
relative productivity (r=0.46, N=881, p<0.001), as did workload expectations (r=0.32, N=882, p<0.001).
In contrast, distractions while working was moderately negatively associated relative productivity (r=-0.41,
N=881, p<0.001). Although significant, distractions while working (r=-0.07, N=860, p=0.035) and
communication with coworkers (r=0.09, N=860, p=0.013) had essentially no relationship to change in time
spent at workstation. Only workload expectations or requirements (r=0.21, N=860, p<0.001) had a weak
positive association with change in time at the workstation.
The third research question was answered by two regression models examining work context variables as
predictors of work performance outcomes were significant: relative productivity (F(29, 444)=14.96,
p<0.001, R2=0.50) and change in time spent at workstation (F(29, 439)=2.43, p<0.001, R2=0.14). The
results are presented in Table 8. Controlling for the other variables, communication with coworkers
uniquely predicted higher relative productivity (b=0.26, p<0.001), as did workload expectations (b=0.30,
p<0.001). In contrast, in this multiple regression, distractions while working (b=-0.26, p<0.001) and having
an infant at home (b=-0.61, p=0.001) both independently predicted lower relative productivity. Likewise,
adjusting work hours (starting earlier/later, switching days of week, having shorter/longer workdays)
uniquely predicted larger increases in change in the time at the workstation (b=0.81, p=0.012), as did
workload expectations (b=0.40, p=0.003). Finally, distractions while working also independently predicted
smaller differences in hours at the workstation (b=-0.24, p=0.031).
Table 8. Regression: Productivity and workstation time changes against work and worker traits
Variable Relative Productivity Change in time
spent at workstation
b SE p-value b SE p-value
Gender -0.11 0.09 0.219 -0.51 0.30 0.095
Age 0.01 0.01 0.339 0.01 0.01 0.574
Engineering, architecture -0.10 0.10 0.352 -0.25 0.35 0.472
Education, arts 0.01 0.10 0.991 -0.03 0.34 0.915
Computer sciences and mathematics -0.15 0.14 0.315 0.49 0.47 0.306
Healthcare and social services -0.11 0.13 0.400 0.96* 0.43 0.028
Service and physical occupations -0.27 0.17 0.114 -1.52* 0.55 0.007
Scientists 0.24 0.17 0.170 0.26 0.58 0.642
Between $50K and $100K -0.04 0.06 0.500 -0.14 0.21 0.512
Between $100K and $150K 0.02 0.07 0.742 0.15 0.24 0.528
More than $150K 0.02 0.08 0.788 -0.01 0.27 0.954
Physical health status 0.05 0.05 0.310 -0.05 0.18 0.753
Mental health status 0.25*** 0.05 <0.001 -0.23 0.17 0.181
I now schedule my work hours around others -0.08 0.09 0.397 -0.21 0.32 0.498
I have adjusted my work hours -0.01 0.09 0.970 0.81* 0.32 0.012
At least 1 independent adult lives with me 0.01 0.10 0.883 0.09 0.34 0.794
At least 1 dependent adult lives with me 0.25 0.17 0.148 0.74 0.59 0.210
At least 1 teenager lives with me 0.06 0.13 0.624 0.57 0.44 0.202
At least 1 school age child lives with me -0.12 0.13 0.384 0.59 0.45 0.187
At least 1 toddler lives with me 0.024 0.15 0.874 -0.32 0.50 0.525
At least 1 infant lives with me -0.61** 0.17 0.001 -0.72 0.57 0.211
At least 1 pet lives with me 0.01 0.08 0.956 0.11 0.28 0.689
I purchased new items for myself -0.120 0.12 0.323 0.34 0.39 0.384

26
My employer purchased new items for me 0.034 0.16 0.838 -0.72 0.55 0.192
I brought items home from my office -0.02 0.11 0.867 0.04 0.36 0.895
I did not get anything new -0.19 0.13 0.164 -0.17 0.45 0.701
Communication with coworkers 0.26*** 0.03 <0.001 0.20 0.11 0.068
Workload expectations or requirements 0.30*** 0.04 <0.001 0.40** 0.13 0.003
Distractions while working -0.26*** 0.03 <0.001 -0.24* 0.11 0.031
*** p<0.001 **p<0.01 *p<0.05
4.3.Discussion
This study showed that overall productivity level of office workers did not change during the WFH
experience due to the stay-at-home orders relative to their productivity before the pandemic. However,
workers indicated a dramatic increase in the number of hours spent at a workstation by 1.46 hours,
approximately 90 minutes, during a typical WFH day. This section provides a discussion of the theoretical
and practical implications relative to worker characteristics, workspace context and work context on our
two work performance outcomes.
4.3.1. Theoretical Implications
Overall, female workers, older workers, and those at higher income levels were found to be significantly
more productive than their counterparts while WFH during the pandemic. The relationship of age and higher
income has been demonstrated in previous research [184], and the literature shows that women are more
inclined towards remote work than male workers and perceive more benefits and less barriers of WFH,
which boosts their productivity in comparison to the typical work from office [185], [186]. Men are
increasingly putting more effort into household duties [187], but women continue to endure the largest
portion of the housework especially when it comes to childcare [188]. WFH has allowed female workers to
create the much-needed balance between work-family-home responsibilities. Specifically, Colley and
Williamson [189] found that women working from home during the pandemic showed a better integration
of work-family responsibilities, which allowed them to be more productive. Further examination of
evolving gender roles relative to WFH and the work-family-home responsibilities can assist in
understanding how these roles and relationship intersect to support positive work performance and work
well-being. Similarly, additional research is recommended to further investigate the effect of age on the
WFH experience, especially among elderly who might not be as familiar with the remote technology as
younger workers [190].
Our data suggest that workers in the “scientists” category showed the highest productivity levels in
comparison to “engineering and architecture,” “computer sciences and mathematics” and “healthcare and
social services.” It is likely that “scientists” might have more flexibility to WFH as some of their work does
not require them to be physically present at their workplaces. Our results also suggest that “healthcare and
social services” and “education and arts” workers are spending more time at their workstations during the
pandemic compared to other occupational categories where workers who are spending the same time as
before. Such a finding can be attributed to increased use of telemedicine for providing healthcare services
[191] and remote learning classrooms in educational settings [192] as opposed to in-person clinical visits
or educational sessions. Alternatively, workers with typical office jobs (receptionist, office manager,
administrative assistant, etc.) did not witness a major shift in their time spent at the workstation. These data
highlight the importance of examining more than just productivity but understanding how transitioning to
a WFH model can differentially affect how workers engaging in their work across different occupations.
Our findings highlight a few occupational categories that may be at most risk for disruption in their work
patterns and eventual health and well-being should WFH continue in the future, namely healthcare [193]
and education [194] Beyond demographic and work characteristics, our results also demonstrate that

27
workers’ productivity was related to better physical and mental health statuses, which is aligned with prior
work. For instance, Singh et al. [195] showed that physical symptoms such as asthma and allergies
negatively affected 16 work hours per month. Similarly, the number of work hours affected negatively by
mental symptoms such as depression and anxiety reached 20 hours per month. Furthermore, it is postulated
that poor mental and physical health statuses can deteriorate workers’ productivity in the form of
“absenteeism” (through sick leaves) and “presenteeism” (not fully functioning) [196]. The health status of
workers working from home is grasping additional attention [197] and researchers should investigate ways
to promote healthy work conditions and proper means to balance between work and well-being.
Our results suggest that productivity levels were higher for workers who have a dedicated workspace at
home in comparison to those who do not have a dedicated workspace. Previous research studies about WFH
have recommended that workers create their own home work area and recognize it as their workspace [198],
which would help workers mentally shift from the home to the work atmosphere, reduce distractions and
improve their productivity and performance. In fact, the lack of a dedicated workspace when working from
home can disrupt the work experience, increase family-work conflicts and degrade worker productivity
[199].
Our results also suggest that sharing the workspace with another household member decreases the worker’s
productivity. The literature on the relation between the office type and productivity is split: while some
research studies show that a private office increases the worker’s productivity in comparison to open plan
offices, others support that an open plan layout is better in terms of productivity [200]. In the case of WFH,
the home-work environment might be dramatically different than the open plan offices where workers share
the workspace with coworkers. For example, during the pandemic’s WFH period , workers might be sharing
their workspaces with their children who are attending online classes and might be disrupted frequently,
which could hamper productivity [201]. To that end, future research directions should investigate whether
sharing the workstation or workspace with a child have the same impact as sharing it with another working
adult.
Furthermore, we found that the satisfaction with IEQ parameters, especially the thermal environment, and
having a dedicated workspace were positively associated with productivity, while having a desk and
adjustable chair were associated with increased time spent at the workstation. The literature provides a wide
variety of studies that are consistent with these findings. For instance, Geng et al. [202] showed that thermal
satisfaction increases office workers' productivity while postulating that it is the most influential IEQ
parameter affecting productivity. Other studies also showed that satisfaction with the indoor air quality
[137], lighting [10], and noise [203] boost workers productivity. Similarly, having a dedicated workspace
that is not intended for other uses decreases the probability of workers being interrupted by distractions,
and that having a dedicated desk and adjustable chair may result in increased comfort allowing workers to
spend more hours at their workstations. Importantly, owning an adjustable chair can reduce musculoskeletal
risk [204], which in can increase workers engagement at the workstation. We found that workers that
purchased their own equipment also reported being at their workstations for more hours than before the
pandemic. This points to an awareness by the workers regarding the need for a supportive workplace and
illuminates the importance of organizations ensuring the workers have the necessary workstation set-up to
support WFH.
Multiple different associations were noted relative to the type of children present in the WFH context.
Workers who had an infant at home had lower productivity levels, likely resulting from high-level of ongoing attention required in infant care that can become a major distraction for WFH workers [205]. On the
other hand, having a teenager boosted productivity; a possible reason could be that teenagers are more
independent and able to help with household tasks and even take care of younger siblings allowing workers

28
to focus more on their jobs, reducing distractions. Schieman et al. [206] found that having a teenager at
home during the COVID-19 era did not impact the work of parents and did not contribute to the work-life
conflict workers might witness during WFH. Workers with toddlers and infants spent less time at their
workstation compared to before the pandemic and those with school-age children had increased time at
their workstation. School-age children require constant attention from their parents to make sure that they
are following up with schoolwork; thus, WFH parents tend to allow their children to share their workstations
while working to keep an eye on them [207], which could lead to increased time at the workstation. On the
other hand, having toddlers and infants forces the workers to leave their workstation more frequently to
care for their children.
Beyond interpersonal relations within the home, our findings indicate that more communication with
coworkers and higher workload expectations were associated with higher worker productivity. Another
recent study has also demonstrated that workers who maintain frequent and effective communication with
their coworkers tend to perform better during the WFH period [208]. With the positive effect of
communication on workers’ productivity, it is important to identify the most useful communication
technologies to support WFH and to further understand factors influencing successful implementation of
these technologies, such as degree of trust and reliability. The sudden shift to WFH during the pandemic
made some employers feel insecure about their businesses which pushed them to increase their expectations
from their workers –maybe unintentionally—[209]. Thus, a reasonable conclusion for the positive
relationship between productivity and workload expectations, is that workers are working harder, putting
more effort into their work and showing higher productivity levels in order to meet the employer
expectations [210], and to prove they can perform well even under extreme and unexpected conditions
[211]. However, distractions while WFH were a major cause for degraded productivity. Workers at home
are susceptible to all kinds of interruptions while remotely working (e.g., take care of children, completing
household tasks, sharing the workspace with others, etc.) which would negatively affect their productivity
and in most cases oblige them to pause their tasks [201].
4.3.2. Practical Implications
Increased WFH arrangements will likely be a reality beyond the COVID-19 pandemic. Conclusions from
this study provide an nuanced understanding of how the WFH experience can impact work performance,
which provides employers, employees, and other supporting professionals (e.g., ergonomists, therapists)
with information relative to the key considerations for how to mitigate factors that might degrade
performance. Importantly, our findings indicate that better physical and mental health statuses were
associated with improved productivity, highlighting the need to develop supportive policies and practices
targeted to key worker and work characteristics that will assist in balancing performance and well-being in
the WFH context. For example, in our data, older, higher-income, and female workers demonstrated higher
productivity levels compared to their counterparts, which highlights types of workers who may require
increased supports through organizational policies and practices to promote successful WFH (e.g., younger,
lower-income, male). These findings are similar to other recent reports published by the Pew Research
Center noting that older workers did not find the transition to WFH during the pandemic as difficult as
younger workers [212], and workers with low income levels faced more financial distress compared to
those earning high income levels [213].
In addition to worker characteristics, our results provide employers and other professionals with
information related to the way in which work is conducted and the space where work is completed when
WFH. For example, communication with coworkers was associated with increased productivity levels
which indicates the necessity for organizations to identify communication tools that can foster collaboration
during remote work. Additionally, our findings inform recommendations establish the WFH workstation to
improve work performance, which includes identifying a dedicated space for work purposes, isolating the
worker from other individuals in the household, and creating an environment with optimal thermal

29
conditions to boost performance. These data illuminate potential equity issues, as many workers may not
have the capability or resources necessary to create an ideal WFH environment. These issues must be
considered and addressed through policies, practices, and supports as organizations consider widespread
adoption of WFH practices.
Finally, our results indicate a shift in the way in which workers are engaging in their work; specifically, that
workers are spending approximately 90 additional minutes engaging at their office workstations as part of
their workday. Prior to the pandemic, musculoskeletal pain was among the most widespread health issues
threatening office workers [70]. On a positive note, our results indicated that workers who spent more time
at their workstation tended to own an office desk or an adjustable chair. With WFH likely resulting in more
time spent at the workstation than when working in the office setting, organizations should prioritize support
for an ergonomic set up of home workstations. Furthermore, we also found that increased time at the
workstation was associated with shifts in the way work hours were scheduled, and decreased time at the
workstation was reported when there was an infant or a toddler at home. Employers and supporting
professionals must be sensitive to individual employee needs or desires to shift work patterns to support
work-family balance, develop methods to monitor employee performance and well-being relative to any
shifts in work patterns, and identify policies, practices, or supports that will ensure successful reorganization of work patterns to promote positive WFH experiences.
4.3.3. Limitations
Multiple limitations must be considered when interpreting the findings in this study. Firstly, caution should
be used in generalizing the results of this study. The vast majority of respondents to this survey were from
the U.S., and nearly half of the respondents were from California. In addition, although respondents were
well distributed across income categories, our overall sample had a relatively high income as compared to
the median income level in the U.S., which is below $50K. Moreover, our sample also had a much larger
proportion of the respondents having college or graduate degrees than typical across all people in the U.S.
Secondly, productivity in this study was measured relative to the productivity prior to work from home
experience due to the pandemic. Slight variations up and down may or may not result in meaningful changes
in the context of the actual job performance. Finally, the change in time spent at the workstation does not
necessarily equate to overall increase in work hours but is an indication that the workers are spending more
time at their workstations when WFH than when conducting work at their place of employment.
4.3.4. Future Research Directions
Further research is needed to study the specific effects of WFH on gender inequalities both at home and at
work. Along those lines, additional research is also needed to examine the effect of family-work conflicts
and the mechanisms used by workers to cope with distractions caused by other family members and daily
home tasks. Furthermore, it is necessary to investigate the role of communication technology on the WFH
experience and to understand the practical and social implications of relying on digital technologies to
perform WFH. In addition, future research is needed to continue investigating the feasibility of WFH across
different work categories and what role technology plays in enabling successful transitions to different
occupations. Finally, the health status of workers working from home should be given additional attention
and researchers should investigate ways to promote healthy work conditions and proper means to balance
between work and well-being.
4.4.Conclusion

30
With the spread of the novel SARS-CoV2 virus, most office workers were obliged to shift to remote
working almost overnight in mid-March 2020, and the adoption of WFH strategies is likely to persist
beyond the pandemic. This work investigated the worker experience during the pandemic’s WFH period
and focused on two outcomes: relative productivity and the change in time spent at the workstation at a
typical workday. Overall, the results suggest that workers’ productivity levels did not change due to the
remote work transition, but that higher productivity was associated with better mental and physical health
status. Several worker characteristics, workspace, and work contexts were found to be associated with
increased and decreased productivity. Specifically, female workers, older workers, and high-waged workers
showed higher productivity levels. Effective communication with coworkers, satisfaction with the thermal
environment, workload expectations, having a teenager at home, not having an infant at home, and
establishing a dedicated workspace for work activities that has no other uses were all associated with higher
productivity. In addition to impacts on productivity, study data indicated that there was an increase in the
number of hours spent at the workstation by approximately 1.5 hours on a typical WFH day in comparison
to a workday before the pandemic. Longer hours spent at the workstation were associated with having a
school-age child at home, having a desk or an adjustable chair at the workstation, and adjustment of specific
work hours. The findings of this work highlight key considerations for organizational policies and practices
that employers, employees, and other worker support professional can use as a foundation for planning
productive and healthy design of WFH in the future of work.
Chapter 5: Associations Among Home Indoor Environmental Quality Factors
and Worker Health while Working from Home during COVID-19 Pandemic
The outbreak of SARS-CoV-2 virus forced office workers to conduct their daily work activities from home
over an extended period. Given this unique situation, an opportunity emerged to study the satisfaction of
office workers with indoor environmental quality (IEQ) factors of their houses where work activities took
place and associate these factors with mental and physical health. We designed and administered a
questionnaire that was open for 45 days during the COVID-19 pandemic and received valid data from 988
respondents. The research questions answered in this chapter include: (1) How did satisfaction with the IEQ

31
factors relate to the prevalence of physical and mental health symptoms while working from home? (2)
What worker demographics (i.e., age, gender, income) were associated with satisfaction with IEQ and
overall mental and physical health while working from home? (3) What insights regarding the impact of
IEQ factors on health can be concluded based on the transition from traditional office environments to home
office environments? This chapter is organized as follows: Section 5.1 outlines our survey-based
methodology to answer the above-mentioned research questions; Section 5.2 introduces the results of the
analysis; Section 5.3 provides discussions of our findings including the limitations and future research
directions. Finally, Section 5.4 summarizes the conclusions.
5.1.Methodology
5.1.1. Procedure
Using Qualtrics[214], an online survey was administered for a period of 45 days from April 27th, 2020, to
June 11th, 2020. An invitation to complete the survey was distributed through newsletters and was posted
on social media platforms (Facebook, LinkedIn, and Twitter). This study was approved as exempt research
by the Institutional Review Board of the University of Southern California (UP-20-00339 IRB study
number). Participants were screened through an initial question that asked if they had transitioned to
working from home during the stay-at-home mandates due to theCOVID-19 pandemic, and if their job
required them to use a workstation (e.g., desk, computer terminal, laptop) most of the day. A total of 1,409
responses were collected, of which 91 responses were eliminated because they did not meet the inclusion
criterion (i.e., screening question). Responses were further screened based on the percentage completion of
the survey; a response was considered incomplete when less than 25% of the survey was completed. The
final number of responses included for analysis was 988.
5.1.2. Participants Characteristics
Of the 988 valid responses, 56.5% were from female respondents, 32.1% were from male respondents and
the remaining 11.4% were unreported. Respondents were between the ages of 18 and 80, with an average
age of 40.9 years (SD=13.1 years). Most of the respondents reported an annual income between $50K and
$100K (40.6%), with the remaining respondents almost equally dispersed among those making less than
$50K (19.0%), $100K - $150K (21.7%), and more than $150K (18.8%). The race and ethnicity were
distributed with 60.9% of the respondents being Caucasians, 24.5% being Asians, 9.4% being
Hispanic/Latinx, 2.8% being African American and the remaining 2.4% were reported as other. 59.6% of
the responses were received from the West region of the U.S., 7.7% from the Northeast region, 9.0% were
from the Midwest region, 6.8% were from the South region,6.4% were from outside the U.S. (International),
and 10.5% were unreported. The U.S. regional division was adopted based on the United States Census
Bureau [215]. The level of education among respondents was distributed as follows: 28.6% had a 4-year
college degree or less, 37.21% had a graduate or professional degree, and 34.19% had a doctorate degree.
Responses were received from office workers working in a variety of occupations including business
(29.1%), engineering and architecture (24.6%), education and arts (22.1%), healthcare and social services
(9.3%), computer sciences and mathematics (8.2%), basic science (4.2%), and service and physical
occupations (2.6%). Most of the respondents were full-time employees (82.8%), while the remaining
respondents were students (8.7%), part-time workers (5.9%), or contractors (2.6%).
We conducted chi-square analyses for all combinations between the demographic variables (gender,
income, race, region), and only two significant relationships emerged: between income and gender
(χ2=54.07, df=3 and p<0.001), as well as income and region (χ2=46.05, df=12 and p<0.001). Not only were
these two relationships the only ones that reached significance, but they would be expected based on the

32
literature (Bee 2012, Torre and Myrskylä 2014). Thus, although the current sample is biased in terms of
income and region (skewed towards higher income and West region), because there are no interactions with
other demographics, such interactions could not be further biasing the sample.
5.1.3. Measures
Indoor Environmental Quality Factors: Respondents rated their satisfaction with different IEQ factors in
their home workspace using a 5-point Likert scale, with 1 being extremely dissatisfied and 5 being
extremely satisfied. Questions regarding satisfaction with natural lighting (access to daylight), electric
lighting (brightness, no shadows), glare (no reflection on work surface and on computer screen), indoor
temperature, humidity (comfortable levels), air quality (fresh, clean air without unpleasant odor, etc.) and
noise were included in the survey.
Physical Health: Respondents provided a rating of their overall physical health status relative to that before
the stay-at-home mandate using a 5-point Likert scale[218], with 1 being much lower and 5 being much
higher. In addition, respondents indicated the physical symptoms they experienced an increase in since
working from home using a predefined list of physical symptoms (selecting all that apply). The list was
established based on a thorough literature review of all the possible physical issues related to the indoor
environment that building occupants usually experience. The list included a total of 9 symptoms: (1)
cardiovascular symptoms (chest pain, blood pressure, heart rate), (2) chest/lung symptoms (shortness of
breath, chest tightness/pain), (3)digestive symptoms (appetite changes, abdominal discomfort, irregularity),
(4) eye-related symptoms (burning, blurry and/or dry), (5) fatigue or tiredness, (6) headaches or migraines,
(7) nose/throat-related symptoms (dry, runny, or bloody nose; hoarseness) (8) skin-related symptoms
(chapped, itchiness, redness), and (9) musculoskeletal discomfort(discomfort or pain in muscles or joints).
Mental Health: Respondents provided a rating of their overall mental health status relative to that before
the stay-at-home mandate using a 5-point Likert scale, with 1 being much lower and 5 being much higher.
In addition, respondents indicated the mental symptoms they experienced an increase in since working from
home using a predefined list of mental symptoms (selecting all that apply). Again, the list was established
based on a literature review of the possible mental issues related to the indoor environment that building
occupants could experience. The list included8 symptoms: (1) anxiety or nervousness, (2) depression,
sadness or feeling blue, (3) insomnia or trouble sleeping, (4) low motivation or slowed actions, (5) mental
stress, rumination or worry, (6) mood swings, (7) social isolation or decreased interest in social engagement,
and (8) trouble concentrating, maintaining attention or focus.
5.1.4. Data Analysis
The analysis was conducted in three phases. First, we conducted descriptive analyses for IEQ satisfaction,
and physical and mental health related responses. Second, we examined the effects of demographics on
IEQ satisfaction, and overall physical and mental health statuses through a series of Pearson correlations,
t-tests, and ANOVA analyses. Finally, logistic regression models were utilized to further investigate the link
between IEQ satisfaction, demographic variables, and physical and mental health symptoms.
5.2.Results
5.2.1. Descriptive Analysis

33
The means and standard deviations for satisfaction with IEQ factors and overall physical and mental health
status are presented in Table 9. Respondents were least satisfied with noise and most satisfied with air
quality. Satisfaction with IEQ data is skewed towards the upper side of the Likert scale. Respondents
reported worse overall physical health and mental health when working from home during the pandemic
relative to that before the stay-at-home mandate.
Table 9. Descriptive statistics of IEQ satisfaction and overall mental and physical health
Variable Mean Standard
Deviation
Median Mode
Satisfaction with air quality 4.19 0.90 4.00 5.00
Satisfaction with natural lighting 4.11 1.09 4.00 5.00
Satisfaction with humidity 4.08 0.94 4.00 5.00
Satisfaction with indoor temperature 4.00 1.07 4.00 5.00
Satisfaction with electric lighting 3.98 0.99 4.00 4.00
Satisfaction with glare 3.70 1.06 4.00 4.00
Satisfaction with noise 3.48 1.22 4.00 4.00
Overall physical health 2.84 0.87 3.00 3.00
Overall mental health 2.70 0.93 3.00 3.00
Fatigue, musculoskeletal discomfort, eye-related symptoms, and headaches were the most frequently
reported physical conditions that increased after transitioning to work from home, each experienced by
more than 20% of the respondents. Conversely, except for mood swings, at least 1 in 5 respondents reported
an increase in all mental health symptoms. The number of responses related to questions about the different
physical and mental health symptoms are presented in Table 10.
Table 10. Number of responses related to the physical and mental health symptoms
Symptom Yes, I experienced an
increase in this symptom
No, I did not experience
an increase in this
symptom
Physical Symptoms
Fatigue or tiredness 380 542
Musculoskeletal discomfort 350 597
Eye-related symptoms 261 661
Headaches or migraines 201 721
Digestive symptoms 133 789
Skin-related symptoms 82 840
Nose/throat-related symptoms 43 879
Cardiovascular symptoms 34 888
Chest/lung symptoms 28 894
Mental Symptoms
Trouble concentrating, maintaining
attention or focus. 342 587
Anxiety or nervousness 329 600
Low motivation or slowed actions 328 601
Mental stress, rumination or worry 307 622
Insomnia or trouble sleeping 245 684
Depression and sadness 236 693
Social isolation or decreased interest in
social engagement 197 732
Mood swings 144 785

34
5.2.2. Associations among Demographics and IEQ Satisfaction and Health
We compared satisfaction with IEQ factors and the overall physical and mental health ratings among
respondents of different income levels using ANOVA tests and by gender using independent t-tests. No
significant differences were noted by gender, but statistically significant differences were noted among the
income groups for ratings of satisfaction with humidity (F(3, 780)=4.49, p=0.004), air quality (F(3,
780)=2.93, p=0.033), indoor temperature (F(3,780)=3.32, p=0.019), and overall mental health(F(3,
776)=4.05, p=0.007).Mean and standard deviations by income category for these variables are presented in
Figure 1. It is worth noting that the value of N changes between different tests, because not every
respondent answered all the questions. Tukey HSD test [219] was used to test differences in IEQ satisfaction
across the four income categories for significance. This test is generally used to test all pairwise differences
among sample means for significance. The results showed that the satisfaction with humidity for the income
category “more than $150K” was statistically significantly higher than every other income category: “less
than $50K”, “between $50K and $100K” and “between $100K and $150K”. Satisfaction with air quality
was statistically significantly higher for the income category “more than $150K” than the income category
“less than $50K”. Also, the income category “more than $150K” showed statistically significantly higher
satisfaction with the indoor temperature, when compared to the categories “less than $50K” and “between
$100K and $150K”. Finally, the results suggest that the income category “less than $50K” presented a
statistically significantly lower overall mental health rating than the “between $50K and $100K” and
“between $100K and $150K” categories.
Figure 1. Plots of the sample distribution in ratings of satisfaction with humidity, air quality, indoor
temperature, and overall mental health across income categories

35
Additional ANOVA analysis was conducted to determine the effect of region and race on the IEQ
satisfaction and the physical and mental health ratings. The results indicate that statistically significant
differences were noted among regions for ratings of satisfaction with noise only (F(4, 788)=3.21, p=0.012).
Additional Tukey HSD tests showed that the satisfaction with noise in the South region (M= 3.76, SD=1.18)
was statistically higher than the West region (M=3.37, SD=1.21).
A correlation matrix among satisfaction with IEQ factors, overall physical and mental health statuses, and
age is presented in Table 11. It is worth noting that, given the previous ANOVA results, a partial correlation
analysis was performed in Table 11 to control for the effect of income on the variables. Pairwise correlations
among the seven IEQ factors were all statistically significant and positive, ranging from 0.21 to 0.67. A
similar moderate, positive correlation was found between overall physical health and overall mental health
(r=0.45, N=881, p<0.001). Satisfaction with air quality was the only IEQ factor significantly associated
with overall physical health rating(r=0.10, N=878, p<0.001), whereas satisfaction with humidity was the
only IEQ factor not significantly correlated with overall mental health. We note that although statistically
significant, the correlations between overall mental health and IEQ satisfaction were weak, including
natural lighting (r=0.11, N=885, p=0.01), electric lighting (r=0.11, N=876, p=0.001), glare (r=0.13, N=874,
p<0.001), air quality (r=0.15, N=879, p<0.001), noise (r=0.21, N=877, p<0.001) and indoor temperature
(r=0.15, N=879, p<0.001). Age had a statistically significant but minimally positive correlation with
satisfaction of natural lighting (r=0.11, N=747, p=0.003), humidity (r=0.11, N=749, p=0.002), air quality
(r=0.12, N=750, p=0.001), noise (r=0.17, N=750, p=0.001), and indoor temperature (r=0.15, N=750,
p=0.002). The results seem to be highly significant, meaning that they are highly unlikely to have arisen by
chance. However, we found some weak correlations between the variables under study, which means that
the variation in one of the variables is not strongly associated with variation in the other.
Table 11. Correlation: Satisfaction with IEQ factors, overall physical and mental health, and age
1 2 3 4 5 6 7 8 9 10
1 1.00
2 0.53** 1.00
3 0.31** 0.46** 1.00
4 0.28** 0.34** 0.35** 1.00
5 0.36** 0.36** 0.28** 0.63** 1.00
6 0.17** 0.25** 0.27** 0.34** 0.33** 1.00
7 0.29** 0.31** 0.27** 0.64** 0.52** 0.34** 1.00
8 0.06 0.01 -0.03 0.01 0.10* 0.05 0.03 1.00
9 0.09* 0.09* 0.10* 0.06 0.12** 0.16** 0.12** 0.45** 1.00
10 0.09* 0.07 -0.02 0.05 0.06 0.17* 0.11* 0.03 0.06 1.00
1: Satisfaction with natural lighting, 2: Satisfaction with electric lighting, 3: Satisfaction with glare, 4:
Satisfaction with humidity, 5: Satisfaction with air quality, 6: Satisfaction with noise, 7: Satisfaction with
indoor temperature, 8: Overall physical health, 9: Overall mental health, 10: Age
**p<0.001 *p<0.01
5.2.3. Associations among Satisfaction with IEQ Factors and Physical Health Symptoms
Logistic regression analyses were conducted to examine if satisfaction across the IEQ factors is associated
with experiencing a new onset of symptoms within each of physical health categories. The Maximum
Likelihood function in logistic regression analysis results in a chi-square (χ2) value. This value determines
the ability to predict a dependent variable by and independent variable [220].Satisfaction with the IEQ
factors was able to predict participants who experienced new eye-related symptoms [χ2=72.35, df=7 and
p<0.001], fatigue and tiredness [χ2=65.03, df=7 and p<0.001], headaches and migraines [χ2=47.61, df=7
and p<0.001], nose/throat related symptoms [χ2=22.19, df=7 and p=0.002], skin related symptoms

36
[χ2=23.60, df=7 and p=0.001] and musculoskeletal discomfort[χ2=47.16, df=7 and p<0.001]. Regression
models for predicting cardiovascular symptoms [χ2=11.06, df=7 and p=0.136], chest/lung symptoms
[χ2=1.94, df=7 and p=0.96], and digestive symptoms [χ2=13.21, df=7 and p=0.067] were not significant.
The models, presented in Table 12, show that low satisfaction with natural lighting (b=-0.25, p=0.002),
glare (b=-0.36, p=0.001) and humidity (b=-0.25, p=0.037) predicted individuals with eye-related
symptoms. Low satisfaction with noise was a strong predictor of fatigue or tiredness (b=-0.35, p=0.002)
and headaches or migraines (b=-0.36, p<0.001). Nose/throat related symptoms (b=-0.69, p=0.003) and skin
related symptoms (b=-0.48, p=0.007) were only uniquely predicted by low satisfaction with humidity.
Finally, low satisfaction with glare (b=-0.18, p=0.02) uniquely and significantly predicted increased
musculoskeletal discomfort.
5.2.4. Associations among Satisfaction with IEQ Factors and Mental Health Symptoms
Logistic regression analyses were also conducted to examine satisfaction across the IEQ factors on
experiencing new mental health symptoms. The results show that satisfaction with IEQ factors significantly
predicted anxiety [χ2=52.73, df=7 and p<0.001], [χ2=43.36, df=7 and p<0.001], insomnia or trouble
sleeping [χ2=23.60, df=7 and p=0.001], mental stress, rumination or worry [χ2=47.29, df=7 and p<0.001],
mood swings [χ2=28.47, df=7 and p<0.001] and trouble concentrating, maintaining attention or focus
[χ2=52.65, df=7 and p<0.001]. IEQ factors were unable to predict individuals with low motivation or
slowed actions [χ2=13.91, df=7 p=0.053] and social isolation or decreased interest in social engagement
[χ2=13.75, df=7 p=0.056].
The results presented in Table 13 show that low satisfaction with noise (b= -0.20, p=0.002) was the only
unique significant predictor of anxiety. Depression and sadness symptoms were significantly predicted by
low satisfaction with natural lighting (b=-0.22, p=0.003) and noise (b=-0.22, p=0.002). Furthermore, low
satisfaction with air quality (b=-0.18, p=0.045) significantly predicted symptoms related to insomnia and
trouble sleeping. Symptoms related to mental stress, rumination or worry were significantly predicted by
low satisfaction with air quality (b=-0.30, p=0.009) and noise (b=-0.33, p<0.001), while those related to
mood swings were predicted by low satisfaction with noise only (b=-0.32, p<0.001). Finally, low
satisfaction with noise (b=-0.29, p<0.001) and indoor temperature (b=-0.33, p<0.001) significantly
predicted the prevalence of symptoms related to trouble concentrating, maintaining attention or focus.

37
Table 12. Logistic regression: Predicting physical health symptoms from IEQ satisfaction
P1 P2 P3 P4 P5 P6 P7 P8 P9
Natural Lighting 0.04 -0.10 -0.09 -0.25**
-0.04 -0.07 0.25 -0.10 -0.12
Electric Lighting -0.37 0.21 -0.07 0.14 -0.07 -0.05 0.05 0.26 0.03
Glare -0.33 -0.13 -0.07 -0.36**
-0.12 -0.01 -0.05 -0.11 -0.18*
Humidity 0.06 -0.12 -0.17 -0.25*
-0.14 -0.20 -0.69**
-0.48**
-0.08
Air quality 0.12 -0.13 -0.04 0.01 0.02 0.11 -0.10 -0.18 0.01
Noise 0.04 0.06 -0.15 -0.09 -0.35**
-0.36***
-0.20 -0.14 -0.13
Indoor temperature 0.14 0.02 -0.02 0.10 0.06 -0.04 0.14 0.21 -0.05
P1: Cardiovascular symptoms, P2: Chest/lung symptoms, P3: Digestive symptoms, P4: Eye related
symptoms, P5: Fatigue or tiredness, P6: Headaches or migraines, P7: Nose/throat related symptoms, P8:
Skin related symptoms P9: Musculoskeletal discomfort
Table 13. Logistic regression: Predicting mental health symptoms from IEQ satisfaction
M1 M2 M3 M4 M5 M6 M7 M8
Natural Lighting -0.12 -0.22**
-0.03 -0.04 -0.05 -.09 -0.04 0.01
Electric Lighting -0.09 0.04 -0.02 -0.15 -0.06 -0.03 -0.01 -0.06
Glare -0.10 -0.12 -0.12 -0.13 -0.03 0.25 .015 -0.01
Humidity -0.26* 0.14 -0.09 0.01 -0.08 -0.23 -0.05 0.07
Air quality 0.04 -0.03 -0.18* 0.12 -0.30** 0.16 -0.02 0.11
Noise -0.20**
-0.22**
-0.14 -0.10 -0.33***
-0.32***
-0.13 -0.29***
Indoor temperature 0.04 -0.29 0.01 -0.15 -0.16 -0.01 -0.14 -0.33***
M1: Anxiety, M2: Depression and sadness, M3: Insomnia or trouble sleeping, M4: Low motivation, M5:
Mental stress, rumination or worry, M6: Mood swings, M7: Social isolation or decreased interest in social
engagement, M8: Trouble concentrating, maintaining attention or focus.

38
5.3.Discussion & Recommendations for Future Work
5.3.1. Associations among Demographics and IEQ Satisfaction and Health
The literature suggests that socioeconomic status has a major effect on the perception of IEQ among
occupants and consequently on their health and well-being[180], [221]. Our results —in agreement with
the literature—show that respondents with high income (more than $150K) are more satisfied with
humidity, air quality and indoor temperature, which might indicate that low-income (less than $50K)
residents lack proper heating, ventilation, and air conditioning (HVAC)systems. We also found that the lowincome category presented a significantly lower overall mental health rating than the high-income
categories (between $50K and $100K, between $100K and $150K). This outcome could be attributed to:
(1) the inability of low-income households to afford satisfactory housing conditions leading to increased
levels of anxiety, distress, and depression [222]; or (2) the economic uncertainty, fear of unemployment,
falling in debt during the pandemic resulting in psychological stress, and poor mental state [223]; or (3) a
combination of both.
Similar reasoning could also be employed to understand the positive correlation between age and
satisfaction with natural lighting, humidity, air quality, noise, and indoor temperature. Studies have shown
strong associations between age and income[184]; as workers gain more experience, their income levels
tend to get higher which can lead to better housing conditions and consequently an increase in IEQ
satisfaction levels. Furthermore, several research studies have shown that older individuals were more
tolerant to IEQ conditions, leading to higher acceptance and satisfaction with the indoor environment
[224].On the other hand, our results are inconsistent with other reports [153], where middle age workers
(35–54years)were less satisfied with IEQ conditions in comparison to their younger and older counterparts.
These conflicting findings indicate the need for further research efforts to understand how people perceive
the indoor environment with age.
Finally, our results showed no statistical difference in IEQ satisfaction between female and male office
workers at home. Among the various IEQ parameters, the literature heavily focuses on gender differences
in terms of thermal comfort and satisfaction; previous work found that female office workers were less
satisfied and comfortable with their office indoor temperature in comparison to their male coworkers [225].
Traditional offices do not take these individual preferences into consideration which results in thermal
satisfaction differences across gender. Working from home might have allowed both female and male
workers to set IEQ conditions based on their personal preferences which might have led to similar
satisfaction levels.
5.3.2. Associations among Satisfaction with IEQ Factors and Physical Health Symptoms
Our regression analysis shows that all else being equal, respondents with higher satisfaction with natural
lighting, glare, and humidity were less likely to present eye-related symptoms. This finding is consistent
with previous studies conducted in office buildings where workers who are satisfied with the spatial
distribution of light in their indoor environment —whether electric or natural lighting—are less irritated by
glare and present less eye fatigue[226]. Our results also agree with Shin et al. (2018)who found that workers
who reported low satisfaction with humidity were prone to more eye problems in office buildings.
Furthermore, our results suggest that respondents who were more satisfied with noise presented less
prevalence of fatigue, tiredness, headaches, and migraines. Witterseh et al. (2004a)found that workers
reported increased levels of fatigue and headache intensity and dissatisfaction with acoustic conditions of
their workspace when background noise levels increased in their open-plan office. In addition, our results
show that low satisfaction levels with humidity predicted higher prevalence of skin (chapped skin, itchiness,
redness) and nose/throat related symptoms (dry, runny, or bloody nose, hoarseness). Similar findings were

39
reported in [227], where low satisfaction with humidity was associated with a feeling of irritation at the
level of noise and throat. Finally, our analysis shows that low levels of satisfaction with glare predicted
higher prevalence of musculoskeletal discomfort. When a person is exposed to glare, gaze stabilization
becomes challenging which requires head, neck or even body posture adjustment to reach a comfortable
visual state. These adjustments might not be optimal and can contribute to muscle pain development [228].
Recently, Mork et al.(2020) concluded that constant exposure to direct glare conditions leads to visual
discomfort and affects the trapezius muscle (back muscle) and leads to the development of neck pain.
5.3.3. Associations among Satisfaction with IEQ Factors and Mental Health Symptoms
We found that lower satisfaction levels with noise predicted higher prevalence of anxiety and depression
among respondents. Alimohammadi et. al (2010) also found that disturbance and annoyance caused by
excessive noise was associated with increased tendency to show symptoms of anxiety and depression
among white collar employees in office buildings. In our study, depression was also predicted by low
satisfaction with natural lighting. Similar findings were reported by Brown and Jacobs(2011)who found
that subjects reporting inadequate natural lighting in their residential apartments were 1.4 times more likely
to show symptoms of depression compared with those who have sufficient access to daylight.
Furthermore, our results show that respondents who were less satisfied with the air quality were more likely
to have insomnia and to experience trouble sleeping. This is in agreement with previous studies that lower
concentrations of CO2 and particulate matter and higher ventilation rates in the sleeping area showed an
improvement in the quality of sleep and sleep latency (time needed to go from being fully awake to sleeping)
while reducing the number of awakenings during the night [232], [233]. We also found that the prevalence
of mental stress symptoms was predicted by low satisfaction levels with air quality and noise. Similarly,
Thach et. al (2020)conducted an environmental health assessment of several office buildings and concluded
that higher satisfaction with air quality and noise levels was associated with reduced stress.
The results from our logistic regression in Table 13 show that respondents who were less satisfied with
noise conditions in their houses were more likely to present symptoms related to mood swings. Blasio et al.
(2019)showed that background noise caused by irrelevant speech in open-plan offices can lead to increased
sense of annoyance, which builds up negative effects. When workers were forced to work from home, many
of them were sharing the workspace with others in their household, creating a similar scenario to open-plan
offices. Finally, our results show that low satisfaction with noise and indoor temperature predicted the
prevalence of symptoms related to trouble concentrating and maintaining attention or focus. The ability to
concentrate and be attentive are directly related to productivity; therefore, many research studies examined
the effect of IEQ on workers’ concentration and attention capabilities in office buildings. Similar findings
were reported by Monteiro et al. (2018) during a controlled experiment, in an office-like environment; it
was found that subjects who were exposed to high noise levels were less satisfied with the acoustic
conditions of their environment and had reduced attention levels. Our results are also in agreement with the
findings of Lipczynska et al.(2018) who concluded that workers’ concentration and alertness levels
increased with higher level of thermal satisfaction in office buildings.
5.3.4. Implications for research and applications in future home office environments
Based on Table 9, respondents, on average, were highly satisfied with IEQ conditions in their homes. This
high satisfaction level can be attributed to the concept of control over one’s indoor environment. For
example, most traditional office buildings are equipped with HVAC systems that operate between 22–
25°C[237] based on ASHRAE’s Standard 55[238]which does not take into consideration individual
preferences and can result in occupants being dissatisfied with the indoor thermal conditions. Also,
controlling the indoor environment in an open plan office becomes difficult when several workers working
in the same office space have different preferences [239]. The same complexity applies to other IEQ factors;

40
one worker might prefer to open the shades to have access to natural lighting, but this might create glare on
the screen of coworkers. At home, office workers might be able to control their indoor environment to
increase their satisfaction or they might have the option to choose a location that pleases them or choose a
location that does not negatively affect others sharing the same space. Therefore, future research directions
should investigate the difference in control over the indoor environment a worker has between traditional
office environments and home offices and the relation between IEQ satisfaction and overall satisfaction
with house layout area.
Our results demonstrate that respondents were most satisfied by air quality but least satisfied with noise.
From a building design perspective, satisfaction with air quality can be attributed to high ventilation rates
either through HVAC systems or natural ventilation. Higher satisfaction with air quality could be attributed
to the fact that most houses have operable windows, different than most office buildings in the U.S. with
non-operable windows. The low level of satisfaction with noise can be associated with the number of people
in the same house; following the stay-at-home mandate, children, adults, and elderly were all obliged to
remain indoors which inevitably increased the noise within a house. It is worth noting that during the period
of survey distribution, stay-at-home mandates were in order in most of the world which limited outside
noise (e.g., traffic). Despite that, our results show that satisfaction with noise was the lowest among all IEQ
parameters, which means that when outside noise is restored, satisfaction with noise will be even worse.
Therefore, assuming work from home will persist after the end of the pandemic, satisfaction with noise
would remain the worst among all IEQ conditions, thus maintaining the same conclusions. However, to
better understand the reasons behind these satisfaction levels and how they differ among different houses,
future research should focus on examining the house conditions and establish standard evaluation schemes
to assess satisfaction considering the physical attributes of home environments and its surrounding.
Furthermore, our results conclude that income has a significant effect on IEQ satisfaction. Before the
pandemic, workers with different income were sharing the same office environment, perceiving similar IEQ
conditions. In other words, workers’ income had no direct effect on the indoor environmental conditions of
the workspace. When workers were forced to work from home, their income level – a direct indicator of
the housing quality- might have become a major driver of IEQ conditions and the workers’ satisfaction and
health and well-being related consequences. To that end, future research directions can focus on
understanding the interrelations between the socio-economic status of people, their housing conditions, and
the associated mental and physical health consequences; such research effort could create a foundation for
policy makers to integrate social equity and justice with the concept of healthy buildings [180].
Finally, our findings on IEQ satisfaction and its relation to various physical and mental health symptoms
showed a strong agreement with previous work conducted in traditional office environments. Therefore, it
is safe to conclude that the relation between IEQ satisfaction and workers’ health and well-being is universal
and not restrained to the physical environment that work is being conducted within. Thus, special
consideration should be given to home office environment design for work, considering more remote work
is likely to take place even after the pandemic [240]. To that end, employers can support their employees in
creating a more comfortable workspace at home. For example, employers can purchase furniture and
equipment to enhance the setup of their home workstations (ergonomic desk and chair, desk light, desk fan,
etc.), or allow their workers to bring items from their formal offices to make their home workstations more
comfortable.
5.3.5. Limitations and future research directions
While this study provides important contributions to the literature on the relationship of IEQ factors and
occupant health within a home office workspace, findings should be interpreted with some limitations in
mind. First, most of the respondents were from the U.S. and nearly two-thirds of them were from the West
region which may limit generalizability of the results. Similarly, although the income categories were well

41
distributed, the sample under study had a relatively high average income, was comprised of highly educated
respondents, and showed high discrepancy of ethnicities which restricted the analysis based on education
level and race. In consideration of the foregoing, the uneven distribution of the sample could explain the
correlations between satisfaction with the IEQ and between the physical and mental health measures. To
that end, future research should examine the regional, and cultural differences, as well as other demographic
factors in considerations of home office IEQ and worker health. Second, it is important to note that the
abrupt transition to work at home and the additional psychological and physical manifestations caused by
the pandemic itself likely impacted the health and well-being of workers who responded to this survey.
However, we did not ask whether of not the respondents had had COVID-19. As such, our investigations
and conclusions relative to satisfaction with IEQ factors in the home office and responses to health-related
symptoms are limited to associations and do not serve to indicate any direct cause or effect between the
environment and health conditions. Instead, these data provide insight into potential areas where further
investigation or support may be needed as working from home decisions are made in future work. In
addition, spatial-related data was limited to the name of the state the worker is in, without going into the
details of the counties. Given that climate zones are best identified based on counties [241], our analysis
could not examine the effect of climate or living in a rural vs. an urban area on the IEQ satisfaction and as
such its mediator effects on the physical and mental health symptoms. Such analysis become even more
complicated when looking at participants outside the U.S., because international participants were not asked
about their country of residence. Finally, future research directions should investigate the means to collect
physical measurements rather than relying on questionnaires and subjective assessment if regulations,
codes, standards, and design and construction guidance are to be developed.
5.4.Conclusions
Following the spread of the SARS-CoV-2 virus, work from home became a necessity for many office
workers, creating an opportunity to evaluate the relationship between home office environments and worker
health. In our survey sample, higher income workers were more satisfied with humidity, air quality and
indoor temperature of their work environment at home, and these workers reported better overall mental
health in comparison to low-income workers. Age was positively correlated with satisfaction with natural
lighting, humidity, air quality, noise, and indoor temperature. Low satisfaction with natural lighting, glare
and humidity significantly predicted the onset of new instances of eye related symptoms, while low
satisfaction with noise was a strong predictor of increased fatigue or tiredness, headaches, or migraines.
Nose/throat related symptoms and skin related symptoms were only uniquely predicted by low satisfaction
with humidity, and low satisfaction with glare uniquely predicted increased musculoskeletal discomfort.
Low satisfaction with noise predicted new anxiety symptoms, and low satisfaction with air quality predicted
symptoms related to insomnia and trouble sleeping. Low satisfaction with noise and air quality together
predicted increased mental stress, rumination or worry. Finally, low satisfaction with noise and indoor
temperature predicted the prevalence of symptoms related to trouble concentrating, maintaining attention
or focus. The findings provide new insights into IEQ factors within home environments and their
associations to workers health where further investigation or support may be needed to ensure positive
health for employees who opt to continue or transition to working from home in future work.

42
Chapter 6: Stress Appraisal in the Workplace and its Associations with
Productivity and Mood: A Multimodal Machine Learning Analysis
Previous studies have primarily focused on predicting stress arousal, encompassing physiological,
behavioral, and psychological responses to stressors, while neglecting the examination of stress appraisal.
Stress appraisal involves the cognitive evaluation of a situation as stressful or non-stressful, and as a
threat/pressure or a challenge/opportunity. In this study, we investigated several research questions related
to the association between states of stress appraisal (i.e., boredom, eustress, coexisting eustress-distress,
distress) and various factors such as stress levels, mood, productivity, physiological and behavioral
responses, as well as the most effective ML algorithms and data signals for predicting stress appraisal. The
following sections of this chapter are organized as follows: Section 6.1 explains in detail the methodology
adopted. Section 6.2 provides a summary of the results and a detailed interpretation of the study findings.
Finally, Section 6.3 summarizes the conclusions, explains the practical implications of the study as well as
its limitations.
6.1.Methodology
We conducted a controlled experimental procedure to collect physiological and behavioral data for stress
appraisal prediction. This study was conducted according to the guidelines of the Declaration of Helsinki
and approved by the Institutional Review Board of the University of Southern California (UP-21-00484,
Effective Approval Date: 22 July 2021). All participants provided written informed consent. Data collection
occurred between March 11, 2022, and July 5, 2022.
6.1.1. Participants
A total of 48 healthy individuals (20 males and 28 females), primarily undergraduate and graduate students,
volunteered for the experiment. Their mean age was 22.6 years (±2.1 years), and there were no dropouts;
all 48 initially enrolled participants completed the study. Participants underwent a rigorous screening
process using a self-report questionnaire. It is important to note that one prospective participant, who did
not meet our inclusion criteria, was excluded before the study's commencement. This exclusion was based
on predefined criteria, such as vision impairments hindering computer use, psychological sensitivity to
stress-inducing activities, pregnancy, or the use of medication affecting physiological signals. With this
exception, all other participants successfully completed the study without any dropouts.
6.1.2. Physiological, behavioral, and human-computer interaction data
During the experiment, participants were equipped with two distinct physiological monitoring devices,
chosen to align with our research objectives and data collection requirements. Firstly, participants wore an
Empatica E4 wristband [242] , which was selected for its versatility in capturing multiple physiological
parameters. The wristband collected Electrodermal Activity (EDA), Skin Temperature (ST), Blood Volume
Pulse (BVP), and x, y, and z wrist acceleration. These parameters were chosen for their relevance to our
study, as EDA and ST provide insights into emotional arousal and stress levels, while BVP offers
information about cardiovascular responses. Additionally, the wrist acceleration data allowed us to monitor
fine-grained motion-related metrics. Secondly, heart rate data was collected using an H10 Polar chest strap
[243]. We chose this device due to its precision in measuring heart rate, a critical physiological metric for
our study, as it reflects stress levels, emotional states, and physical exertion. In addition to physiological
data, we employed behavioral data collection methods. A Microsoft Azure Kinect DK camera [244] ,

43
strategically positioned atop the computer screen and facing the participant, recorded facial expressions
throughout the experiment. This video data was instrumental in complementing the physiological
measurements and facilitating the analysis of emotional responses. Furthermore, to gain insights into
participants' interactions with the computer, we ran the Mini Mouse Macro logging application [245] in the
background. This application meticulously recorded participants' activities involving the computer's mouse
and keyboard.
6.1.3. Experimental protocol
The experiment took place in a quiet private office using a standard desktop computer. The experiment
simulated 2 different work conditions: low-stress work and high-stress work.
Low-stress work: In this condition, participants had 40 minutes to prepare a PowerPoint presentation about
their favorite movie, book, or television series, that is, a familiar topic. In this condition, participants worked
without being monitored.
High-stress work: In this condition, participants had only 30 minutes to prepare a PowerPoint presentation
about an unfamiliar topic. Participants had to present the scientific and philosophical achievements of two
ancient Greek philosophers and provide their opinions about how these achievements are still shaping
modern human life. The requirements (time and topic) were carefully established to make the completion
of the presentation achievable but at the same time to create a sense of time pressure, heavy workload, and
unfamiliarity with the task. Furthermore, a confederate played the role of a university professor who
monitored the participants using live video, audio, and screen sharing via Zoom video conferencing. An
application on the computer screen showed the professor’s rating of their work, which began at 100 points
and then decreased and increased in a standardized manner across all participants. Changes in the score
appeared at uneven intervals such that the participants could not recognize a pattern, instead, associating
the scoring with the professor noticing a flaw or correction. Participants were informed that the highestscoring individuals would receive the maximum compensation (i.e., $50) and the lowest-scoring individuals
would receive minimal compensation for their time (i.e., $5). At the conclusion of the study, all participants
were informed that their score did not equate to the level of compensation, and everyone received the
maximum compensation (i.e., $50).
Throughout both conditions, a pop-out questionnaire appeared on the computer screen every 5 minutes,
asking the participants to rate their perceived stress level, mood, and productivity. These metrics were rated
using a 0-100 slider, with 0 indicating “I am not stressed at all,” “I am in a bad mood,” or “I feel extremely
unproductive” and 100 indicating “I am extremely stressed,” “I am in a good mood” or “I feel extremely
productive”. In addition, participants appraised their stress as distress (pressure) and eustress
(opportunity/challenge) using the Valencia Eustress-Distress Appraisal Scale (VEDAS), an efficient and
validated tool for appraising stress as perceived levels of distress and eustress [246], [247]. Distress was
assessed using a 6-point scale as 1 (very definitely is not a source of pressure), 2 (definitely is not a source
of pressure), 3 (generally is not a source of pressure), 4 (generally is a source of pressure), 5 (definitely is
a source of pressure), or 6 (very definitely is a source of pressure). Eustress was assessed using a similar 6-
point scale going from 1 (very definitely is not a source of opportunity/challenge) to 6 (very definitely is a
source of opportunity/challenge).
Each condition started with a 5-minute baseline phase, in which participants remained idle and relaxed
while we collected a baseline for all physiological signals (which is typical for stress detection research).
At the end of the baseline phase, participants rated their perceived stress level and mood. Participants were
given the option of taking a maximum 10-minute break if needed between the low and high stress phases;
however, all 48 participants opted to proceed immediately with the next phase of the experiment. The total
duration of the experiment was around 100 minutes.

44
6.1.4. Feature Extraction and Data Processing
The collected data was segmented into 30-second time frames to extract physiological and behavioral
features. This particular time window was chosen based on the findings of Bernardes et al. [248], who
determined that a 30-second duration is the smallest timeframe suitable for obtaining dependable HRV
features that accurately evaluate psychological stress. Therefore, our dataset consisted of 48 participants,
each with 70 minutes of data collection divided into 30-second time windows, resulting in 6720 datapoints.
The final dataset comprised 83 features, including 34 physiological features, 48 behavioral features
(including 3 related to human-computer interactions), 39 facial-related features, 6 features for wrist
acceleration, and 1 feature indicating the participant's gender. Table 14 provides a summary of all the
features analyzed.
Table 14. Features dataset
Type (Number of features) Signal Features Included
Physiological (34)
Electrodermal activity (EDA)
Blood volume pulse (BVP)
Skin temperature (ST)
Mean, Standard deviation,
Median, Minimum, Maximum,
25th & 75th percentile, slope
fitted through the data.
Heart rate
Heart rate variability (HRV)
Mean, Standard deviation,
Minimum, Maximum, rmsdd,
LF peak, HF peak, LF power,
HF power, LF/HF
Behavioral (48)
Facial action units (AUs)
Head rotation
Eye gaze direction
Mean, Standard deviation
Blink Count
Wrist acceleration Mean, Standard deviation
Mouse right clicks
Mouse left clicks
Keyboard keystrokes
Count
Gender (1) Female
Male Binary
Rmsdd: Root Mean Square of the Successive Differences, LF peak: Low-frequency peak, HF peak:
High-frequency peak, LF power: Low frequency power, HF power: High frequency power
Kubios software [249] was utilized to process HRV data and extract multiple time and frequency-domain
indices of the heart rate signal. A moderate artifact correction was employed to identify R-R intervals
varying above or below 0.25 seconds from the average. This method preserved the data's variability while
addressing the presence of any artifacts. Kubios also applied a piecewise cubic spline interpolation method
to generate corrupted or missing values, resulting in a cleaner and more accurate HRV signal. The RRinterval was excluded from the feature set to prevent duplication because of its inverse proportionality with
heart rate, and there was a strong 94% correlation between the two features in the dataset.
Signals obtained from the Empatica E4 were processed before feature extraction to reduce noise, similar to
what was done in a previous study [250]. The BVP and ST signals were filtered using winsorization [251],
a statistical technique that replaces extreme values beyond the 2nd and 98th percentiles. We used the
MATLAB Ledalab toolbox [252] to clean and process the EDA data using a Butterworth low-pass filter,
Hanning smoothing with a window size of 4 adjacent datapoints, and manual artifact correction to remove

45
any noise that might have been caused by movement or other sources of interference. After the cleaning
procedure, we calculated the mean, standard deviation, median, minimum, maximum, 25th and 75th
percentiles, and slope of BVP, EDA, and ST to ensure a thorough assessment of the various dimensions
involved in stress appraisal [253].
For behavioral data in each 30-second time window, we calculated the mean and standard deviation for x,
y, and z wrist accelerations obtained from the Empatica E4, and we utilized OpenFace software [254] to
extract the mean and standard deviation of participants' facial action unit (AU) intensities from the RGB
video captured by the Kinect camera. AUs are predefined facial muscle movements associated with
emotions and are classified as main AUs, head movement AUs, and eye movement AUs. Facial expressions
are a reliable indicator of stress and are suitable for stress detection research [255]. We removed the head
translation vector in the x, y, and z planes from the analysis as it depended on the participant's height and
position in the camera frame. Similarly, we excluded head rotation in the x and y planes as it was highly
correlated with the gaze vector, resulting in duplicate information. A correlation analysis confirmed the
close relationship between these variables, with a Pearson correlation ranging from 89% to 94%. By
removing these features, we avoided redundancy in our analysis. Lastly, we aggregated keyboard strokes
and mouse clicks, which are known to be influenced by cognitive and emotional states and is a relatively
innovative approach to stress prediction that has exhibited encouraging results in recent studies [256], [257].
Some data was missing from the dataset due to technical problems, including keyboard and mouse files for
three participants in the low-stress condition and RGB video files for two others in the high-stress condition.
To address this, an XGBoost model was trained using the existing data from 43 participants to impute the
missing data. Hyperparameter tuning was conducted to optimize the model's performance, and the best
values for the learning rate, maximum tree depth, and number of trees were selected. The optimized
XGBoost model was then used to predict missing data points, as this method preserves the standard
deviation and shape of feature distribution and avoids data loss from deleting rows with missing entries.
Mean/median imputation methods were avoided as they are less accurate [258]. Additionally, robust scaling
was used per participant, a data preprocessing technique to normalize features in machine learning [259].
It employs the interquartile range (IQR) instead of mean and standard deviation, making it robust against
outliers. By linearly transforming the data using the 25th and 75th percentiles (IQR), robust scaling ensures
fair comparisons and accurate modeling, especially in the presence of outliers.
All physiological data were subtracted from individual baseline levels established before the first
experimental condition to increase between-participant validity. Behavioral features were not normalized
against a baseline because facial activation is closely tied to the intensity of facial expressions, which does
not necessarily require normalization across participants. Perceived stress and mood data values were
subtracted from the baseline; thus, the range for the perceived stress and mood variables was between -100
and 100. The baseline for perceived productivity was 0 as the participants were not performing any work;
thus, the range for the perceived productivity variable was between 0 and 100.
6.1.5. Outcome formulation
This study aimed to examine stress appraisal and thus required the creation of a metric to describe the four
states: boredom, eustress, eustress-distress coexistence, and distress. The 6-point scales for eustress and
distress were condensed by grouping the lower three and top three categories into a binary variable as
"Stress not appraised as eustress" or "Stress appraised as eustress" and "Stress not appraised as distress" or
"Stress appraised as distress." The two ratings were combined to classify each datapoint into one of the four
stress states of interest, as shown in Table 15. The resulting dataset was unbalanced, with approximately

46
49% indicating eustress-distress coexistence, 28% indicating boredom, 18% indicating eustress, and 5%
appraised as distress.
Table 15. Stress appraisal formulation and data distribution across four stress appraisal states
Eustress Appraisal Distress Appraisal Stress Appraisal
State Datapoints
Stress not appraised as eustress Stress not appraised as distress Boredom 1890
Stress appraised as eustress Stress not appraised as distress Eustress 1230
Stress appraised as eustress Stress appraised as distress Eustress-distress
coexistence 3270
Stress not appraised as eustress Stress appraised as distress Distress 330
6.1.6. Prediction assessment
The evaluation of prediction performance involved the use of accuracy and average F1 score. All models
presented in the results were subjected to k-fold cross-validation, with a value of 10 for k. This approach
ensured that no participant was used in both the training and testing sets during any iteration of the crossvalidation process, thereby enhancing the reliability and generalizability of our results.
6.2.Results and discussion
6.2.1. Perceived stress, mood, and productivity levels across stress appraisal states
To address the first research question of this study, we conducted three ANOVA tests to explore how
different stress appraisal states relate to the perception of stress, mood, and productivity. Figure 2 presents
the means and variances of these metrics across the four stress appraisal states. Statistically significant
differences were identified in perceived stress (F(3, 6716) = 271.82, p<0.001), mood (F(3,
6716) = 236.62, p =<0.001), and productivity (F(3, 6716) = 135.41, p<0.001) between the four stress
appraisal states. Post-hoc Tukey analysis found no significant differences in the outcomes between
"eustress-distress coexistence" and "distress," but this analysis indicated all other pairwise comparisons
were significant for all three metrics: productivity, mood, and stress (i.e., arousal).

47
Figure 2. Boxplots of perceived productivity, mood, and stress across stress appraisal states
The results of perceived stress and productivity generally follow the Yerkes-Dodson law. Firstly, low, or
insignificant stress arousal is associated with boredom, lack of motivation, limited interest, and low
performance. This is supported by our findings showing that boredom appraisals had the lowest stress level
(M=5.58±13.17) and were associated with significantly lower perceived productivity (M=27.35±27.67)
compared to the three other stress appraisal states. The Yerkes-Dodson law states that moderate stress levels
can increase alertness, attention, and motivation resulting in better performance. In our results, eustress
appraisals had the highest perceived productivity (M=46.02±32.94) and a slight but statistically significant
increase in stress level (M=8.98±15.41) compared to boredom. This represents the “sweet spot” or optimal
level of stress arousal that sets the grounds for eustress and maximizes performance. Next, the YerkesDodson law states that as stress arousal builds up, distress takes over eustress, leading to impaired
performance due to increased anxiety. This is also supported by our results, where the eustress-distress
coexistence showed increased stress (M=19.91±23.00) and reduced perceived productivity
(M=41.87±28.37). Finally, as distress became dominant, the average change in stress level was the highest
(M=21.61±18.01), and productivity showed a decreasing trend (M=38.30±26.69).
The lack of significant difference in perceived stress and productivity between distress and the eustressdistress coexistence condition may indicate that higher stress arousal is mainly associated with distress,
irrespective of whether eustress coexists with distress. Alternatively, the differentiation of these appraisal
states may not have been possible due to the small number of time points appraised as distress in the dataset.
Moreover, the outcomes associated with distress appraisal may be time-dependent or have a cumulative
component not maximally elicited within the 40-minute experimental condition.
Changes in mood perceptions were much smaller than stress and productivity across the four appraisal
states. The average rating of mood increased during times appraised as boredom and eustress, with the best
mood (M=3.67±11.03) occurring along with the eustress state. In contrast, mood was consistently rated
lower than the baseline when distress was indicated, either in combination with eustress or occurring alone
(M=-6.70±15.26). These results support the division of stress appraisal into 2 constructs: positive and
negative. While eustress (positive construct of stress) is often associated with excitement, enthusiasm, and
fulfillment [24], distress (negative construct of stress) leads to intense negative feelings [43]. It is worth
noting that the negative feelings associated with distress seem to engulf the positive feelings of eustress, as

48
demonstrated by the average perceived mood (M=-6.57±15.17) under the eustress-distress coexistence
condition being almost equal to that of the distress condition.
To the best of our knowledge, this work is the first attempt to quantify the Yerkes-Dodson law using stress
appraisal relative to stress arousal, performance, and mood. These findings contribute to a deeper
understanding of the complex relationship between stress and human performance while highlighting the
importance of optimizing stress arousal to achieve maximum productivity and well-being.
6.2.2. Comparison between different ML models for stress appraisal prediction
In this section, we address the second research question of this study: What ML algorithms are best suited
for predicting stress appraisal? Using all 83 features, we tested the following algorithms: Naïve Bayes,
Adaboost, Logistic Regression, Linear Discriminant Analysis, Decision tree, Multilayer perceptron,
Support Vector Machine (with polynomial -degree between 2 and 10- and radial kernel), K-nearest neighbor
(for K-values between 2 and 15, and Euclidean or Manhattan distances), Random Forest, and XGBoost. In
selecting these machine learning models, our aim was to encompass a wide spectrum of algorithmic
approaches, from simple and interpretable models like Naïve Bayes and Logistic Regression to more
complex and powerful methods such as neural networks (Multilayer Perceptron), ensemble techniques
(AdaBoost, Random Forest, and XGBoost), and versatile kernel-based methods (Support Vector Machines
and K-nearest neighbor). This comprehensive selection allows us to thoroughly evaluate the performance
of various models on our dataset, ensuring that the best-suited algorithm for predicting stress appraisal is
identified.
To overcome the problem of unbalanced classes, we applied an oversampling method using the synthetic
minority oversampling technique (SMOTE) algorithm [260] to generate new synthetic samples in the
minority classes. The algorithm draws a random sample from the minority class, identifies the k-nearest
neighbors, and creates synthetic data points in the direction of the vector connecting the minority instance
and its neighbors. The SMOTE algorithm was applied to the training set but not to the testing set. The
results of modeling the 83 features in the training set among the ML algorithms are presented in Table 16.
Table 16. Comparison of ML model accuracy between different classifiers
Algorithm Accuracy F1-score
Naïve Bayes 39.27% 40.58%
AdaBoost 48.09% 49.91%
Logistic Regression 54.94% 51.22%
Linear Discriminant Analysis 55.90% 53.61%
Decision Tree 61.07% 61.18%
Multilayer Perceptron 64.67% 63.97%
Support Vector Machine – Linear 64.55% 60.39%
Support Vector Machine – Radial 66.25% 63.57%
Support Vector Machine – Polynomial degree 5 69.06% 68.83%
K-Nearest Neighbor, K = 5, Euclidean distance 70.55% 70.42%
K-Nearest Neighbor, K = 4, Manhattan distance 75.56% 73.59%
Random Forest 77.33% 77.30%
XGBoost 82.78% 82.28%
Naïve Bayes, AdaBoost, logistic regression and linear discriminant analysis did not perform well. AdaBoost
is sensitive to imbalanced datasets [261]; if one class dominates the other, it will not generalize well to new
data. This is the case in our dataset, where the data distribution (Table 15) shows that the distress condition
represents only 5% of the total datapoints. Naïve Bayes performs poorly with irrelevant features as its
predictions may be influenced by these features that have no relation with the outcome under study.

49
Furthermore, Naïve Bayes and logistic regression both assume that all features are independent [262],
which was not true in our dataset as physiological and behavioral stress indicators are often affected by one
another. For instance, sympathetic neural activity in response to stress autoregulates ST and results in EDA
peaks [263]. If the input features are highly correlated, models might not be able to accurately capture the
relationship between the input features and the outcome under study [261]; a correlation matrix among the
input features shows that some features hold up to 80% correlation among each other. In addition, both
logistic regression and linear discriminant analysis algorithms are linear models and assume that the
relationship between the input variables and the output variable is linear. If the relationship is non-linear,
models that can handle non-linear relationships between the input features and the output are better suited;
hence decision trees, multilayer perceptron, and support vector machine algorithms led to better
performance.
K-NN algorithm coupled using K = 4 and Manhattan distance showed a decent accuracy in predicting stress
appraisal conditions (75.56%). K-NN is an instance-based algorithm, which makes predictions based on
the similarity of new data points to the training data [264]. This means that the outcome under study is
separable and can be divided into distinct classes based on the input features. However, it is worth noting
that K-NN can be sensitive to the hyperparameters choice of distance metric, and the value of K as our
analysis showed significant fluctuations in performance when changing these hyperparameters.
Random forest and XGBoost performed the best in predicting stress appraisal states. Random forest and
XGBoost are ensemble learning methods known for their good performance on classification problems that
use decision trees as the base learners, which can handle large numbers of features and are robust to
overfitting. Random forest creates many decision trees and combines their predictions through majority
voting, providing further robustness to overfitting and allowing for the capture of complex patterns in the
data [265]. XGBoost advances this approach by using gradient boosting to optimize the decision trees and
improve their performance and allows for fine-tuning of the model parameters. Additionally, XGBoost uses
regularization to prevent overfitting and optimizes the tree structure to minimize the loss function. These
optimizations allow XGBoost to generalize better to new data and capture more complex patterns, typically
resulting in slightly better performance than random forest in classification problems [266], as demonstrated
in our dataset.
6.2.3. Comparison between different modalities for stress appraisal prediction
In this section, we address the third research question of this study: What data modalities are best suited for
predicting stress appraisal? Our analysis first looked at clusters of data obtained from each of the primary
data collection tools (e.g., E4, H10, Kinect, Mini Mouse Macro) and then explored which of the 83
individual features were most useful. Our findings here can inform which data collection methods and
individual features are critical and which might be dropped to increase the feasibility of data collection in
real-world settings.
First, to examine the clusters of features, we used the XGBoost algorithm for our analysis as it led to the
best results, as shown in Section 6.2.2. The SMOTE algorithm was applied to all models, and gender was
always included due to differences in how stress can be perceived. We trained a baseline XGBoost model
based on the gender feature without any physiological or behavioral features. This model’s accuracy was
41.56%. Then, we tested several subsets of the data, as shown in Table 17.
Table 17. Comparison different features and data collection tools in the stress appraisal prediction
Features under study* (total number of features) Data collection tools Accuracy F1-score
Gender (1) - 41.56% 26.13%
1 Monitoring Device
EDA, ST, BVP, ACC (31) E4 73.43% 65.67%

50
HR&HRV (11) H10 63.94% 51.05%
Facial (40) Kinect 69.46% 60.62%
Computer (4) Mini Mouse Macro 44.87% 24.33%
2 Monitoring Devices
EDA, ST, BVP, ACC, HR&HRV (41) E4 + H10 79.55% 76.60%
EDA, ST, BVP, ACC, Facial (70) E4 + Kinect 81.08% 78.74%
EDA, ST, BVP, ACC, Computer (34) E4 + Mini Mouse Macro 70.42% 67.53%
HR&HRV, Facial (50) H10 + Kinect 72.25% 65.28%
HR&HRV, Computer (14) H10 + Mini Mouse
Macro 62.17% 51.38%
Facial, Computer (43) Kinect + Mini Mouse
Macro 70.05% 60.63%
3 Monitoring Devices
EDA, ST, BVP, ACC, HR&HRV, Facial (80) E4 + H10 + Kinect 82.09% 81.56%
EDA, ST, BVP, ACC, HR&HRV, Computer (44) E4 + H10 + Mini Mouse
Macro 79.45% 76.44%
HR&HRV, Facial features, Computer (53) H10 + Kinect + Mini
Mouse Macro 75.22% 67.56%
4 Monitoring Devices
EDA, ST, BVP, ACC, HR&HRV, Facial, Computer (83) E4 + H10 + Kinect +
Mini Mouse Macro 82.78% 82.28%
EDA: Electrodermal Activity; ST: Skin Temperature; BVP: Blood Volume Pulse; HR: Heart Rate; HRV:
Heart Rate Variability; ACC: Acceleration
When using only 1 monitoring device, data from the E4 (accuracy=73.43%) and facial features
(accuracy=69.46%) had the best prediction performances, while HR and HRV features resulted in weaker
performance (accuracy=63.94%), and the computer features only managed to improve the prediction
accuracy of the baseline gender model by ≈3%. When adding a second monitoring device, combinations
that included the E4 data had the best performance, with an accuracy as high as 81.08% when E4 data was
coupled with the facial features. The results indicate that the EDA, ST, BVP, and wrist acceleration are
among the biggest contributors to the accurate prediction of the different states of stress appraisal. Minimal
improvement by 1% occurred when combining data from three devices (accuracy=82.09%), almost equal
to the accuracy for the full dataset comprised of all features (accuracy=82.78%).
Importantly, a pure physiological dataset (E4+H10) resulted in a 79.55% accuracy, only 3% lower than the
highest accuracy reached using all four devices (accuracy=82.78%). This finding is important as it indicates
that physiological data alone may be adequate to accurately predict stress appraisal. These results offer
flexibility to users interested in implementing automated stress appraisal prediction at the workplace. If the
goal is to maximize the prediction performance, collecting as many features as possible is necessary;
however, when computation, time, and financial resources are limited, relying only on physiological
features can provide useful results. Nowadays, a single wearable device can offer various physiological
features. For instance, the Fitbit wristwatch, Oura rings, and WHOOP wristbands can collect many
physiological signals such as heart rate, HRV, EDA, wrist acceleration, and oxygen saturation all at once
[267], which makes them a feasible alternative for simple and unobtrusive data collection protocols and a
reliable option for stress appraisal prediction in office settings.
To identify the individual features that had the greatest impact on predicting stress appraisal, we conducted
a feature importance analysis on the best-performing ML model. We used the feature importance attribute
of the model to measure how much each feature contributed to the overall prediction accuracy. An
importance score for each feature was calculated to assess the contribution of each feature to the overall

51
prediction accuracy of the ML model. Our results revealed that the 15 most important features accounted
for a significant proportion of the model's overall accuracy (accuracy top15=80.96%), leading to an almost
equal prediction performance to that of the whole dataset (accuracy whole=82.78%). The 15 most important
features of the stress appraisal model using combined physiological and behavioral data are presented in
Figure 3.
Figure 3. Feature importance for the stress appraisal prediction model
The feature importance results indicate that EDA, ST, and BVP were the most significant physiological
features for stress appraisal. Additionally, wrist acceleration (Y-axis) ranked as the third most important
feature. These findings support our previous conclusions and demonstrate why the E4 wristwatch played a
crucial role in improving the predictive performance of the model. The Kinect camera contributed the
second most important feature (AU-07) and five other top features (AU-45, -06, -04, -14) and gaze angle
(Y-direction). This highlights why a model based solely on facial features achieved a respectable prediction
accuracy of 69.46% (as shown in Table 17). Heart rate and HRV features were less significant and less
prevalent. The least important feature among the top 15 was maximum heart rate, which explains why a
model utilizing only these features achieved an accuracy of 63.94%.
It is worth noting that gender was ranked 9th in the list of important features for predicting stress appraisal
states. Research has shown that gender differences affect stress arousal due to the biological nature of stress
[268], and previous studies on stress detection have found that gender is an important factor that must be
considered to achieve better prediction performance [269], [257]. Our results support that gender is also
crucial for accurately identifying stress appraisal states using predictive models. Future studies should
consider other personal factors not examined in this study that might also impact stress appraisal (e.g., age,
ethnicity).
Finally, data from the Mini Mouse Macro did not contribute value to the overall prediction models, and
none of the top 15 features included human-computer interaction features. While previous studies have
shown that computer-related features (keyboard strokes and mouse clicks) play an important role in stress
detection, our results did not find them to be a useful in predicting stress appraisal. The limited influence
in our study may have been due to the controlled nature of the selected tasks that limited variability in the
human-computer interactions across appraisal states. Additionally, other computer-related measures not
collected in our study (e.g., keystroke pressure, pause rate, typing speed, mouse movement and wheel usage,

52
number of open windows) might be more useful in predicting stress appraisal. Future research directions
must consider additional features and examine a broader range of task conditions to provide better insights
into the relationship between human-computer interaction and stress appraisal.
6.2.4. Variation in physiological and behavioral signals across stress appraisal states
This section addresses the fourth research question of this study: How do physiological and behavioral
responses vary across different stress appraisal states? To the best of our knowledge, this is the first study
to address this research question, which makes it a unique contribution to the state of the art. We focused
this analysis on the most important physiological and behavioral features for predicting stress appraisal
states identified by the feature importance analysis. Table 18 summarizes the average and standard
deviation values of all physiological and behavioral features based on the stress appraisal states.
Table 18. Physiological and behavioral data changes across the stress appraisal states
Boredom Eustress Eustress-distress
coexistence Distress
Physiological Features
Electrodermal activity, mean µS* -0.90±3.83 -0.11±0.43 0.07±0.43 0.06±0.13
Blood volume pulse, standard deviation* 38.55±27.48 46.54±26.70 51.94±35.43 50.54±22.67
Electrodermal activity, 75th percentile µS* -0.96±3.94 -0.13±0.40 0.08±0.29 0.09±0.15
Skin Temperature, median °C* -0.48±0.77 -0.32±0.79 -0.43±0.74 -0.55±1.21
Electrodermal activity, maximum µS* -0.82±3.80 -0.03±0.46 0.10±0.38 0.11±0.15
Skin temperature, mean °C* -0.46±0.73 -0.42±0.76 -0.54±0.73 -0.64±1.05
Heart rate, maximum bpm* 10.74±8.21 12.96±8.58 12.32±8.29 13.27±11.25
Behavioral Features
AU07 (lid tightener), mean 0.32±0.48 0.35±0.45 0.51±0.42 0.52±0.50
Wrist Acceleration in Y-axis, mean g* 4.40±17.10 10.66±14.78 9.29±14.28 5.03±15.28
AU45, blink count 7.45±4.13 7.87±4.08 6.58±4.11 6.53±4.09
Gaze angle in Y-axis, mean degrees 0.28±0.09 0.32±0.10 0.24±0.11 0.26±0.09
AU06 (cheek raiser), mean 0.15±0.28 0.19±0.25 0.17±0.26 0.04±0.12
AU04 (brow lowerer), standard deviation 0.29±0.43 0.32±0.50 0.39±0.45 0.37±0.44
AU14 (dimpler), mean 0.49±0.38 0.85±0.58 0.59±0.47 0.48±0.38
*Feature calculated as a change from the pre-experimental state with positive values indicating an increase and
negative values indicating a decrease from the baseline state.
The findings indicate that EDA played a significant role in predicting stress appraisal, with three out of the
top 15 features related to EDA (Table 18). Analysis of Variance (ANOVA) revealed statistically significant
differences in the EDA parameters between the appraisal states: mean (F(3, 67216) = 71.09, p < 0.001),
75th percentile (F(3, 67216) = 72.98, p < 0.001), and maximum (F(3, 67216) = 74.88, p < 0.001). As in the
productivity, mood, and stress analyses, post-hoc Tukey HSD tests indicated significant differences between
all pairwise comparisons of EDA outcomes across the appraisal states, except for the comparisons between
eustress-distress coexistence and distress conditions. All three EDA features exhibited a similar pattern.
Specifically, when participants perceived their situation as boring compared to the baseline condition, there
was a notable decrease in EDA with the mean, 75th percentile, and maximum values of EDA of -0.90±3.83,
-0.96±3.94, and -0.82±3.80, respectively. When participants perceived their situation as eustress, there was
a slight reduction in EDA compared to the baseline with mean, 75th percentile, and maximum values of
EDA of -0.11±0.43, -0.13±0.40, and -0.03±0.46, respectively. Conversely, when participants experienced
eustress-distress or distress, their EDA levels were slightly higher than baseline by 0.06-0.11 µS, an upward
trend suggesting increased stress arousal [270].
Among the top 15 features used in predicting stress appraisal states, two ST-related parameters, specifically
median and mean, were identified. ANOVA tests revealed a statistically significant difference in ST
parameters between stress appraisal states for the median (F(3, 67216) = 14.78, p < 0.001) and mean (F(3,

53
67216) = 16.06, p < 0.001) values. Tukey HSD analysis demonstrated significant differences in pairwise
comparisons across all four stress appraisal states. Perceptions of boredom and eustress had lower median
(-0.48°C and -0.32°C, respectively) and mean (-0.46°C and -0.42°C, respectively) ST than baseline. In
contrast, when participants assessed their work conditions as co-existing eustress-distress or distress alone,
there was a larger decrease in both median (-0.43°C and -0.55°C, respectively) and mean (-0.54°C and -
0.64°C, respectively) ST. The relationship between stress and ST is intricate and not always straightforward,
which may explain why conditions characterized by low arousal (such as boredom and eustress) also
exhibited a decrease in ST in comparison to the baseline. Furthermore, ST regulation involves numerous
factors, including external environmental conditions, blood flow, and metabolic processes [271], which
may require further exploration or control in future studies.
BVP and heart rate were the final two physiological features examined. BVP is a physiological indicator of
stress arousal that measures the amount of blood pumped by the heart in one minute. When experiencing
stress, the sympathetic nervous system can cause vasoconstriction, which narrows the blood vessels and
decreases blood volume at the sensor placement site. This can result in lower BVP readings. Conversely,
BVP readings tend to be more stable and consistent during non-stressful situations [272]. Our results
support this description; the ANOVA analysis showed a significant difference in the standard deviation of
BVP (F(3, 67216) = 60.28, p < 0.001) between appraisal states. More specifically, participants who
appraised their work as boredom had the lowest standard deviation in BVP readings (38.55±27.48),
participants who perceived their working situation as eustress showed a higher level of variation in BVP
readings (46.54±26.70), and those who experienced distress had the highest level of variation (51.94±35.43-
50.54±22.67). Maximum heart rate was only different in boredom states, with a significantly lower
maximum heart rate (10.74±8.21) than the other appraisal states. The difference is represented by a small
approximate increase of 2 beats per minute (bpm) in the maximum heart rate when comparing boredom
and the other states. Maximum heart rate may be useful for studies interested in differentiating boredom,
but these findings explain why maximum heart rate was the least important for predicting between all four
states (Figure 3).
Among the behavioral features, our results show that wrist acceleration (F(3, 67216) = 51.50, p < 0.001)
differed across stress appraisal states. wrist acceleration is the speed of movement of the hand, which can
be influenced by factors such as boredom and stress. For example, boredom can result in slower hand
movements, as the individual may lack motivation or interest in the task, while an adequate amount of stress
can cause the hand to move more quickly, as the individual might present more engagement and be
motivated and excited to finish work tasks [273]. Our results follow this pattern such that appraising the
working situation as boring had the lowest wrist acceleration (4.40±17.10), and when participants appraised
the work conditions as eustress, they showed the highest level of wrist acceleration (10.66±14.78). When
participants appraised the work as distress, a lower wrist acceleration was noted (5.03±15.28), which aligns
with an understanding that higher stress levels can interfere with motor performance and motor skills due
to muscle tension and decreased dexterity, contributing to lower wrist acceleration. No statistically
significant differences were noted for wrist acceleration between appraisals of the coexistence of eustress
and distress (9.29±14.28) and the eustress (10.66±14.78). Just as previous studies have indicated the value
of acceleration data for predicting stress arousal [274], our data indicate this feature may be useful in
differentiating ideal stress appraisal states (i.e., those containing eustress) from less ideal states (i.e.,
boredom and distress).
Eustress, distress, and boredom can all have a significant impact on facial features and body language.
When a person experiences eustress, their facial expression can reflect excitement and engagement, with
increased animation in the eyes, eyebrows, and mouth. On the other hand, when a person experiences
distress, their facial expression can be tense and tight, with furrowed eyebrows, a tight mouth, and a
downward gaze. Boredom can result in a neutral or expressionless face, little movement in the eyes or
eyebrows, and a passive or downward gaze. These emotions can also impact body language, such as head

54
rotation or gaze direction, as people tend to lean away from negative stimuli and towards positive ones.
These changes in facial features can provide valuable insights into a person's emotional state and can be
useful in fields such as psychology, human-computer interaction, and neuroscience. In general, when a
person is experiencing stress arousal, (eustress or distress), the intensity of AU07 (lid tightener) tends to
increase [275], while the changes in the intensity of AU04 (brow lowerer) also increases, although to a
lesser extent [276]. This pattern of facial movements is often associated with a tense or anxious expression,
which may reflect a heightened state of arousal in response to stress. On the other hand, when a person is
experiencing boredom, the intensity of both AUs 07 and 04 tends to decrease, which is associated with a
relaxed or neutral expression. This pattern of facial movements may reflect a decreased state of arousal in
response to a lack of stimulation or challenge. This description is in line with the findings presented in
Table 18.
Other facial features can be analyzed from the perspective of engagement, interest, and enthusiasm about
work tasks, and therefore their relation to productivity and performance of individuals. High levels of
engagement and productivity are associated with positive emotions such as happiness, interest, and
motivation. These emotions can cause an increase in the activation of the zygomaticus major muscle (AU06:
cheek raiser), which pulls the corners of the mouth upwards and creates a subtle smile [45]. Additionally,
high engagement and productivity can lead to an increase in the activation of the orbicularis oculi muscle
(AU14: dimpler), which raises the upper lip. This action unit has been widely used in ML applications for
the detection of increased attention and activity engagement [277], [278] . Our findings align with these
concepts, showing that the intensity of the action units is low when participants experience boredom or
when distress arises as performance degrades. On the other hand, the AU14 intensity reaches high levels
during a eustress state associated with optimal performance. It is important to note that these relationships
are complex and can vary between individuals.
The ANOVA results also suggest significant differences between stress appraisal states in blink count (F(3,
67216) =5.01, p =0.02). The blink count of 7.45±4.13 while experiencing boredom was slightly lower than
during eustress states (7.87±4.08). Although not statistically significant, this observation indicates that
eustress may contribute to a heightened level of attention or cognitive engagement, leading to an increased
blink rate. In contrast, participants experiencing distress, representing a negative and aversive emotional
state, exhibited a lower blink count. Specifically, participants demonstrated blink rates of 6.58±4.11 and
6.53±4.09 under eustress-distress coexistence and pure distress conditions, respectively. Post-hoc tests
indicated significant differences between the pairwise comparisons of eustress-distress coexistence with
both boredom and eustress states. These findings suggest that distress may elicit a distinct physiological
response compared to boredom and eustress. The lower blink rate during distress could indicate heightened
vigilance or cognitive load, as individuals may be more focused on the distressing stimuli, resulting in a
reduced blink rate.
Lastly, the mean gaze angle in the y-direction is an important feature for predicting stress appraisal.
However, these features are often affected by other factors, such as bodily postures [279], which were not
examined in our analysis. Further analysis is needed to determine the specific impact of this feature beyond
the effects of posture to ensure a more accurate and comprehensive understanding of the relationships
between stress arousal and appraisal, and also performance, and the role played by mean gaze angle in these
relationships. Furthermore, while it is noteworthy that individuals can exhibit unique variations in their
facial expressions, the connection between stress arousal and certain facial movements may differ from
person to person. Additionally, context, personality, personal traits, and other factors can impact the
correlation between performance, engagement, and activation of facial muscles. To gain a comprehensive
understanding of the intricate relationship between stress arousal, eustress, distress, boredom, and
performance with facial expressions and gaze, further research is needed.
6.3.Conclusions

55
To the best of our knowledge, this research is the first to use a ML framework to forecast stress appraisal.
The stress appraisal was separated into four categories: boredom, eustress, combined eustress-distress, and
distress. The study simulated various work scenarios with two levels of stress arousal: low-stress and highstress work conditions. To build the ML prediction models, both physiological and behavioral signals were
utilized. The findings demonstrate that the relationship between perceived stress and performance aligns
with the Yerkes-Dodson principle, where moderate stress is associated with improved performance but too
much or too little stress is linked with degraded performance. After evaluating thirteen different classifiers,
our results indicate that the XGBoost algorithm had the best performance for predicting stress appraisal
states. By utilizing this model, a combination of physiological and behavioral features resulted in an
accuracy rate of 82.78% in predicting stress appraisal. The feature importance analysis showed that
physiological and facial features had the greatest impact on prediction performance, while human-computer
interaction features had little effect. Finally, we explored how the intensities of physiological and facial
features varied across the different stress appraisal conditions and analyzed these differences in relation to
previous research findings in the field.
6.3.1. Practical implications
Our research extends beyond the realm of academic inquiry, intertwining the fields of general psychology,
organizational psychology, and affective computing to offer practical solutions for enhancing the well-being
and health of employees within the workplace. Rooted in the principles of general psychology, our study
delves deep into the intricacies of stress appraisal and its implications for human physiology, behavior and
cognition. From an organizational psychology perspective, in the context of today's fast-paced and
demanding work environments, addressing stress-related issues is paramount to eliminating burnout,
mitigating intense stress, and preventing extended stress exposure. First and foremost, our research
empowers managers with a valuable tool to enhance the well-being of their teams. By utilizing our stress
appraisal model, managers can gain insights into their employees' unique stress appraisals, which
encompass not only distress but also eustress – the beneficial form of stress associated with motivation and
performance enhancement. Armed with this knowledge, they can make more informed decisions when it
comes to task allocation. This approach goes beyond mere task distribution; it considers individual stress
thresholds and preferences, creating opportunities to harness the power of eustress, prevent boredom
situations, and eliminate extended periods of excessive distress which can be significant contributors to
burnout.
Acknowledging the financial considerations that organizations often face, we propose a cost-effective
approach to monitoring employee well-being. We recommend the utilization of a single monitoring device
for data collection. Drawing from the domain of affective computing, this pragmatic approach strikes a
balance between effectiveness and cost-efficiency, leveraging technology to understand and respond to
human emotions and physiological responses. It is vital to emphasize that the implementation of such
monitoring systems should be accompanied by transparent communication and a steadfast commitment to
respecting privacy rights. When employees are informed about the purpose behind data collection and have
the agency to choose their level of participation, trust is fostered, and sustainable well-being initiatives are
better supported.
Employee engagement and job satisfaction are pivotal components of a healthy work environment. Our
stress appraisal model offers a valuable tool for identifying instances of employee boredom, distress, and
eustress. Recognizing the positive impact of eustress on motivation and performance, organizations can
develop tailored strategies not only to eliminate distress but also to harness the power of eustress, thereby
contributing to improved well-being. Effective stress management is essential in maintaining a healthy
workplace. Beyond assessing stress arousal, our model can evaluate the effectiveness of stress-management
training programs. By considering stress appraisal alongside stress arousal, organizations gain a more

56
comprehensive understanding of how well these programs are truly benefiting employees. This insight can
lead to more targeted and effective stress management initiatives, further supporting well-being.
To further enhance our framework, we propose coupling it with a notification system that can alert workers
to prolonged distress experiences. This proactive approach allows for timely intervention suggestions to
limit unhealthy stress exposure. By intervening early and promoting the appropriate balance between
distress and eustress, organizations can prevent the escalation of stress-related issues, promoting the wellbeing and health of their employees.
6.3.2. Limitations and future research directions
The study, which represents the first effort to use ML to identify stress appraisal states, has some limitations
that should be addressed in future research. One limitation is that the experiment was not a true reflection
of office work, as participants were assigned predesigned tasks and subjected to prescribed work conditions.
To that end, it should be noted that the importance of different signals, such as wrist acceleration, may vary
depending on the nature of the task or job description and should. Hence, future research is needed to
examine stress appraisal within real-life work environments. The ML models in this study focused on
gender as a moderator in stress appraisal, but they did not encompass all the personal factors affecting
stress. While our study advanced stress appraisal through machine learning, it is crucial to recognize that
factors beyond gender, like age, socioeconomic background, early life experiences, and ethnicity, also
impact stress responses. To fully understand stress experiences, future research should explore these factors
to enhance ML models and establish personalized prediction frameworks for distinct worker groups based
on their unique profiles. Finally, the results showed that both head movement and gaze are important
predictors of stress appraisal, suggesting the importance of considering body posture in future research.

57
Chapter 7: Predicting Office Workers Productivity: A Machine Learning
Approach Integrating Physiological, Behavioral, and Psychological Indicators
This study pioneers the application of a machine learning framework to predict perceived productivity of
office workers using physiological, behavioral, and psychological features. Two approaches were
compared: the baseline model, predicting productivity based on physiological and behavioral
characteristics, and the extended model, incorporating predictions of the psychological states such as stress,
eustress, distress, and mood. The following sections of this chapter are organized as follows: Section 7.1
explains in detail the methodology adopted. Section 7.2 provides a summary of the results and a detailed
interpretation of the study findings. Section 7.3 offers an overview of the study limitations. Finally, Section
7.4 summarizes the conclusions.
7.1.Methodology
We performed an experimental study to predict individuals' perceptions of stress, eustress, distress, and
mood, which are utilized to gauge their perceived productivity levels. To capture a broad spectrum of signals
from each participant, our 70-minute experiment consisted of two phases: the first phase involved lowstress engagement with a computer workstation, followed by a second phase that incorporated several
stressful elements. The aim of this experimental method was to create diverse work conditions that could
impact the emotional states of participants and subsequently their productivity.
7.1.1. Participants
The research included 48 volunteers who took part willingly, consisting of 28 women and 20 men. The
participants were a mix of graduate and undergraduate students, averaging 22.6 years old with a standard
deviation of 2.1 years. Individuals with vision problems that could hinder computer-related tasks,
psychological disorders that make them more vulnerable to stress, pregnant women, and those using
medication that might impact physiological signals were excluded from the study. Informed consent was
obtained from all subjects involved in the study.
7.1.2. Data Collection
During the study, the participants were equipped with two types of sensors: an E4 Empatica wristband [242]
and an H10 polar chest strap [243]. These sensors were employed to gather various physiological
information, including HR, EDA, ST, Blood Volume Pulse (BVP), and wrist accelerations in the x, y, and
z directions. Additionally, a Microsoft Azure Kinect DK camera [280] was positioned to capture
participants' facial expressions throughout the experiment. To record their computer interactions, such as
keystrokes and mouse clicks, a logging application called Mini Mouse Macro [245] was utilized. Figure 4
illustrates the configuration of the workstation used in the experiment.

58
Figure 4. Participant taking the experiment showing the used sensors and the data collection platform
The experimental procedure comprised two distinct phases: low-stress work and high-stress work. At the
commencement of each phase, participants were instructed to remain motionless for a duration of 5 minutes
to record their baseline physiological data while at rest. Following these initial resting periods, participants
were required to rate their baseline subjective stress and mood levels on a scale ranging from 0 to 100. A
rating of 0 denoted the absence of stress or a negative mood, whereas a rating of 100 indicated the presence
of extreme stress or a positive mood. During both phases, at intervals of 5 minutes, participants were
presented with a pop-up questionnaire on the computer screen. This questionnaire prompted them to rate
their perceived stress and mood levels using the 0-100 scale. Furthermore, participants were requested to
evaluate the nature of their work experience as either eustress or distress using the VEDAS. Eustress was
evaluated based on the perception of opportunity or challenge, employing a 6-point scale ranging from
“very definitely is not” to “very definitely is.” Similarly, distress was assessed as a source of pressure using
the same 6-point scale. Additionally, participants were asked to rate their perceived productivity on a scale
of 0 to 100, with 0 representing an extreme lack of productivity and 100 indicating an exceptional level of
productivity. The sequence of the two phases was kept consistent for all participants, beginning with the
low-stress tasks followed by the high-stress tasks. Randomization was not employed in this study as its
primary focus was on data collection rather than examining the effects of varying task sequences. Prior to
the commencement of the experiment, participants were provided with guidelines and definitions to help
them understand and interpret the scales accurately. For instance, for stress, a score of 0 was explained as
feeling completely at ease and relaxed, whereas a score of 100 represented feeling overwhelmed and unable
to cope with the pressure. Analogously, mood ratings were clarified with 0 representing feelings of sadness
or frustration and 100 reflecting feelings of happiness or excitement.
The objective of the low-stress phase was to provide participants with a perception of autonomy and
minimize external demands or pressures associated with the task. In contrast, the high-stress phase was
deliberately designed to introduce a workload that was challenging within a restricted timeframe. During
the low-stress phase, participants were allotted 40 minutes to construct a slide deck centered around a topic
of their preference, such as a beloved book, television series, or movie. Subsequently, participants were

59
granted a break before resuming the experiment. This break was intentionally brief, limited to no more than
two minutes, in order to preserve the participants’ focus and work-oriented mindset.
Upon returning and recording their resting physiological data, participants were informed that they would
no longer be working on their initial topic. Instead, they were instructed to engage with an unfamiliar topic
for a duration of 30 minutes. To identify a topic that would provoke a high-stress response, we conducted
a pilot study, testing various themes. Ultimately, we settled on a discussion regarding the scientific and
philosophical contributions of two ancient Greek philosophers and the enduring impact of their ideas on
contemporary human life.
In the experiment's high-stress condition, additional stressors were applied to elevate pressure. Participants
had to activate video cameras and share screens via Zoom, with a confederate posing as a professor
specializing in optimal work environments for productivity. This authoritative figure was introduced to
increase expectations and induce performance anxiety and stress due to fear of negative evaluations from
someone significant in the field. The confederate stated they would monitor and potentially decrease
participants' scores based on performance, simulating a high-pressure work environment with critical
evaluations. A computer application displayed fluctuating ratings for the professor throughout the 40-
minute task, with these manipulations standardized across participants. Participants were informed their
scores would be compared to others', introducing a competitive element and inducing stress through fear of
inadequacy or failure. Compensation varied based on performance, with the highest-scoring participants
receiving $50 and the lowest $5. This variable compensation introduced financial stressors, playing on loss
aversion and stress associated with potential monetary loss. However, during debriefing, participants
learned the confederate was not an actual expert, and their scores did not affect compensation. All
participants received maximum compensation regardless of performance.
7.1.3. Data Processing
The participants' perceived stress and mood ratings were calculated by subtracting their ratings during the
resting period from the corresponding ratings during the experimental phases. This resulted in perceived
stress and mood ratings ranging from 0 to 100, representing the deviation from the baseline. The appraisals
of eustress and distress were transformed into a binary outcome. Responses falling within the categories of
"very definitely is not a source of," "definitely is not a source of," and "generally is not a source of" were
combined to create a category indicating that the stress was not appraised as either eustress or distress.
Similarly, the categories of "very definitely is a source of," "definitely is a source of," and "generally is a
source of" were merged to form a category indicating that the stress was appraised as either eustress or
distress. Given that participants were not engaged in any work tasks during the baseline phases, the
perceived productivity baseline was assumed to be 0, reflecting the absence of any productivity during
those periods.
The data collected in the study was segmented into 30-second time windows to extract physiological and
behavioral features. This choice was informed by Bernardes et al.'s research [248], which demonstrated that
30-second windows provided the smallest reliable timeframe for extracting HRV features that accurately
assess psychological states. To analyze HRV and obtain time and frequency-domain indices of the HR
signal at each 30-second interval, the Kubios software [249] was employed. A moderate artifact correction
technique was applied to pinpoint R-R intervals that deviated more than 0.25 seconds from the mean. This
approach retained the variability of the data while managing any existing artifacts. Furthermore, Kubios
incorporated a piecewise cubic spline interpolation process to fill in flawed or missing data, ensuring a
more refined and precise HRV reading.

60
Data gathered from the Empatica E4 underwent processing prior to the extraction of features to minimize
noise, akin to the methodology adopted in an earlier research [250]. BVP and ST signals were refined using
winsorization [251], a statistical method that removes outliers beyond the 2nd and 98th percentiles. For
processing the EDA signal, we employed the MATLAB Ledalab toolbox [252]. This included the use of a
Butterworth low-pass filter, Hanning smoothing encompassing a span of 4 consecutive data points, and
manual artifact rectification to eliminate any noise potentially due to motion or other external disruptions.
Post this cleaning phase, we computed the average, variance, median, minimum, maximum, 25th and 75th
percentiles, and the fitted slope (i.e., linear regression slope) of BVP, EDA, and ST to provide a
comprehensive evaluation of the multiple facets inherent in psychological assessment [253].
The OpenFace tool [254] was utilized to retrieve the mean and standard deviation values of facial action
unit intensities, gaze angles, and head movement and orientation at 30-second intervals, derived from the
RGB video recorded by the Kinect camera. Furthermore, keyboard strokes and mouse clicks were recorded
for each 30-second time window. Finally, the mean and standard deviation of the wrist acceleration in the
x, y, and z planes were recorded. A summary of the features dataset, including the various measures and
statistics derived from the different physiological and behavioral sources, can be found in Table 19.
Table 19. Feature Dataset
Type (Number of features) Signal Features Included
Physiological (34)
Electrodermal activity (EDA)
Blood volume pulse (BVP)
Skin temperature (ST)
Mean, Standard deviation,
Median, Minimum, Maximum,
25th & 75th percentile, slope
fitted through the data.
Heart rate (HR)
Heart rate variability (HRV)
Mean HR, Standard deviation
HR, Minimum HR, Maximum
HR, rmsdd, LF peak, HF peak,
LF power, HF power, LF/HF
Behavioral (48)
Facial action units (AUs)
Head rotation
Eye gaze direction
Mean, Standard deviation
Blink Count
Wrist acceleration Mean, Standard deviation
Mouse right clicks
Mouse left clicks
Keyboard keystrokes
Count
Gender (1) Female, Male Binary
The final dataset utilized in this study consisted of 6,720 data instances, collected from 48 participants.
Each participant contributed 70 minutes of data, with 2 instances per minute using 30-second time windows.
The final dataset comprised 83 features, encompassing 34 physiological features, 48 behavioral features
(including 3 related to human-computer interactions, 39 facial-related features, and 6 hand wrist
acceleration features), and a feature indicating the participant's gender. To facilitate proper normalization,
robust scaling was applied on a per-participant basis to ensure consistency across the dataset.

61
7.1.4. Analysis Plan
Our analysis consists of four distinct steps. In the first step, we focus on predicting productivity using the
physiological and behavioral dataset solely. In the second step, we extended our productivity prediction
model by incorporating the predictions of psychological states (stress, mood, eustress, and distress) into the
dataset. To achieve this, we employed various ML algorithms and selected the best performing one for each
outcome under study. Continuous metrics such as stress, mood, and productivity are predicted using
regression models, including linear regression, ridge regression, lasso regression, random forest, gradient
boosting regressor, and the Extreme Gradient Boosting (XGBoost) regressor. On the other hand, binary
outcomes like eustress and distress require classification models, and we evaluate algorithms such as
logistic regression, random forest, gradient boosting classifier, XGBoost classifier, decision tree, and
Support Vector Classifier (SVC). To evaluate the performance of these ML models, we employed an 80%-
20% split, where 80% of the dataset was utilized for training and the remaining 20% for testing purposes.
To assess the regression models, we considered the Mean Absolute Error (MAE) and R-squared (R2
) as
evaluation metrics. For the classification models, we utilized accuracy and F1-score to evaluate their
performance. In the third step, a feature importance analysis was conducted on the extended productivity
model. This process aimed to elucidate significant predictors of productivity by calculating importance
scores for each feature, utilizing the model’s feature importance attribute. This analysis identified and
examined the top 15 features influencing productivity predictions, aiding in understanding the model's
predictive mechanisms. In the fourth step, we evaluated the effectiveness of physiological and behavioral
features individually in predicting productivity. Figure 5 presents a summary of the analysis.
Figure 5. Overview of analysis
7.2.Results and Discussion
7.2.1. Predicting mood, stress, eustress, and distress
In this study, we conducted a regression analysis to evaluate the performance of various algorithms in
predicting stress and mood levels. The regression analysis aimed to assess the effectiveness of each model
in capturing the underlying patterns and relationships. Additionally, a classification analysis was performed

62
to determine the accuracy of different models in predicting eustress and distress. The classification analysis
focused on evaluating the models' ability to classify individuals into the appropriate stress categories. Table
20 provides a summary of these results.
Table 20. Summary of mood, stress, eustress, and distress prediction models performance
Regression Analysis
Algorithms Mood Stress
R
2 MAE R
2 MAE
Linear Regression 0.06 10.36 0.09 14.74
Ridge Regression 0.04 10.40 0.03 15.40
Lasso Regression 0.01 11.13 0.01 15.68
Random Forest 0.38 8.23 0.44 9.81
Gradient Boosting 0.31 8.98 0.31 12.74
XGBoost 0.44 7.91 0.43 10.01
Classification Analysis
Algorithms Eustress Distress
Accuracy F1-score Accuracy F1-score
Logistic Regression 0.79 0.65 0.53 0.68
Random Forest 0.84 0.88 0.81 0.82
Gradient Boosting 0.79 0.86 0.75 0.77
Decision Tree 0.74 0.81 0.67 0.69
Support Vector 0.65 0.78 0.68 0.51
XGBoost 0.88 0.91 0.85 0.85
Linear regression models are ineffective in predicting mood and stress levels, with weak correlations (R2
:
0.06-0.09) and high average deviations (MAE: 10.36-15.68). In contrast, tree-based ensemble models
(Random Forest, Gradient Boosting, XGBoost) outperform linear regression, showing stronger correlations
(R2
: 0.31-0.44) and lower average deviations (MAE: 7.91-12.74). These results indicate that tree-based
models better capture the complexities and nuances in mood and stress data.
Among the tree-based models, XGBoost and Random Forest excel in predicting mood and stress,
respectively. Their superior performance can be attributed to the employment of ensemble techniques that
leverage multiple decision trees to capture intricate relationships and interactions within the data. These
models are adept at handling nonlinearities, outliers, and high-dimensional feature spaces, allowing them
to effectively capture the nuanced aspects present in mood and stress data.
When considering eustress and distress predictions, Logistic Regression shows moderate accuracy (0.79)
and F1-scores (0.65) for eustress, indicating reasonably accurate predictions. However, its performance is
less satisfactory for distress, with lower accuracy (0.53) and F1-score (0.68). This limitation can be
attributed to the linear nature of Logistic Regression, which hinders its ability to capture the complexity
and nonlinearity associated with distress instances.
The Decision Tree model performs satisfactorily for Eustress (accuracy: 0.74, F1-score: 0.81), indicating
accurate predictions and a good balance between precision and recall. However, its performance for Distress
(accuracy: 0.67, F1-score: 0.69) suggests a relatively lower ability to accurately classify Distress instances,
possibly due to overfitting and capturing noise or idiosyncratic patterns that don't generalize well. Support
Vector Classifier exhibits inferior performance for both Eustress (accuracy: 0.65, F1-score: 0.78) and

63
Distress (accuracy: 0.68, F1-score: 0.51), indicating less accurate predictions and imbalanced precision and
recall. This can be attributed to the sensitivity of Support Vector Classifier to feature scaling and
hyperparameter selection, making it less effective when there is overlap between eustress and distress
instances.
Gradient Boosting demonstrates strong performance for both Eustress (accuracy: 0.79, F1-score: 0.86) and
Distress (accuracy: 0.75, F1-score: 0.77), capturing complex relationships effectively. Random Forest
performs even better with higher accuracy for both Eustress (accuracy: 0.84, F1-score: 0.88) and Distress
(accuracy: 0.81, F1-score: 0.82), integrating multiple decision trees to capture a wider range of patterns.
XGBoost emerges as the top performer, achieving the highest accuracy for both Eustress (accuracy: 0.88,
F1-score: 0.91) and Distress (accuracy: 0.85, F1-score: 0.85), utilizing advanced ensemble techniques,
regularization, and optimization strategies to handle complex datasets effectively.
Comparing our models with those from the literature is challenging due to variances in data signals,
features, ML algorithms, and experimental conditions. Nevertheless, our top-performing models appear
competitive. For instance, a study by Yu et al. [281] utilized wrist acceleration, EDA, and ST from mobile
phone data to predict mood and stress on a 0-100 scale, achieving a MAE of 13.7 and 12.8 for stress and
mood prediction, respectively. In comparison, our models achieved MAE values of 7.91 and 9.81. In a
separate study, Li et al. [282] combined computer usage with HR and HRV data, achieving a prediction
accuracy of 71% for eustress, whereas our model reported an accuracy of 88%. Additionally, a study
focused on detecting distress events using a ML model based on EDA and BVP attained a F1-score of 0.71
[283], compared to our model's F1-score of 0.85.
7.2.2. Baseline vs. extended productivity models
We begin by establishing the baseline productivity model, which predicts productivity solely using the
physiological and behavioral features collected during the experiment. Subsequently, the extended
productivity model incorporates additional predictions from the best performing models of mood, stress,
eustress, and distress, in addition to those features. Specifically, we utilized the XGBoost models outlined
in Table 20 for predicting mood, eustress, and distress. For stress prediction, we employed the Random
Forest model, also presented in Table 20. We tested several ML algorithms, and the outcomes are presented
in Table 21. It is noteworthy to highlight that in the extended model, stress, mood, eustress, and distress
were already predicted using the physiological and behavioral features. Consequently, we conducted a
thorough assessment to address the potential issue of multicollinearity, employing the well-established
statistical measure known as Variance Inflation Factor (VIF). Multicollinearity arises when two or more
independent variables within a regression model exhibit high correlation, leading to undesirable
consequences such as unstable and unreliable coefficient estimates. VIF precisely quantifies the extent to
which the variance of an estimated regression coefficient is inflated due to multicollinearity. Through our
analysis, we ascertained that the correlation levels were moderate, with VIF values ranging from 1 to 8.
This finding indicates that the multicollinearity in the extended model is not severe and does not necessitate
any alterations.
Table 21. Performance comparison between the baseline and extended productivity models
Productivity Regression Analysis
Algorithms Baseline Model Extended Model
R
2 MAE R
2 MAE
Linear Regression 0.25 21.59 0.27 16.42

64
Ridge Regression 0.12 23.43 0.13 18.60
Lasso Regression 0.10 24.16 0.15 22.60
Random Forest 0.44 17.19 0.57 10.91
Gradient Boosting 0.40 18.29 0.46 13.67
XGBoost 0.48 16.62 0.60 10.52
The productivity regression analysis highlighted the consistent superiority of the extended model over the
baseline model in predicting productivity. Remarkably, XGBoost was the top-performing algorithm for
both models. In the comparison, the extended model posted an impressive R2 of 0.60 and a lower MAE of
10.52, against the baseline's R2 of 0.48 and MAE of 16.62. Such improved performance was consistent
across various algorithms, substantiating the extended model's enhanced effectiveness. While the extended
model demonstrates a noteworthy improvement over the baseline, it is important to acknowledge that the
R
2 value of 0.60 represents a moderate correlation. This indicates that, despite its superior performance, the
extended model still has room for refinement to capture the complexities of productivity more accurately.
What set the extended model apart was its integration of additional features. Instead of relying solely on
physiological and behavioral variables like the baseline model, it incorporated predictions from leading
models on mood, stress, eustress, and distress. This broadened feature set ensured a more in-depth
understanding, capturing the intricate human conditions that influence productivity. The stark difference in
performance metrics, particularly the R2
improvement and MAE reduction for XGBoost, attests to the
potency of a holistic approach. It drives home the point that for nuanced, human-centric predictions like
productivity, it is imperative to embrace a wider spectrum of influencing factors.
It is noteworthy that the direct use of self-reported scores (as opposed to predictions) for stress, mood,
eustress, and distress in the extended productivity model yielded a performance enhancement. Specifically,
improvements ranged from 0.7% to 2.4% when compared to the results obtained from the extended
predictive model based on stress, mood, eustress, distress predictions as presented in Table 21. Utilizing a
model based on psychological state predictions may necessitate initial ground truth data for training the ML
model, after which user input becomes unnecessary. On the other hand, relying solely on self-reported
metrics would require continuous user input. This could compromise the objective of establishing an
automated framework, as it may result in frequent interruptions to work activities. Consequently, it may be
judicious to forgo the slight increase in prediction accuracy in favor of the operational advantages offered
by a fully-automated ML system built on psychological state predictions.
As far as our knowledge extends, there has been no prior investigation into the utilization of physiological
or behavioral features to evaluate productivity among office workers, let alone the incorporation of
psychological states. Thus, these findings provide an essential basis for future endeavors focused on the
development of machine learning-driven solutions to predict and monitor productivity in an office setting.
7.2.3. Analyzing features importance
To gain a deeper understanding of the prediction mechanism employed by the extended model, we
conducted a feature importance analysis. This analysis aimed to reveal the physiological and behavioral
features that exerted the greatest influence in predicting productivity, as well as shed light on the role of
emotional states in this prediction process. We used the feature importance attribute of the model to measure
how much each feature contributed to the overall prediction accuracy. An importance score for each feature

65
was calculated to assess the contribution of each feature to the overall prediction accuracy of the ML model.
The 15 most important features of the extended productivity model are presented in Figure 6.
Figure 6. Feature importance for the extended productivity prediction model
The analysis of feature importance in predicting productivity revealed several key predictors, including
physiological, behavioral, and emotional features. Of particular interest are the emotional states, which
emerged as significant contributors to productivity. Understanding why these emotional states play a crucial
role in predicting productivity requires a closer examination of their underlying mechanisms. Emotions, as
complex and dynamic psychological states, have long been recognized for their influence on cognitive
processes and behavior. In the context of productivity, emotions can shape an individual's motivation,
attention, decision-making, and overall cognitive functioning. The appearance of emotional states as
important predictors in the extended model highlights their importance in capturing the multifaceted nature
of productivity.
One prominent emotional characteristic to consider is the predicted eustress, a form of stress that conveys
positive implications. Eustress denotes a moderate degree of stress that individuals perceive as
advantageous or motivating. It emerges in circumstances where individuals encounter a sense of challenge,
excitement, or anticipation. Eustress can heighten cognitive performance, foster adaptive coping
mechanisms, and facilitate goal-oriented conduct [177]. By integrating predictions of eustress, the expanded
model acknowledges the potential advantages of stress in optimizing productivity. It posits that an optimal
level of stimulation and challenge can cultivate engagement, concentration, and the mobilization of
cognitive resources. Findings from the t-test provide support for this assertion (t(6718) = 3.95, p = 0.04); the
projected productivity level was higher (42.75±27.05) when the model predicted eustress compared to
scenarios without eustress (29.17±26.32).
Among the emotional features under investigation, the predicted mood emerged as a notably influential
factor in the prediction of productivity. Mood encompasses an individual's overall emotional state,
encompassing various degrees of positivity, negativity, or neutrality. Positive mood states, characterized by
emotions such as enthusiasm, joy, and contentment, have been consistently linked to heightened cognitive
flexibility, enhanced creative thinking, and improved problem-solving capabilities [284]. Such positive
affective experiences contribute to increased motivation, active engagement in tasks, and effective

66
information processing. On the other hand, negative mood states, including emotions such as sadness,
anxiety, or frustration, can detrimentally affect cognitive functioning, leading to decreased levels of
productivity. A correlation analysis revealed a statistically significant yet weak positive Pearson correlation
between mood and productivity (r = 0.10, N = 6720, p < 0.001). The extended model's ability to capture and
integrate these mood predictions enables a more comprehensive understanding of the impact of affective
experiences on productivity.
In our study, we initially hypothesized that distress would have a negative impact on productivity. However,
contrary to our expectations, our findings revealed a different pattern. The results of our t-test analysis
contradicted the expected trend, as individuals experiencing distress (41.56±25.72) reported higher levels
of productivity compared to those without distress (34.73±29.01). To gain a deeper understanding of this
unexpected outcome, we further investigated how the simultaneous presence of eustress and distress
influenced individuals' perception of productivity. Specifically, we examined four combinations of
predictive outcomes: "no-eustress" and "no-distress," "eustress" but "no-distress," "eustress" and "distress,"
and "no-eustress" but "distress."
Significant statistical differences were observed in predicted productivity levels (F(3, 6716) = 142.30,
p =<0.001). These findings suggest that the interplay between eustress and distress, rather than the presence
of distress alone, may have a nuanced impact on individuals' productivity levels. Specifically, when both
"no-eustress" and "no-distress" were simultaneously predicted by the model, the average predicted
productivity level was found to be (M=27.52±26.37). In cases where the model predicted "eustress" but
"no-distress," the average predicted productivity level was (M=45.95±29.73). When both "eustress" and
"distress" were predicted by the model, the average predicted productivity level was (M=41.92±25.84).
Conversely, when the model predicted "no-eustress" but "distress," the average predicted productivity level
was (M=38.21±24.78). These results indicate that the presence of pure distress can lead to decreased
predicted productivity compared to a state of pure eustress. However, the lowest predicted productivity
level was observed when the model predicted a worker's state of no eustress and no distress, which may
signify a state of boredom or disengagement from work [285].
The study's results should be interpreted cautiously as the sample exclusively consists of young students,
whose stress responses and perceptions of productivity may not generalize to wider or diverse populations.
Their unique academic stressors, coping mechanisms, and perhaps elevated resilience to distress might lead
to different productivity outcomes compared to non-student groups.
It is important to acknowledge that emotional states exhibit considerable predictive power in the expanded
model. However, it is crucial to consider them in conjunction with physiological and behavioral
characteristics. This is because emotional states, while influential, do not exist in isolation; they are
intertwined with our physiological responses and the actions we take. By adopting this holistic approach,
we gain a more comprehensive understanding of how these factors collectively impact productivity. Table
22 presents the correlation analysis between predicted productivity and the most significant physiological
and behavioral features in the predictive framework. In our analysis, we conducted Shapiro-Wilk tests, and
all obtained p-values exceeded the significance threshold of 0.05, confirming that the assumption of
normality for our data is met and validating the use of correlation analysis.
Table 22. Correlation between predicted productivity and physiological, behavioral features

67
The observation of a positive relationship between ST and productivity suggests a potential interconnection
between physiological and cognitive processes. One plausible explanation for this phenomenon lies in the
amplified blood flow and metabolic activity that occur during engaged and productive tasks [286]. This
heightened physiological response facilitates effective heat transfer from the body's core to the peripheral
regions, consequently leading to an elevation in ST. However, it is noteworthy that while four EDA-related
features emerged as primary predictors of productivity, only the median of the EDA signal exhibited a
statistically significant positive correlation with the predicted productivity. Although the correlation
coefficients between the EDA-related features and predicted productivity are relatively small, ranging from
0.02 to 0.04, they do show a positive association. This finding could be attributed to the activation of the
sympathetic nervous system, which occurs during focused attention and heightened arousal in productive
tasks. This activation leads to increased sweat secretion, resulting in higher EDA values. [286]. These
findings lay the foundation for future investigations aimed at elucidating the underlying relationship
between EDA and productivity.
Higher wrist acceleration is found to be negatively correlated with productivity. In tasks involving a
keyboard and mouse, optimal productivity is often associated with stable, precise, and limited hand
movements. Conversely, higher hand acceleration, suggesting excessive or erratic hand movement, tends
to be negatively correlated with productivity as it might indicate non-focused or inefficient activity [287].
Conversely, a positive correlation exists between the standard deviation of AU06 Cheek Raiser and AU10
Upper Lip Raiser and productivity, suggesting that greater variation in the movements of these action units
is associated with higher productivity. This may be attributed to increased facial expressiveness, reflecting
active engagement and emotional responsiveness as well as lower stress levels [288], thus contributing to
enhanced productivity. Lastly, a negative correlation is found between the mean of AU04 Brow Lowerer
and productivity, suggesting that higher levels of brow lowering movements are associated with lower
productivity. Brow lowering often accompanies negative emotions or concentration, potentially indicating
increased cognitive load or negative affect, which may impede productivity [275].
Mean in head rotation (Z-axis) showed a positive correlation coefficient of 0.06. This finding suggests that
greater head rotation is associated with higher productivity levels. The increased head rotation may reflect
heightened attentiveness and active involvement in tasks, indicating an individual's active scanning of the
environment or engagement in complex cognitive processes. These cognitive processes likely contribute to
enhanced productivity by facilitating information processing and task engagement. Additionally, the mean
Pearson Correlation P-value
Physiological features
Max: Skin Temperature 0.16 <0.001
75th Percentile: Skin Temperature 0.16 <0.001
Mean: Electrodermal Activity 0.02 0.07
25th Percentile: Electrodermal Activity 0.02 0.07
75th Percentile: Electrodermal Activity 0.03 0.06
Median: Electrodermal Activity 0.04 0.04
Behavioral features
Mean: Wrist Acceleration (Y-axis) -0.11 <0.001
Standard Deviation: AU06 Cheek Raiser 0.08 <0.001
Mean: Head Rotation (Z-axis) 0.06 <0.001
Mean: AU04 Brow Lowerer -0.05 <0.001
Standard Deviation: AU10 Upper Lip Raiser 0.06 <0.001
Mean: Gaze Angle (X-axis) 0.13 <0.001

68
in gaze angle (X-axis) feature demonstrated a notable positive correlation coefficient of 0.13 with predicted
productivity. This correlation suggests that a more direct and focused gaze is associated with higher levels
of productivity. A concentrated gaze directed toward a task or relevant stimuli signifies sustained attention
and cognitive engagement. This focused visual attention is indicative of an individual's ability to maintain
cognitive resources on the task at hand, resulting in improved productivity.
It is crucial to acknowledge that the interpretations put forth are grounded in observed correlations,
signifying a relationship between the variables under investigation. However, to advance our understanding
and draw more definitive conclusions, further research is warranted to establish causality and unveil the
specific mechanisms that underlie the intricate relationships between these features and their impact on
productivity. Moreover, it is imperative to recognize the interconnection between these findings and the
experimental results conducted in our study. To ensure the generalizability and applicability of these results
in real-world settings, a comprehensive longitudinal data collection approach becomes indispensable. By
systematically gathering physiological and behavioral data over an extended duration and employing the
ecological momentary assessment method to continuously inquire about participants' psychological states
and productivity, we can enhance the generalizability and robustness of our findings. This approach allows
us to capture the dynamic nature of these variables in a real-world context and provides a more holistic
view of their influence on productivity. Additionally, such an approach enables us to gain insights into the
temporal aspects and potential causal pathways, shedding light on the underlying mechanisms that govern
these associations.
7.2.4. Comparing between different modalities
In this section, our investigation aimed to explore the impact of various modalities on the prediction of
productivity. Specifically, we conducted a comparative analysis between wearable devices, namely the
Empatica E4 and H10 Polar, and workstation addons, specifically the Kinect camera and Mini Mouse
Macro. Our focus was directed towards evaluating the performance of the extended productivity model, as
we had previously demonstrated its superior predictive capabilities.
Table 23. Comparative analysis between wearable devices and workstation addons
E4 Empatica & H10 Polar Kinect & Mini Mouse Macro
Performance
Metrics
Best Performing
Algorithm
Performance
Metrics
Best Performing
Algorithm
Mood R
2=0.43
MAE=8.02 XGBoost R
2=0.23
MAE=9.12 Random Forest
Stress R
2=0.40
MAE=11.17 Random Forest R
2=0.27
MAE=13.19 XGBoost
Eustress Accuracy=0.86
F1-score=0.90 XGBoost Accuracy=0.81
F1-score=0.86 XGBoost
Distress Accuracy=0.82
F1-score=0.83 XGBoost Accuracy=0.77
F1-score=0.76 XGBoost
Productivity:
Extended Model
R
2=0.56
MAE=12.97 XGBoost R
2=0.50
MAE=15.55 XGBoost
Results from Table 23, revealed that the data collected from wearable devices exhibited superior predictive
capabilities, as indicated by an R2 value of 0.56 and a MAE of 12.97. In contrast, the data derived from
workstation addons yielded comparatively lower predictive accuracy, with an R2 value of 0.50 and an MAE
of 15.55. The favorable predictive performance of the model utilizing wearable device data, is particularly

69
noteworthy. These results are highlighted further when considering the performance of the model that
incorporates all available data streams, which yielded an R2 value of 0.60 and an MAE of 10.52 (Table 21).
While the combined model achieved slightly better accuracy, the wearable device model's performance
remains comparable, emphasizing its potential as a standalone predictive tool.
The findings unequivocally establish that integrating data from wearable devices into productivity models
yields markedly superior predictive outcomes compared to relying solely on workstation addons. The
essence of this disparity lies in the precision and granularity of the data captured. Wearable devices,
carefully selected for this study, exhibited a remarkable concentration of physiological data, boasting highfrequency sampling rates. For instance, the Empatica E4 [289] meticulously recorded BVP (64Hz), ST
(4Hz), EDA (4Hz) and wrist acceleration (32Hz), whereas the H10 Polar collected heart Rate (1kHz) [290]
at near real-time intervals. This high-resolution data allowed us to discern nuanced shifts in an individual's
physiological responses, thus enabling more accurate productivity assessment.
Conversely, the Kinect camera, operating at a somewhat modest 10 frames per second (fps), while proficient
in capturing facial expressions and body movements, may have occasionally missed subtler cues. In
particular, minor fluctuations in facial expressions that could indicate nuanced emotional states might not
have been entirely captured, potentially limiting the depth of contextual information obtained. Furthermore,
the human-computer interaction features, such as mouse clicks and keyboard keystrokes, while valuable in
understanding participant engagement with computer-based tasks, might not provide the most
comprehensive representation of productive work. Participants might have been engaged in cognitively
demanding activities, such as reading and processing information on the screen or formulating ideas for
written responses, which might not manifest through these interaction metrics but still constitute productive
work. Therefore, solely relying on these metrics could underestimate the actual productivity levels of
participants during computer-based tasks.
Furthermore, wearable devices allow for unobtrusive data collection without altering participants' natural
work routines, mitigating concerns about participant awareness and potential bias [291]. Lastly, the discreet
nature of wearable devices somewhat addresses privacy concerns associated with camera-based systems or
tracking applications, ensuring participant comfort and the representation of natural work behaviors.
In practical terms, organizations striving for maximum predictive accuracy in productivity monitoring may
choose to invest in a fully equipped workstation setup (wearable devices & workstation addons), as it allows
for maximum productivity prediction accuracy. However, recognizing that not all organizations possess the
necessary resources—be it financial, time, or data infrastructure—to support extensive data collection from
workstation addons like Kinect cameras, our research underscores the practicality of utilizing simple
wearable devices. These devices offer a cost-effective and efficient alternative for monitoring productivity,
allowing organizations to achieve a predictive accuracy that compares favorably to the maximum prediction
accuracy attainable. This flexibility empowers organizations to tailor their productivity monitoring to suit
their unique circumstances, objectives, and resource constraints.
7.3.Limitations and Future Work
While this study provides valuable insights into the predictors and mechanisms of productivity, it is
important to acknowledge several limitations. The research was conducted in a controlled laboratory
setting, limiting the generalizability of the findings to real-world office environments. To address this, future
research should validate the findings in diverse work contexts. Additionally, the correlational nature of the

70
study restricts the establishment of causality between predictors and productivity outcomes. Longitudinal
and experimental designs are needed to uncover causal relationships and underlying mechanisms.
Furthermore, it is imperative to consider the demographic specificity of the studied population as a
limitation. The participants primarily comprised young undergraduate and graduate students within a
confined age range, possessing distinctive educational backgrounds, high motivation, and elevated
cognitive reserves. Future research endeavors should consciously aim to engage a more heterogeneous
participant pool, encompassing varied age groups, educational levels, and cognitive reserves, to enhance
the generalizability and applicability of the findings in real-world, diverse work settings.
Another limitation is the use of an 80%-20% model evaluation approach, which may affect the
generalizability of the results. The results obtained using the leave-one-participant-out method and crossvalidation did not demonstrate high prediction accuracy compared to the 80%-20% split method. This
discrepancy in performance could be attributed to the substantial personal variability present within our
dataset, thus compromising the generalizability of the findings. For that, future research should explore
ways to incorporate individual characteristics, such as age and personality traits, into the prediction model.
Moreover, it is crucial to acknowledge the significance of privacy concerns. Nevertheless, it is worth
mentioning that engagement in productivity monitoring programs can be optional, granting employees the
freedom to decide whether to participate. Upholding privacy rights and guaranteeing clear and open
communication regarding data utilization will be vital when integrating these monitoring systems. Practical
implementation and integration of the model within workplace systems should be explored, along with
evaluating its effectiveness in improving productivity and employee well-being. By addressing these
limitations, researchers can advance our understanding of productivity and its management in office
environments.
7.4.Conclusions
To the best of our knowledge, this research represents the pioneering application of a ML framework to
predict perceived productivity based on physiological and behavioral features in the context of smart
workstations. The results showed incorporating predictions of office workers’ psychological states such as
stress arousal, eustress, distress, and mood alongside physiological and behavioral features result in
improved productivity prediction. The feature importance analysis conducted in this study aimed to uncover
the key predictors of productivity and shed light on the role of emotional states in the prediction process.
Emotional states emerged as significant contributors, with mood, eustress and distress playing influential
roles. The study also identified important physiological and behavioral features related to productivity, such
as ST, EDA, wrist acceleration, facial movements, head rotation, and gaze angle. Finally, a comparative
analysis between wearable devices (Empatica E4 & H10 polar) and workstation addons (Kinect camera &
Mini Mouse Macro) showed that data collected from wearable devices outperformed workstation addons
in predicting productivity, highlighting the potential value of wearable devices as a standalone tool for
productivity assessment.
This research has significant implications for office design and management, specifically in enhancing
productivity. The extended productivity prediction model, considering emotional states, physiological
responses, and behavioral characteristics, enables an office that is aware of workers' emotional and
cognitive states. This allows for an adjustable workspace that dynamically adapts to foster productivity. For
instance, integrating a smart workstation with productivity prediction features into office systems enables

71
real-time monitoring and response to workers' emotional and physiological states. Through intelligent
lighting, temperature control, and ambient music, the office environment can be optimized to promote
positive mood states and high levels of eustress, thereby enhancing productivity [291]. Also, the proposed
model allows for targeted interventions and personalized approaches to productivity enhancement.
Individual workers can receive feedback and guidance based on their unique profiles, enabling them to
understand and regulate their emotional states and behaviors for optimal productivity. By employing the
extended productivity prediction model, organizations can take a proactive approach to enhancing
productivity, resulting in higher job satisfaction, improved performance, and increased overall well-being
among office workers [292].

72
Chapter 8: Cognitive Performance, Creativity and Stress Levels of
Neurotypical Young Adults Under Different White Noise Levels
Of all IEQ factors, environmental noise is a primary reason for dissatisfaction for employees, resulting in
degraded productivity, deteriorated performance, and increased stress levels. To solve this problem, office
managers rely on white noise masking solutions. Masking is the practice of using sound to cover and
interfere with disruptive noise in an office environment, by adding a layer of barely noticeable, speciallytuned, and electronically generated sound (e.g., white noise), distracting conversations and sudden noises
are minimized. However, there is limited research about the impact of masking white noise on the
performance and stress of office workers. The following sections of this chapter are organized as follows:
Section 8.1 explains in detail the methodology adopted. Section 8.2 provides a summary of the statistical
analysis conducted. Section 8.3 offers a detailed interpretation of the study findings. Finally, Section 8.4
summarizes the conclusions.
8.1.Methodology
8.1.1. Participants
Forty adults participated in this study voluntarily. A power analysis was conducted using G*Power version
3.1.9.7 to determine the sample size[293]. For an effect size (f = 0.2) and an α error probability = 0.05, the
obtained sample was sufficient for a power of 80%. Of the 40 participants, 24 identified themselves as male
and 16 as female. The average age of the participants was 25.82 ± 7.53. Also, 37 specified that they consider
their right hand as the dominant one, and the remaining 3 indicated that their left hand is dominant. All 40
participants were graduate students at the University of Southern California. The study was limited to
participants between 18 and 64 years old. Individuals with visual problems, hearing deficits, noise
sensitivity, and/or physical injuries, making it uncomfortable to sit for a long period, were not eligible to
participate. A screening survey was used for determining eligibility. If a participant felt uncomfortable
during the experiment, they were given the option to discontinue at any point. Every participant performed
an online hearing screening[294] to ensure they had normal hearing sensitivity in both ears. This test
consists of three main parts: (1) four self-evaluation questions about hearing abilities, (2) tone testing at
500, 1000, and 4000 Hz, and (3) conversation comprehension where participants would listen to a short
conversation and respond to related questions[294]. At the end of the test, participants would be provided
with a hearing report stating their hearing loss level. Only participants with no hearing loss were eligible to
be part of the experiment. One participant was excluded from the analysis due to unrealistically fast
response speed on survey responses; data from the remaining 39 participants were included in all analyses.
The study was approved by the Institutional Review Board of the University of Southern California (UP20-00389 IRB study number). All participants reviewed the informed consent and agreed to participate in
the experiment. All experiments were performed in accordance with relevant guidelines and regulations.
8.1.2. Auditory conditions
The experiment consisted of three auditory conditions: white noise at 45 dB, white noise at 65 dB, and
ambient noise. According to the Center for Disease Control and Prevention (CDC), people exposed to a 70
dB noise level for a prolonged time might feel overwhelmed and annoyed[295]. To eliminate this effect on
our experimental results, we chose a 65 dB as the high white noise condition. On the other hand, the U.S.
Environmental Protection Agency (EPA) recommends a 45 dB for indoor activities which is why a 45 dB
was chosen our low white noise condition[296]. White noise was presented via Bose QuietComfort 35

73
headphones with active noise cancellation to reduce background noise. The white noise was generated using
Audacity software version 3.0.2; this software has been used previously by several research studies to
generate white noise tracks[297],[298],[299]. The experiments were conducted using a Lenovo ThinkPad
X390 Yoga (Intel Core i7-8565U CPU @ 1.8GHz equipped with 16GB RAM) with Realtek Audio drivers
(version 6.0.8757.1). The baseline condition was set by asking participants to complete their tasks without
wearing the headphones; in this condition, participants were exposed to the ambient noise of the office
space. During this condition, the noise level was measured continuously using a BAFX digital sound
meter[300]. The noise meter was positioned immediately behind the participant at ear level. The average
ambient noise level across all participants was 42.3 dB with a standard deviation of 1.2 dB.
8.1.3. Test battery
Five different tests were employed in this study. Cognitive performance assessment included attention using
the continuous performance test, learning and inhibition via the Stroop test, and memory using a two-back
test. Creativity was evaluated via the remote associate test, and the speed and accuracy of work were
measured using a writing performance test. All tests were completed in Psychopy software version
2021.1.0[301]. A brief explanation was added before every test to inform the participant of the task’s nature
and to provide brief instructions for proper completion. It is worth noting that the tests did not have a fixed
time, because the progress of the test is response-dependent; as soon as the participant provides a response,
the next question is presented immediately. A thorough description of each test follows.
Continuous performance test: This test measures sustained attention, which is defined as the ability of an
individual to focus on a stimulus for a certain period while ignoring distracting stimuli[302]. Previous
studies report a moderate reliability level of the continuous performance ranging between 0.4 and 0.7[303].
The test includes 16 different stimuli formed by combinations of four shapes (i.e., star, circle, square, and
triangle) and four colors (i.e., yellow, red, white, and blue). During the test, participants were presented
with a total of 320 stimuli, each appearing on the screen for 0.3-sec followed by a 1-sec inter-stimulus
(blank screen) period before presenting the next stimulus. Participants were asked to press the “Enter”
keystroke whenever they saw the target stimulus: a red star. If a participant failed to react within the 1.3-
secs time span when a red star appeared or pressed the “Enter” keystroke when a shape other than the red
star was presented, the response was marked as incorrect. The target stimulus accounted for 30% of the
images. Color-conjunctive distractors (red non-star) appeared in 17.5% of the trials, and shape-conjunctive
(non-red star) appeared in 17.5% of the trials. The remaining 35% of the trials were non-conjunctive
distractors, where the shape and color were different from the target stimulus. The order of appearance of
stimuli changed every time the test was run to limit any learning effect across the different auditory
conditions.
Stroop test: This test assesses an individual’s selective attention and inhibition, that is, the capability to
overcome a learned response[304]. The Stroop test is widely accepted as a reliable assessment for inhibition
and selection attention. For instance, Siegrist[303] reported a 0.73 reliability for the Stroop test among
adults. In this test, participants were presented with 16 different combinations of four-color words in the
same four ink colors (i.e., blue, red, yellow, and green). The test consisted of 120 trials, with 50% of the
trials showing consistent word and ink color combinations and the remaining 50% presenting color words
printed in an inconsistent ink color (e.g., the word “yellow” written in “red” ink). Each word appeared for
1 sec followed by 1 sec of blank screen before presenting the next word. Participants were required to
indicate the ink color, not the color represented by the word itself, with a keystroke of numbers 1 through
4 each associated with a respective color. To help the participants make this association, colored pieces of

74
paper covered each of the number keys according to the color they represent (e.g., a blue piece of paper
covered the key for number 1). If a participant failed to react within the 2 sec period or pressed the wrong
key, the response was marked as incorrect. The order of the trials changed every time the test was run to
limit any learning effect across the different auditory conditions.
Two-Back test: This test assesses working memory[305]. The test is reported to have a moderate to high
reliability level[306]. Participants were presented with a sequence of letters and pressed the “Enter” key
when the current letter was the same as the letter presented 2 steps earlier in the sequence. The full sequence
was composed of 120 letters, each appearing on the screen for 0.5 sec followed by a 1.5 sec of blank screen
before proceeding to the next letter. If a participant failed to react within the 2 sec time span when they
should have pressed the “Enter” keystroke or pressed the “Enter” keystroke falsely, the response was
marked as incorrect. Out of the 120 trials, 30% were target letters while the remaining 70% were non-target.
Participants were presented with a different list of letters in each condition to limit any learning effect.
Remote associate test: This test assesses creativity levels, particularly an individual’s ability to make
associations[307]. The test has been extensively used and is highly reliable. The Spearman-Brown
reliability reported by Mednick was 0.92[308]. Participants were presented with three cue words and were
asked to determine a fourth word that links the other three words together (e.g., cottage, Swiss and cake are
three cues linked by the word “cheese”). Word selections for this study were acquired from a word bank
developed from previously published studies[309]. No time limit was allocated for this test; participants
could take as much time as they needed to provide an answer before moving to the next set of words. Each
trial had only one correct answer, and sores were calculated as the number of correct answers out of 10
trials for each auditory condition.
Typing performance test: This test measures an individual’s speed and accuracy in preforming computer
work[310]. Participants were presented with a printed paragraph and were asked to type the text into a
digital format on a computer. Automatic spelling and grammar checks were disabled in the word processing
software during the test. Writing speed was measured as the time needed to type the paragraph, and accuracy
was measured as the number of errors made. Participants were provided with a different paragraph of
similar size (300 words) and difficulty level (elementary level) for each auditory condition. These
paragraphs were acquired from a public resource[311].
8.1.4. Electrodermal activity
EDA is a measure of variation in electrical conductance at the surface of the skin[312]. EDA is associated
with emotional arousal, stress intensity, and increased cognitive workload of individuals[312], and
therefore, is considered a valid physiological indicator of stress[313]. In this study, EDA was monitored
continuously using a wrist band sensor (Empatica E4), which applied an unnoticeable yet continuous
voltage to the skin surface to measure variation in skin conductance. EDA was measured in microSiemens
(μS) with a sampling frequency of 4 Hz (non-customizable) and a range of 0.01 μS to 100 μS[314]. EDA
analysis was completed using two components[313]: (1) the tonic component which refers to slow
variations in the EDA signal over time measured through the tonic component and (2) the phasic component
which refers to rapid and smooth transient events noticeable in the EDA signal. The MATLAB Ledalab
toolbox was used to analyze the EDA raw data[252]. The software uses the “Continuous Decomposition
Analysis” to decompose the raw EDA data into the tonic and phasic components. The full analysis
comprises four steps: estimation of the tonic component, nonnegative deconvolution of phasic SC data,
segmentation of driver and remainder, and reconstruction of SC data[315]. However, in this study, the
analysis is solely focused on the tonic component of the EDA since the experimental procedure did not

75
include any specific stress-inducing events that required identifying sudden changes via EDA’s phasic
component. The tonic component of the EDA data is computed based on the mathematical process of
deconvolution[316], where only data intervals that do not reflect any phasic activity are used to estimate
the tonic component. Significant peaks in the EDA data are detected whenever a local maxima shows a
difference of 0.2 μS, in comparison to a preceding and succeeding local minima. These peaks are the
indicators of phasic activity. Thus, the tonic component is calculated by averaging the values of the driver
function governing the EDA data outside the phasic activity intervals. For more details about the
“Continuous Decomposition Analysis”, please refer to the following studies: [315], [316] The difference in
the mean tonic activity between baseline EDA and the EDA during the auditory conditions was calculated
for every participant. This difference was compared across the different conditions to determine the effect
of noise on the EDA.
8.1.5. Procedure and Experimental Design
The experiment took place in a private office at a time when no occupants were present in neighboring
offices to limit external distractions. The window’s blinds were kept shut and the same artificial lighting
conditions were maintained for all participants to limit the effect of lighting on participants’ performance.
Similarly, a 24 °C indoor temperature was set and maintained throughout the experiment. This setup mimics
a private office setting with standard lighting and thermal conditions and no distracting noise (e.g., no
telephone rings, printer noise, chat, etc.) At the outset, participants indicated their gender, age, and dominant
hand, reported any sensitivity to noise and completed the online hearing screening. The E4 [314] was placed
on the participant’s wrist, and the participant remained still for 5 minutes to collect an EDA baseline. Then,
participants completed the 5 tests in the same order under each auditory condition, starting with the
continuous performance test, followed by the Stroop test, two-back test, remote associate test, and finally
the typing performance test. Participants were exposed to the 45 dB and 65 dB white noise continuously
through the headphones while performing the tests. During the ambient noise condition, participants were
asked to complete their tests without wearing the headphones. The study follows a within-subject
experimental design, where every participant completed all three conditions. The order of the three auditory
conditions was randomized for each participant using a Latin square design[317]. The total duration of the
experiment was around 2.5 hours.
8.1.6. Data Analysis
A repeated-measures analysis of variance (ANOVA) was used to analyze the outcomes under study during
the various auditory conditions as a within subjects’ factor. In this study, the dependent variables are
sustained attention, selective attention, inhibition, working memory, creativity, performance, and stress
level. The independent variables are the three noise conditions: white noise at 45 dB, white noise at 65 dB,
and ambient noise. Tukey HSD analysis was employed to examine the significant differences in the
outcomes between each of the three conditions. A p-value of 0.05 was used to determine statistical
significance. The statistical analysis was conducted using the IBM SPSS statistics software, version
27[318].
8.2.Results
8.2.1. Sustained Attention – Continuous Performance Test
Sustained attention was significantly different among the auditory conditions [F(2,114) = 3.92, p=0.02, d=
0.51, 95% CI [0.01, 0.15]]. Specifically, participants’ scores on the continuous performance test were

76
significantly higher in the 45 dB white noise condition (M= 95.23%, SD= 4.06%) compared to the ambient
noise condition (M= 93.12%, SD= 3.34%). No significant differences in sustained attention were found
between the 65 dB white noise (M= 93.29%, SD= 3.58%) and the remaining two auditory conditions.
8.2.2. Selective Attention and Inhibition – Stroop Test
No significant effect was noted on participants’ selective attention or inhibition assessed by the Stroop test
[F(2, 114) =0.49 , p=0.61, d= 0.20 , 95% CI [0.00, 0.05]].
8.2.3. Working Memory – Two-Back Test
Significant differences in working memory [F(2, 114) =3.34 , p=0.04, d= 0.46, 95% CI [0.00, 0.14]], were
reflected in participants’ scores on the Two-Back test being significantly better in the 65 dB white noise
condition (M= 66.38%, SD= 3.74%) compared to the ambient noise condition (M= 64.26%, SD= 3.86%).
No significant differences in working memory were found between the 45 dB white noise (M= 64.71%,
SD= 3.80%) and the remaining two auditory conditions.
8.2.4. Creativity – Remote Associate Test
ANOVA results comparing remote associate test scores showed a significant impact of the auditory
condition on participants’ creativity levels [F(2, 114) =3.89 , p=0.02, d= 0.51, 95% CI [0.00, 0.16]]. These
differences aligned with sustained attention as being significantly higher with 45 dB white noise (M=
65.13%, SD= 24.69%) compared to ambient noise (M= 50.77%, SD= 24.53%). No significant differences
in the creativity measure were found between the 65 dB white noise (M= 54.28%, SD= 27.16%) and the
two other auditory conditions.
8.2.5. Performance – Typing Performance Test
Auditory conditions had a significant effect on both the number of mistakes made by the participants [F(2,
114) =5.13, p=0.01, d= 0.59, 95% CI [0.01, 0.18]] and the time required to complete the typing task [F(2,
114) =4.62 , p=0.01, d= 0.55, 95% CI [0.00, 0.17]]. Participants made more mistakes (M= 10.77, SD= 6.21)
and took more time (M=543, SD= 171) when working in the 65 dB white noise condition. This difference
was statistically significant compared to the mistakes made during the 45 dB white noise (M=7.38, SD=
3.92) and ambient noise (M= 7.33, SD=5.85) and compared to the time required in the 45 dB white noise
condition (M=441, SD=136). No significant differences were noted between the number of mistakes made
under the 45 dB white noise condition and the ambient noise condition, nor in the amount of time required
during the ambient noise condition (M=466, SD= 149) and the two other auditory conditions.
8.2.6. Stress – Change in Mean Tonic Activity
Average changes in the tonic activity from baseline were noted to be different across the conditions [F(2,
114) =3.26 , p=0.04, d= 0.46, 95% CI [0.00, 0.14]]. Specifically, changes from the baseline tonic activity
during the 45 dB white noise condition (M=−0.20, SD=0.91), was found to be significantly different than
changes in the tonic activity during the 65 dB white noise condition (M=0.22, SD=0.74). On the other hand,
changes from baseline tonic activity during the ambient noise condition (M=0.13, SD=0.62) were not
significantly different from the two other auditory conditions.

77
Average scores for the five tests and changes in the mean tonic activity across the three auditory conditions
are provided in Table 24.
Table 24. Statistical analysis of study measures under different noise conditions
Dependent Variable White noise 45 dB Ambient noise White noise 65 dB F(1, 114) p value M SD M SD M SD
Sustained attention (%
correct) 95.23 4.06 93.12 3.34 93.29 3.58 3.92 0.02
Selective attention and
inhibition (% correct) 91.99 8.40 90.11 8.32 90.58 8.64 0.49 0.61
Working memory (%
correct) 64.71 3.80 64.26 3.86 66.38 3.74 3.34 0.04
Creativity level (%
correct) 65.13 24.69 50.77 24.53 54.28 27.16 3.89 0.02
Performance (Number
of mistakes) 7.38 3.92 7.33 5.85 10.77 6.21 5.13 0.01
Performance (Time in
seconds) 441 136 466 149 543 171 4.62 0.01
Δ Mean Tonic Activity
(Microseconds) −0.20 0.91 0.13 0.62 0.22 0.74 3.26 0.04
Additionally, a summary of the post hoc analysis is presented in Table 25.
Table 25. Post-hoc analysis summary
Dependent Variable Conditions under Comparison |Mean Difference| p value
Sustained attention (%
correct)
WN 45 dB x Ambient Noise 0.02 0.03
WN 45 dB x WN 65 dB 0.01 0.06
WN 65 dB x Ambient Noise 0.01 0.98
Working memory (%
correct)
WN 45 dB x Ambient Noise 0.01 0.861
WN 45 dB x WN 65 dB 0.01 0.134
WN 65 dB x Ambient Noise 0.02 0.041
Creativity level (%
correct)
WN 45 dB x Ambient Noise 0.14 0.038
WN 45 dB x WN 65 dB 0.14 0.047
WN 65 dB x Ambient Noise 0.01 0.996
Performance (Number
of Mistakes)
WN 45 dB x Ambient Noise 0.05 0.999
WN 45 dB x WN 65 dB 3.42 0.019
WN 65 dB x Ambient Noise 3.43 0.017
Performance (Time in
seconds)
WN 45 dB x Ambient Noise 24.5 0.760
WN 45 dB x WN 65 dB 101.1 0.012
WN 65 dB x Ambient Noise 76.6 0.074
Δ Mean Tonic Activity
(Microseconds)
WN 45 dB x Ambient Noise 0.33 0.147
WN 45 dB x WN 65 dB 0.42 0.043
WN 65 dB x Ambient Noise 0.09 0.848
8.3.Discussion

78
This study examined the effect of white noise levels on cognitive performance, creativity, and stress levels
in neurotypical young adults, which is not well studied in the literature. In general, white noise level at 45
dB resulted in better cognitive performance in terms of sustained attention, accuracy, and speed of
performance as well as enhanced creativity and lower stress levels. The 65 dB white noise condition led
only to improved working memory. Results related to creativity, performance and stress levels are especially
important to note because white noise condition at 45 dB resulted in significantly better creativity levels
compared to the ambient noise at around the same dB level. This points out to the signal characteristics of
white noise at 45 dB supporting creativity and shows that white noise condition at 65 dB neither reduces
nor improves creativity compared to ambient noise at 45 dB. In addition, white noise condition at 45 dB
resulted in significantly better performance both in terms of accuracy and speed compared to the white
noise condition at 65 dB. Moreover, participants had lower levels of stress during white noise condition at
45 dB compared to the white noise condition at 65 dB.
Previous studies have shown that white noise results in improved recognition memory [319], speed of
arithmetic computations [320] and lexical acquisition of novel word forms [297] in neurotypical adults.
Our results extend these findings by showing that white noise at 45 dB and 65 dB enhanced sustained
attention and working memory, respectively, in comparison to the ambient noise. Yet, the differences in
these two outcomes were not substantial to have major consequences from a practical aspect. Nevertheless,
the fact that white noise conditions –not the ambient noise condition– triggered better cognitive
performance remain worthy of note. This may be as a result of the white noise sound characteristics (pitch
and frequency), which makes it resemble the sound of rain, waves or the wind going through tree leaves
[321], and allows it to be perceived as pleasant to the senses [322] in comparison to the ambient noise.
Previous studies suggest that low to moderate white noise levels can be enough to induce a high construal
level leading to better abstract processing thus enhancing creative thinking [323]. On the other hand, the
literature also presents a plethora of studies demonstrating that high levels of white noise can impede the
creativity of individuals. For instance, Martindale and Greenough [324] concluded that a 75 dB white noise
resulted in the lowest scores on the remote associate test in comparison to the control (no white noise)
condition. Similarly, results from the study conducted by Hillier et al.[325] showed that a 90 dB white noise
would hinder creative thinking compared to the control condition. This is because high white noise levels
have been associated with increased distraction, resulting in deteriorated information processing, and thus
degraded creativity [323]. In our study, white noise at 65 dB was not too high to impede creativity as
suggested by previous studies in related literature. However, our findings show that white noise of 45 dB
could support creative thinking in comparison to ambient noise (42.3 dB). This is an important finding as
the ambient noise level was relatively equal to the white noise level at 45 dB, which highlights the unique
properties of white noise in supporting creativity.
The white noise level at 45 dB allowed for better typing performance in terms of speed and accuracy and
led to reduced EDA levels when compared to the white noise at 65 dB. EDA levels have been widely used
as indicators of stress in experimental procedures related to environmental interventions [326], [327]. Thus,
our results support the conclusion that white noise at 45 dB resulted in reduced physiological stress whereas
the 65 dB white noise exposure led to increased stress levels while performing cognitively demanding tasks.
These findings agree with previous research studies showing that long exposure to high-level white noise
results in a stressful response. For example, Liu et. al. [328] showed that the exposure to an 80 dB white
noise level was enough to induce more stress than a mental arithmetic task. Similarly, Kraus et al. [329]
argue that at higher noise intensities, activations of sympathetic nervous activity (bodily response to
stressors) are prominent. To that end, we postulate that high-stress levels during the 65 dB white noise

79
condition were associated with deteriorated speed and accuracy. This conclusion was further supported by
the work of Loewen and Suedfeld [330] who found that masking office noise with a 60 dB white noise led
to high arousal and stress as well as reduced task performance among office workers in comparison to a nonoise condition. On the other hand, our 45 dB white noise condition results are unique; to the best of our
knowledge, there has been no study that investigated the effect of low-level white noise on task performance
or stress levels among neurotypical adults.
Finally, our results indicate that different tasks might require different white noise levels for optimal
performance: at 45 dB white noise level, sustained attention, accuracy, and speed were optimal but working
memory improved under the 65 dB white noise level. Research suggests that the necessary dopamine levels
for optimal cognitive performance can vary depending on the type of task [297]. For example, memory
tasks are usually highly mentally demanding and thus, require higher dopamine levels [331], which could
explain why the 65 dB white noise condition boosted the memory performance of our participants.
While this study contributed to the literature in unique ways as highlighted above, it also holds some
limitations. The study did not present enough variability in age and gender to determine the effect of
demographics on the relationship between white noise levels and cognitive performance. Also, the study
covered two white noise conditions only (45 dB and 65 dB), thus future research directions could perform
more studies to uncover the relationship between various white noise levels and cognitive performance, by
recruiting more participants as well as examining more white noise levels (e.g., 55 dB and 75 dB) and the
personal differences based on gender, age, etc. Moreover, researchers can investigate the effect of different
noise colors (e.g., pink, brown, etc.) [332] on the cognitive performance of neurotypical adults, using the
same experimental procedure presented in this chapter. Additionally, this study did not perform a concurrent
analysis of participants’ dopamine levels to biologically explain the results at hand. To that end, future
research efforts can measure dopamine levels to confirm our conclusions and determine the governing
relationship between the dopaminergic circuitry and cognitive performance under various white noise
levels. This can be accomplished by measuring the concentration of injected radioligand (radioactive
biochemical substance) using a positron emission tomography camera which helps detect the dopamine
released in the brain [333]. On the practical side, results from this chapter can be used to enhance our
understanding of customized workspaces. Hence, future research can investigate the means and methods to
integrate the use of white noise as a performance booster in the workplace and customize the exposure to
different white noise levels to fit into the requirement of the work task at hand.
8.4.Conclusion
This study examined the effect of two white noise conditions, white noise level at 45 dB and white noise
level at 65 dB, on the cognitive performance, creativity, and stress levels of neurotypical young adults in a
private office space. Our findings showed that white noise level at 45 dB resulted in better cognitive
performance in terms of sustained attention, accuracy, and speed of performance as well as enhanced
creativity and lower stress levels. On the other hand, the 65 dB white noise condition led to improved
working memory but higher stress levels, which leads to the conclusion that different tasks might require
different noise levels for optimal performance. These findings are significant, as they extend previous
research results about the positive effects white noise has on the cognitive performance of neurotypical
adults. Future research directions presented include studying more white noise levels and different noise
colors. Similar research might perform a concurrent analysis of participants’ dopamine levels to biologically
explain the results.

80
Chapter 9: The Impact of Light’s Color Correlated Temperature and
Illuminance Levels on Stress and Cognitive Functions Restoration
Lighting significantly affects people's physiological reactions and cognitive abilities. However, no research
has yet concentrated on assessing light's effectiveness in lessening stress or enhancing cognitive
performance. This research conducts a controlled experiment to examine how various lighting strategies,
characterized by different color temperatures and brightness levels, can alleviate stress following workrelated tension and rejuvenate cognitive functions after mental exhaustion. The following sections of this
chapter are organized as follows: Section 9.1 explains in detail the methodology adopted. Section 9.2
provides a summary of the statistical analysis conducted and offers a detailed interpretation of the study
findings. Finally, Section 9.3 summarizes the conclusions.
9.1. Methodology
9.1.1. Experimental Design
A between-subjects design was employed in the implementation of this experimental study. The procedural
framework of the experiment involved inducing stress among participants in the initial phases, while
depleting their cognitive capacities, followed by an attempt to subsequently ameliorate these abilities
through lighting interventions. Consequently, the utilization of a within-subjects design proves impractical,
as it would present challenges in the reinduction of stress to participants who would possess anticipatory
awareness regarding the nature of tasks they are to engage in.
9.1.2. Research Settings
The research was carried out in a private office, with an area of 25m2
, situated at the University of Southern
California. The ambient conditions were constant across participants and during the experimental
procedure. To ensure the ecological integrity of the experimental environment, the indoor environmental
quality was subject to continuous monitoring via the Awair sensor [334]. The temperature was set at a
comfortable 22 degrees Celsius for everyone, and the noise level averaged around 47±2 decibels. The
recorded humidity levels sustained a mean value of 50±7%, while the concentration of carbon dioxide
(CO2) consistently registered below the threshold of 300ppm. Furthermore, the concentration of particulate
matter with a diameter of 2.5 micrometers or smaller (PM2.5) remained consistently below 3µg/m³. To
mitigate potential confounds arising from natural daylight, the window blinds were consistently maintained
in a closed position throughout the entire experimental duration.
9.1.3. Participants
One hundred adults willingly took part in this study, with twenty individuals assigned to each of the five
experimental conditions: 100 lux & 7000K, 100 lux & 3000K, 1000 lux & 7000K, 1000 lux & 3000K, or
the baseline of 500 lux at 3700K. The adequate sample size was determined through a power analysis using
G*Power version 3.1.9.7. This analysis aimed at identifying an appropriate sample size based on an effect
size (f = 0.3) and a significance level (α) of 0.05, resulting in a sample size that provided 80% statistical
power. Out of the 100 participants, 60 identified as male, 39 as female, and one as non-binary. Participants
were allocated to the conditions to ensure an equal gender distribution across the various conditions. The
number of male participants varied from 11 to 13, while the number of female participants ranged from 7
to 9 across the five conditions. The participants' average age was 24.43 ± 2.84 years. All participants were
students at the University of Southern California, and the study's scope encompassed individuals aged 18
to 64. Those with visual impairments, sensitivity to lighting, or physical injuries hindering prolonged sitting
were excluded from participation. Additionally, individuals taking medications influencing physiological

81
indicators like heart rate and blood pressure were ineligible for the experiment. A screening survey was
employed to assess eligibility, and participants had the freedom to withdraw from the experiment if they
felt uncomfortable. The Institutional Review Board of the University of Southern California approved the
study. Every participant thoroughly reviewed the informed consent and willingly agreed to partake in the
study, which adhered to all pertinent guidelines and regulations.
9.1.4. Experimental Procedure
Upon arrival at the office space, participants were briefed about the experiment. They were then instructed
to wear the Empatica E4 wristwatch and to keep their hand still throughout the experiment to avoid motion
artifacts in the physiological readings. Participants only needed one hand to complete the experiment and
used the one not adorned with the E4.
The experiment initiated with a 5-minute physiological baseline, aimed at stabilizing physiological signals
such as heart rate and electrodermal activity, which could have been disturbed due to the physical exertion
participants experienced (e.g., walking, biking) enroute to the experiment site. They then completed the 6-
item State Trait Anxiety Inventory (STAI-6) to gauge initial perceived stress and answered questions from
the Swedish Occupational Fatigue Inventory (SOFI) to determine baseline cognitive fatigue. Further details
on the physiological signals, the SOFI and STAI-6 can be found in the section below.
Following this baseline assessment, participants engaged in 30 minutes of mentally demanding tasks,
comprising the Stroop Test and Mental Arithmetic Task. Their cognitive performance was subsequently
assessed using the Continuous Performance Test (CPT) and the Visual Backward Digit Span (VBDS),
which are tools tailored to evaluate attention and memory. Upon task completion, participants filled out the
STAI-6 and SOFI once again to identify any changes in their perceived stress and cognitive fatigue.
The subsequent 10-minute phase marked the period of restoration. Prior research has employed restoration
periods for controlled experimental investigations centered around stress or cognitive functions restoration,
ranging from 5 to 10 minutes [118], [335], [336]. To adequately establish any effects of restoration, we
allocated a full 10 minutes for this purpose. During this interval, participants were subjected to one of five
lighting conditions: 100 lux & 7000K, 100 lux & 3000K, 1000 lux & 7000K, 1000 lux & 3000K, or the
baseline of 500 lux at 3700K. Further details about these lighting conditions are available in the Lighting
Conditions section. During the restoration period, participants were prompted to detach from their tasks
and purposefully rest and relax.
Following the restoration phase, participants once again completed the STAI-6 and SOFI to gauge their
perceived stress levels and the success of the cognitive functionsrestoration process. To wrap up the session,
they retained the same lighting condition from the restoration phase and revisited the CPT and VBDS for
10 minutes.
All tests and surveys were completed in Psychopy software version 2021.1.0 [301]. A brief explanation was
added before every test to inform the participant of the task’s nature and proper completion. Figure 7
provides an overview of the experimental procedure implemented.

82
Figure 7. Experimental Procedure
9.1.5. Lighting Conditions
The baseline lighting condition, set at 500 lux and 3700K, aligns with the University of Southern
California's office lighting standards as defined by Facility Management and complies with the IES
(Illuminating Engineering Society) guidelines for office environments [337]. We compared this baseline
with four intervention conditions: 1000 lux & 7000K, 1000 lux & 3000K, 100 lux & 7000K, and 100 lux
& 3000K. By choosing 1000 lux and 100 lux, we encompass both the high and low extremes of the lighting
intensity spectrum, effectively simulating the conditions of bright task lighting and soft ambient lighting.
The CCT values of 7000K and 3000K are chosen to explore the impacts of cool and warm light. These
conditions were carefully chosen to represent both higher and lower illumination and CCT levels than the
baseline.
To set up the lighting conditions, we used floor lamps with Torkase 10W Smart Light Bulbs [338]. Both
CCT and illumination level were consistently measured at each participant's eye level using the Color LED
Chroma Light Meter [339] to ensure accuracy and reliability of results. The standard deviation of the
illumination levels among participants within the same group was 6.4 lux, while the standard deviation for
the CCT levels within the same group was 18.45 K.
9.1.6. Measures
Physiological measures: During the experiment, various physiological signals were gathered from the
Empatica E4 wristband [314], including EDA, and heart rate. These signals serve as objective indicators
of individuals' stress responses. Notably, heightened EDA values are commonly associated with elevated
stress levels. The EDA signal was sampled at a frequency of 4 Hz and measured in microsiemens (μS). To
process EDA data, a series of steps were taken. A Butterworth low-pass filter was utilized, in combination
with Hanning smoothing employing a 4-point window, and manual artifact correction to eliminate any noise
stemming from movement or external interference. The mean values of EDA were then utilized for
evaluating stress responses.
For a comprehensive analysis of heart rate (HR) and HRV, the software Kubios was used [249]. An artifact
correction approach was employed, targeting R-R intervals(the time intervals between consecutive R waves
on an electrocardiogram, representing one cardiac cycle) deviating more than 0.25 seconds from the mean.
This correction method preserved data variability while rectifying artifacts. Furthermore, Kubios
implemented a piecewise cubic spline interpolation technique to enhance HRV data quality, resulting in a
more precise and cleaner signal. To assess stress responses, mean HR, and LF/HF ratio (the ratio of lowfrequency to high-frequency power in heart rate variability) were used. Both metrics are widely used in
scientific literature as indicators of stress, with numerous studies showing that higher HR or LF/HF ratios
typically correlate with increased stress, while lower values are associated with relaxation [340].

83
Psychometric tests: Two tests were employed to induce stress and cognitive fatigue: the Stroop test and the
mental arithmetic test. The Stroop test is commonly used in lab-based experiments to induce both stress
and cognitive fatigue [341], [342]. The Stroop test induces stress and mental fatigue by presenting
participants with conflicting visual and linguistic information, requiring intensive cognitive effort to resolve
the interference quickly and accurately. In the same vein, repetitive mathematical calculations in mental
arithmetic tests are known to induce cognitive fatigue and stress [343]. These tests were piloted and refined
several times before the start of data collection to ensure an increase in perceived stress and cognitive
fatigue as well as physiological arousal.
Stroop test: During the test, participants encountered 600 trials featuring five color words displayed in
contrasting ink colors, such as the word "yellow" inked in "red". Words were displayed for 1 second,
succeeded by a 0.5-second blank screen. Participants needed to identify the ink color using keys 1 to 5,
each labeled with a corresponding color paper (e.g., key 1 was covered with red), rather than the word's
meaning. If they failed to respond within 1.5 seconds or made an error, a "wrong" alert with a buzz sound
was triggered to heighten their stress and frustration.
Mental arithmetic test: The test included 100 mathematical equations, each displayed for 6 seconds. These
equations involved addition, subtraction, and multiplication. Participants had to determine the correctness
of each equation within the time frame. If they responded correctly, it was marked as "correct." Any late or
incorrect answers were labeled "incorrect" and triggered a "wrong" alert accompanied by a buzz sound to
amplify stress and create frustration. The set of equations was split evenly between correct and incorrect
solutions. Though each equation was solvable within the allotted 6 seconds, they were crafted to invoke a
sense of time pressure. All numbers in the equations were integers, with a maximum of two digits, and the
results were always positive with one or two digits. Example equations include: 43-9-23=21, 6×7-12=31,
and 23+2×9=41.
Cognitive performance was evaluated using the Continuous Performance Test (CPT) and the Visual
Backward Digit Span (VBDS). These tools are frequently employed in stress restoration research to gauge
the effect of interventions on cognitive recovery following fatigue [344], [345]. The CPT assesses sustained
attention—defined as an individual's capacity to concentrate on a stimulus over time while disregarding
distractions [302]. Meanwhile, backward digit span tasks primarily tests directed-attention mechanisms and
serves as a measure of working memory, a reliable indicator of cognitive performance [346], [335].
Continuous performance test: The test consists of 16 unique stimuli, created from combinations of four
shapes (star, circle, square, and triangle) and four colors (yellow, red, white, and blue). Participants were
shown 160 stimuli in total, with each displayed for 0.3 seconds, followed by a 1-second blank interval
before the next stimulus. They were instructed to press the “y” key for the target stimulus, a red star, and
the “n” key for all other stimuli. If a participant did not respond within 1.3 seconds or mistakenly pressed
the wrong key, it was deemed an incorrect response. The red star represented 30% of the stimuli. Red shapes
that weren't stars constituted 17.5% of trials, while stars not in red made up another 17.5%. The remaining
35% were distractors, neither matching the shape nor color of the target. The sequence of stimuli was
randomized for each test to minimize learning effects between the pre-restoration and post-restoration
phases. The CPT score was determined as the number of correct answers.
Visual backward digit span: For 10 seconds on the computer screen, participants were shown a number.
They were then tasked with memorizing it and typing it out in reverse order. The numbers began with two

84
digits and increased in length until the participant made an error. The VBDS score was determined by the
longest sequence of digits the participant could correctly memorize.
Questionnaires: Two questionnaires were administered to evaluate participants' perceived stress levels and
cognitive fatigue: STAI-6 and SOFI-CF.
STAI-6: The STAI-6 is a streamlined version of the State-Trait Anxiety Inventory, specifically developed
to assess immediate stress levels. With just six items, it measures current feelings of tension, apprehension,
and worry. Its frequent use in stress restoration research underscores its reliability and effectiveness for this
study [347]. This abbreviated inventory, yields similar average scores to the full 20-item STAI focused on
state anxiety [348].
The STAI-6 comprises of three stress-positive questions, such as "I am nervous" or "I am sad", and three
stress-negative ones, like "I am content" or "I am happy". Participants rate their feelings in the moment on
a scale ranging from "Not at all" to "Very Much". For stress-positive items, a higher score indicates more
stress, while the opposite holds true for stress-negative items. The mean score of the six questions
determines the stress level. To ensure consistency in interpretation, the scores for the stress-negative items
were inverted so that an increase in the total score consistently represented higher stress.
SOFI-CF: The Swedish Occupational Fatigue Inventory (SOFI) is a specialized tool developed to evaluate
and quantify fatigue within occupational environments [349]. The SOFI was derived from a factor analysis
that discerned five distinct components addressing both the cognitive and physical facets of fatigue. These
primary factors encompass Lack of energy, Physical exertion, Physical discomfort, Lack of motivation, and
Sleepiness. Together, these components integrate 25 specific descriptors, with each aimed at capturing
nuances of fatigue-related sensations or states.
For our study, the focus was narrowed to cognitive fatigue. Hence, only three components related to it—
Lack of energy, Lack of motivation, and Sleepiness—were incorporated. To elucidate, terms such as
“drained”, “exhausted”, and “overworked” were used to measure Lack of energy; “passive”, and
“uninterested” gauged Lack of motivation; and descriptors like “sleepy”, assessed Sleepiness. These were
rated by participants on a scale ranging from 1 to 10, with 1 signifying subdued sensations and 10
representing heightened intensity. The overall score from the six questions, which ranges between 6 and
60, indicates the level of cognitive fatigue: a higher score signifies greater cognitive fatigue. A similar
assessment of cognitive fatigue was employed in [118], which sought to determine the effect of naturerelated interventions on the restoration of cognitive performance.
9.2. Results & Discussion
9.2.1. Evaluating the effectiveness of inducing stress and mental fatigue
The initial phase of our analysis aimed to determine whether the first 30 minutes of testing resulted in
increased stress and mental fatigue levels. To assess this, we conducted a paired sample t-test for all 100
participants. We performed four paired sample t-test comparisons: comparing perceived stress and mental
fatigue levels using the STAI-6 and SOFI-CF before and after the tests and comparing the difference in
physiological data (average EDA and HR) between the first 5 minutes and the last 5 minutes of the test.
Regarding mental fatigue, the results revealed that the post-test SOFI-CF score (M = 21.75, SD = 10.50)
was significantly higher than the pre-test score (M = 19.39, SD = 9.35; t(99) = -2.34, p = 0.02). In terms of
stress levels, the post-test STAI-6 score was significantly higher (M = 11.58, SD = 3.19) than the pre-test

85
score (M = 8.78, SD = 2.67; t(99) = -8.51, p < 0.001). The difference in EDA between the first 5 minutes
and the last 5 minutes was marginally significant (t(99) = -1.94, p = 0.05). However, the EDA level during
the last 5 minutes (M = 0.60, SD = 0.86) was higher than that during the first 5 minutes (M = 0.50, SD =
0.82). Conversely, the HR data exhibited a more pronounced significance (t(99) = -6.40, p < 0.001) with
the average heart rate during the first 5 minutes of testing (M = 72.25, SD = 4.34) compared to the last 5
minutes (M = 75.36, SD = 6.48).
In summary, these findings suggest that the 30-minute tests administered to the participants effectively
induced an increase in stress and mental fatigue. This lays a solid groundwork for studying the effects of
lighting interventions on alleviating stress and mental fatigue.
9.2.2. Impact of lighting interventions on stress responses
The analysis of physiological responses to various lighting conditions revealed significant alterations in
EDA, heart rate, and LF/HF ratio. Specifically, a considerable reduction in mean EDA was observed (t(19)
= 2.12, p = 0.04), moving from pre-restoration levels (M=0.53, SD=0.66) to post-restoration (M=0.38,
SD=0.40) under the lighting condition of 100 lux, 3000K.
Regarding heart rate, three distinct lighting conditions successfully reduced elevated pre-restoration heart
rates. Under the 100 lux, 3000K condition, heart rates decreased significantly (t(19) = 3.82, p < 0.001) from
a mean of (M=79.62, SD=12.06) to (M=76.03, SD=10.95). Similarly, exposure to 1000 lux, 7000K lighting
resulted in a significant reduction (t(19) = 6.81, p < 0.001) from a mean of (M=81.23, SD=10.73) to
(M=77.57, SD=10.33). Finally, in the 1000 lux, 3000K setting, the mean heart rate declined significantly
(t(19) = 4.20, p < 0.001) from (M=84.21, SD=10.74) to (M=79.94, SD=9.96).
Additionally, the LF/HF ratio, a measure of autonomic balance, demonstrated notable changes. Under 1000
lux, 3000K lighting, the LF/HF ratio decreased significantly (t(19) = 2.32, p = 0.03) from a mean of
(M=2.49, SD=1.48) pre-restoration to (M=1.66, SD=1.03) post-restoration. In the baseline lighting
condition of 500 lux, 3700K, the LF/HF ratio also showed a reduction (t(19) = 2.34, p = 0.03), moving from
(M=2.67, SD=1.58) to (M=1.88, SD=1.29).
The subjective assessment of stress, as measured by the STAI-6, revealed significant reductions in perceived
stress levels under certain lighting conditions during the restoration phase. Notably, exposure to a lighting
condition of 100 lux, 3000K led to a significant decrease in STAI-6 scores (t(19) = -3.04, p = 0.007), with
(M=12.35, SD=3.68) pre-restoration to (M=10.10, SD=2.53) post-restoration. Furthermore, similar stressreducing effects were observed with the 1000 lux, 3000K lighting condition (t(19) = -3.10, p = 0.006). Here,
STAI-6 scores showed a significant decline from a pre-restoration (M=11.15, SD=3.52) to a post-restoration
(M=8.95, SD=2.54).
All these results are presented in Figure 8.

86
Figure 8. Comparison of stress responses metrics pre- and post-restoration

87
9.2.3. Impact of lighting interventions on cognitive performance
As mentioned before, the cognitive performance was assessed through the CPT (indicator of attention) and
the VBDS (indicator of memory). A paired sample t-test was conducted to compare cognitive performance
before and after the restoration phase across five different lighting conditions. This analysis aimed to
ascertain the influence of each lighting intervention on cognitive abilities. For the CPT, we assessed the
accuracy of responses. The data suggest that altering lighting conditions (from the baseline to any of the
four specified lighting conditions) can potentially improve performance, thereby increasing response
accuracy. However, maintaining the baseline lighting environment (500 lux, 3700 K) before and after the
adjustment period did not lead to a notable change in CPT scores.
In the VBDS test, preserving the baseline lighting conditions (500 lux, 3700 K) both before and after
restoration led to a marked decrease in performance. The average score declined from 7.55 (SD = 1.79)
before restoration to 6.80 (SD = 2.28) after restoration, (t(19) = 2.51, p = 0.02). In contrast, altering the
baseline condition (500 lux, 3700 K) to one of the two post-restoration conditions (100 lux, 7000 K or 1000
lux, 7000 K) resulted in a significant improvement in VBDS test scores.
Finally, the findings indicate that a restoration phase under 7000K CCT lighting correlates with lower
Swedish Occupational Fatigue Inventory (SOFI) scores, suggesting a reduced perceived level of mental
fatigue. Specifically, lighting at 100 lux, 7000 K led to a noteworthy reduction in SOFI score (t(19) = 2.28,
p= 0.03) from pre-restoration (M=23.70, SD=9.57) to post-restoration (M=20.10, SD=8.86). Similarly,
lighting at 1000 lux, 7000 K resulted in a significant drop in SOFI score (t(19) = 2.21, p= 0.04) from prerestoration (M=21.15, SD=8.09) to post-restoration (M=17.65, SD=7.90).
All these results are presented in Figure 9.

88
Figure 9. Comparison of cognitive performance metrics pre- and post-restoration

89
9.3. Discussion
9.3.1. Impact of lighting interventions on stress responses
The present study provides compelling evidence that lighting conditions, particularly those at a color
temperature of 3000K, significantly reduce physiological arousal and contribute to stress reduction. These
findings are consistent with existing literature, which suggests that warmer lighting can have a calming
effect on individuals. The study by Kuijsters et al. [350] demonstrated that an ambiance of 125 lux with
2700K lighting creates a more pleasant and low-arousing atmosphere compared to a brighter setting of 325
lux with 4000K cool lighting. This calmer ambiance was effective in soothing anxiety, as indicated by
subjective feedback and physiological markers such as decreased heart rate and electrodermal activity.
The profound impact of lighting at a color temperature of 3000 K on physiological arousal and stress
reduction may be deeply rooted in the evolutionary adaptation of human physiology to the Earth's natural
light-dark cycle. The warm, red-orange hues predominant at twilight, normally signaling the transition from
daytime vigilance to nocturnal rest, are closely emulated by a 3000K color temperature. Such environmental
cues are integral to the suprachiasmatic nucleus (SCN) in the brain, which orchestrates a neuroendocrine
response preparing the body for sleep—markedly through the secretion of melatonin, known for its sleepinducing and stress-reducing properties [351].
The effectiveness of a lighting setup combining a 3000K color temperature with 100 lux illumination is
particularly notable. This setup markedly decreased both heightened EDA and heart rate, as well as the
perceived stress level, as assessed by the STAI-6. Although there was not a significant drop in the low
frequency/high frequency (LF/HF) ratio, it still reduced from an average of 1.93 to 1.57 after the restoration
period. This particular lighting does not cause eye strain or excessive brain stimulation, thus fostering a
relaxing environment akin to the natural dimming of light at dusk. It mirrors the body's diurnal rhythms,
which naturally pivot towards a state of rest as daylight wanes. As cortisol levels drop and melatonin
production ramps up in the evening, the body gears itself towards a reparative state. By mimicking these
natural environmental transitions during the restoration phase, 3000K lighting at a 100 lux provides a signal
that resonates with the body's internal clock, encouraging a seamless progression towards relaxation and
recuperation [352].
To that end, lighting conditions at 3000 K are likely to engage the intrinsically photosensitive retinal
ganglion cells (ipRGCs), which are crucial for regulating circadian rhythms and managing physiological
responses related to the autonomic nervous system (ANS) [353]. The ANS is dichotomous, comprising both
the sympathetic and parasympathetic nervous systems. The sympathetic division is typically associated
with the 'fight or flight' response, preparing the body for acute stress or threats. In contrast, the
parasympathetic division is often referred to as the 'rest and digest' system, promoting conservation of
energy and relaxation. Heart rate is modulated by these systems, with increased sympathetic activity leading
to elevated heart rates and parasympathetic activity slowing it down [291]. Hence, a decrease in heart rate
following exposure to 3000 K lighting at 100 lux is indicative of enhanced parasympathetic activity,
underscoring a state of relaxation and recovery from stress. A similar rationale can be applied to the LF/HF
ratio, a quantifiable measure of autonomic balance. A lower ratio post-exposure to 3000 K lighting at 1000
lux suggests a tilt towards parasympathetic dominance, corroborating the notion that such lighting
conditions can facilitate a physiological environment favorable to stress recovery.
9.3.2. Impact of lighting interventions on cognitive performance
The observed enhancement in cognitive performance under 7000K lighting in this study is a finding that is
increasingly supported by the body of scientific research. The implication that higher color temperatures
exert a stimulating effect on the brain is not unprecedented. For instance, the study conducted by Grant et

90
al. [354] shows that blue-enriched white light at 5000K improves processing speed, working memory, and
procedural learning among young adults, in comparison to a 3000K CCT lighting. The research by Lehrl et
al. [355] reinforces this perspective, suggesting that light with a greater component of blue wavelengths
aids in mitigating sleepiness and boosting performance. These insights dovetail with the current study's
demonstration that a color temperature of 7000K can facilitate cognitive functions post-restoration.
From a physiological perspective, the impact of 7000K CCT on cognitive performance can be attributed to
the fundamental relationship between light exposure and circadian rhythms. The circadian system, serving
as the body's internal clock, is acutely responsive to the blue light spectrum. This spectrum is predominant
in natural daylight and is similarly present in high CCT light sources. Studies have demonstrated that
exposure to this light spectrum effectively suppresses melatonin production, a hormone responsible for
inducing sleepiness. Consequently, this suppression enhances alertness and cognitive performance. This
concept was further explored in the research conducted by Chellappa et al. [356], which provided insights
into the specific effects of 6500K light. Their findings revealed that exposure to 6500K light results in a
more pronounced suppression of melatonin, alongside increased subjective alertness and improved
cognitive performance. Notably, 6500K light exposure was associated with significantly quicker reaction
times in tasks requiring sustained attention. This enhancement in cognitive abilities was closely linked to
decreased levels of salivary melatonin, particularly under the 6500K lighting condition.
Delving deeper into the biological underpinnings, it is known that light receptors in the eye, particularly the
intrinsically photosensitive retinal ganglion cells (ipRGCs), are directly responsive to blue light. These cells
project to various brain regions that regulate alertness, cognitive functions, and the sleep-wake cycle [353].
The activation of these pathways by light that simulates a high CCT such as 7000K is thought to lead to
increased cognitive arousal and improved performance on tasks requiring sustained attention.
In an evolutionary context, humans are diurnal beings, naturally more alert and cognitively capable during
daylight hours. The high CCT light's mimicry of natural daylight could, therefore, be invoking an innate
physiological response that primes the brain for heightened cognitive activity. This response could be an
adaptive advantage, enhancing the ability to perform complex tasks during optimal daylight conditions
when our ancestors were most active.
The interplay between lighting conditions and cognitive performance also finds theoretical support in
Kaplan's attention restoration theory (ART) [357], which suggests that exposure to elements reminiscent of
natural environments can replenish depleted cognitive resources. The high CCT lighting may emulate the
revitalizing characteristics of natural daylight [358], thereby aligning with ART. This environmental
mimicry supports restorative processes, potentially aiding in recovery from cognitive fatigue and
contributing to improved performance on tasks that measure attention and memory, such as the CPT and
VBDS.
9.3.3. Practical applications in office environments
The findings from this research have practical implications for enhancing the health and productivity of
office workers through the application of smart office space design. The integration of adaptive lighting
systems is key to creating environments that support productivity and employee well-being. These systems
go beyond simple automation; they aim to establish a dynamic interaction between the workspace and its
occupants.
For instance, to optimize worker well-being, physiological sensing technology plays a critical role. This
system monitors the stress levels of employees in real time by assessing physiological arousal. Upon
detecting signs of stress, the office's automated lighting system could respond by adjusting the environment

91
to a warmer color temperature with reduced illuminance, thereby facilitating the stress recovery process
and contributing to a calm atmosphere.
Concurrently, the office environment could leverage machine learning algorithms to analyze workers'
interactions with their computers, utilizing facial recognition and posture analysis to gauge the nature of
activities and levels of mental fatigue. Should the system identify indicators of diminished concentration or
escalating cognitive fatigue, it could proactively alter the lighting to a cooler temperature, which is
conducive to reinvigorating cognitive performance and sustaining attention. This intelligent office
infrastructure thus serves to maintain an environment that is responsive to the immediate needs of its
occupants, ensuring both health and productivity are addressed with precision.
Previous research has underscored the significance of biophilic elements in facilitating stress recovery.
Many of these studies concentrated on the design of breakout rooms incorporating biophilic design
principles, requiring workers to leave their workstations to relax in a designated space. Our study, however,
reveals that the color and intensity of lighting at the workspace itself play a pivotal role in reducing stress
in a non-intrusive manner. Our findings indicate that a lighting intervention with a color temperature of
3000 K and an illuminance level of 100 lux at employees' desks can markedly improve stress recovery,
eliminating the need for separate breakout rooms. This method offers a simple, non-intrusive way to
enhance well-being and stress recovery in the workplace.
9.3.4. Limitations and future research directions
Considering the results of this study, a range of future research avenues emerges. Firstly, this study's scope
was limited in capturing diverse personal characteristics. Therefore, it is essential to broaden future research
to include a more diverse range of demographic variables, such as age, gender, and personality traits. This
expansion will enable a more thorough understanding of the varied responses to lighting conditions among
different groups. Secondly, the study's restoration period was limited to only 10 minutes. Exploring the
minimum duration required for the restoration period to fully return physiological and cognitive functions
to their original state would be a compelling area of research. Additionally, applying adaptive lighting
systems in varied settings beyond the typical office environment is pivotal. This expansion into spaces like
educational institutions, healthcare facilities, and residential areas could deepen our understanding of the
potential benefits these systems offer for enhancing cognitive performance and overall well-being.
Furthermore, an integrated approach that considers the interplay between lighting and other environmental
factors, such as ambient noise, temperature, and air quality, is necessary. This holistic perspective could
provide a more complete picture of how environmental elements collectively impact cognitive and
psychological health. Finally, implementing interventions in naturalistic environments through longitudinal
studies represents a vital future research direction. This approach would allow for the assessment of the
effectiveness, reliability, and sustainability of adaptive lighting systems over time. This methodological
shift towards naturalistic, longitudinal studies would not only validate the findings from controlled
experiments but also provide a comprehensive understanding of the practical applications and long-term
benefits of lighting interventions in everyday environments.
9.4. Conclusions
This study investigated how different lighting conditions can help with reducing stress arousal and restoring
cognitive functions after stressful and mental draining office tasks. The major findings reveal that certain
lighting conditions significantly influence cognitive performance and stress levels. Specifically, lighting at
7000K color temperature notably enhanced cognitive functions post-restoration, likely due to its stimulating
effect on the brain, akin to natural daylight. This was evidenced by improved scores in the Continuous
Performance Test (CPT) and the Visual Backward Digit Span (VBDS), indicating heightened attention and
memory capabilities. Conversely, lighting at a 3000K color temperature, particularly at 100 lux,

92
demonstrated a substantial reduction in physiological markers of stress, suggesting a calming effect that
aligns with the relaxation and readiness for sleep associated with warmer light.
These findings hold significant implications for office environments and broader settings where cognitive
performance and stress management are crucial. They underscore the potential of adaptive lighting systems
to enhance well-being, health, and productivity in professional settings. By using the appropriate lighting
interventions highlighted in our study, it is possible to create environments that not only foster better
cognitive performance but also promote a more conducive atmosphere for stress recovery. This research
contributes to a growing body of evidence supporting the strategic use of environmental elements to
improve psychological and physiological outcomes in workplace settings. Future research directions
include longitudinal studies, expanding demographic variables, exploring varied environments, and
integrating lighting with other environmental factors, to further understand and optimize these benefits.

93
Chapter 10: Conclusion
This dissertation establishes the groundwork for developing intelligent workspaces attuned to human
health, capable of adapting indoor environmental conditions to alleviate adverse health effects. It employs
a diverse array of research methodologies, such as experiments involving human subjects, machine learning
frameworks, questionnaires, and utilizes this knowledge to guide the design of healthy buildings.
Chapters 4 and 5 derive from a questionnaire that was crafted and distributed over a period of 45 days
during the COVID-19 pandemic, securing valid responses from 988 participants. The results from Chapter
4 reveal that overall worker productivity remained unchanged compared to pre-pandemic in-office levels,
with increases in productivity reported by female, older, and high-income workers. Factors such as better
mental and physical health, having a teenager, improved communication with coworkers, and a dedicated
workspace were associated with higher productivity, while the average time spent at a workstation increased
by about 1.5 hours during a typical work-from-home (WFH) day. Chapter 5's findings indicate that low
satisfaction with natural lighting, glare, and humidity was linked to eye-related symptoms, while
dissatisfaction with noise strongly correlated with fatigue, headaches, and mental health issues.
Additionally, low satisfaction with certain indoor environmental quality (IEQ) factors like air quality and
noise was associated with increased mental stress and difficulty concentrating. Workers with higher
incomes reported greater satisfaction with humidity, air quality, and indoor temperature, and older
individuals were more satisfied with IEQ factors overall. Together, the findings from both chapters can
inform future design practices focused on hybrid home-work environments by highlighting the impact of
IEQ factors on occupant well-being.
Through a controlled experimental procedure involving 48 participants, physiological and behavioral data
were collected to establish automated frameworks for stress appraisal assessment and productivity
monitoring. The results from Chapter 6 show that moderate stress levels, as per the Yerkes-Dodson law,
are linked to increased productivity and positive mood, while both low and high stress levels are associated
with decreased productivity and negative mood, especially when distress overshadows eustress. An
XGBOOST model demonstrated the highest prediction accuracy for stress appraisal, highlighting the
significance of physiological data, particularly electrodermal activity, skin temperature, and blood volume
pulse. Chapter 7 advances the research by employing a machine learning framework to predict office
workers' perceived productivity, integrating physiological, behavioral, and psychological features. The
extended model, incorporating psychological states, outperformed the baseline physiological and
behavioral model, with mood and eustress identified as critical productivity predictors. Wearable devices
showed superior performance over workstation addons in productivity prediction. Collectively, these
findings advocate for the integration of the proposed models within smart workstations, enabling adaptable
environments that enhance health, amplify eustress, and boost productivity and overall well-being among
office workers.
In Chapter 8, we invited 39 participants to explore the impact of white noise on cognitive performance,
creativity, and stress levels among neurotypical young adults within a private office setting. The study
compared two white noise conditions—45 dB and 65 dB—against a baseline of ambient office noise.
Results indicated that white noise at 45 dB significantly improved cognitive performance in terms of
sustained attention, accuracy, and speed, in addition to enhancing creativity and reducing stress levels.
Conversely, the 65 dB white noise condition was found to improve working memory but also increased
stress levels. This suggests that different tasks might benefit from specific noise levels to optimize
performance. The findings underscore the potential of integrating white noise into office environments as

94
a means to boost workers’ performance across various dimensions. For Chapter 9, the study engaged 100
participants to examine how different lighting conditions affect stress reduction and the restoration of
cognitive functions following mentally taxing office tasks. This investigation revealed that certain lighting
conditions have a marked influence on cognitive performance and stress. Specifically, lighting with a
7000K color temperature significantly improved cognitive functions, likely due to its stimulating effects
similar to natural daylight, as demonstrated by enhanced scores on the Continuous Performance Test (CPT)
and the Visual Backward Digit Span (VBDS). In contrast, a 3000K color temperature at 100 lux was shown
to significantly lower physiological stress markers, suggesting its calming effects conducive to relaxation
and sleep readiness. These outcomes highlight the nuanced effects of lighting conditions on both cognitive
and emotional well-being in work environments. Together, the results of these two studies provide critical
insights into designing healthier and more productive indoor work environments. By carefully considering
the application of white noise and specific lighting conditions, it is possible to create workspaces that not
only promote productivity and cognitive functioning but also contribute to stress reduction. These findings
offer a valuable framework for optimizing office design to support the well-being and efficiency of workers.

95
References
[1] N. E. KLEPEIS et al., “The National Human Activity Pattern Survey (NHAPS): a resource for
assessing exposure to environmental pollutants,” J. Expo. Sci. Environ. Epidemiol., vol. 11, no. 3,
pp. 231–252, Jul. 2001, doi: 10.1038/sj.jea.7500165.
[2] U.S. Bureau of Labor Statistics, “Occupational Employment and Wages, May 2020,” 2021.
https://www.bls.gov/oes/current/oes430000.htm#nat.
[3] A. Kidd, “The marks are on the knowledge worker.,” in Proceedings of the SIGCHI conference on
Human factors in computing systems, 1994, pp. 186–191.
[4] H. Jahncke and D. M. Hallman, “Objective measures of cognitive performance in activity based
workplaces and traditional office types.,” J. Environ. Psychol., vol. 72, 2020.
[5] G. G. Fisher, M. Chacon, and D. S. Chaffee, “Theories of cognitive aging and work.,” Work
across Lifesp., 2019.
[6] M. D. Lezak, D. B. Howieson, D. W. Loring, and J. S. Fischer, Neuropsychological assessment.
USA: Oxford University Press, 2004.
[7] N. Hähn, E. Essah, and T. Blanusa, “Biophilic design and office planting: a case study of effects
on perceived health, well-being and performance metrics in the workplace.,” Intell. Build. Int., vol.
13, no. 4, pp. 241–260, 2021.
[8] W. H. Ko et al., “The impact of a view from a window on thermal comfort, emotion, and cognitive
performance.,” Build. Environ., vol. 175, 2020.
[9] S. H. Kang, J. Lee, and S. Jin, “Effect of standing desk use on cognitive performance and physical
workload while engaged with high cognitive demand tasks.,” Appl. Ergon., vol. 92, 2021.
[10] A. Jamrozik et al., “Access to daylight and view in an office improves cognitive performance and
satisfaction and reduces eyestrain: A controlled crossover study.,” Build. Environ., vol. 165, 2019.
[11] Y. Al Horr, M. Arif, A. Kaushik, A. Mazroei, M. Katafygiotou, and E. Elsarrag, “Occupant
productivity and office indoor environment quality: A review of the literature,” Build. Environ.,
vol. 105, pp. 369–389, Aug. 2016, doi: 10.1016/j.buildenv.2016.06.001.
[12] M. Awada et al., “Occupant health in buildings: Impact of the COVID-19 pandemic on the
opinions of building professionals and implications on research,” Build. Environ., vol. 207, p.
108440, Jan. 2022, doi: 10.1016/j.buildenv.2021.108440.
[13] M. J. Mendell et al., “Improving the Health of Workers in Indoor Environments: Priority Research
Needs for a National Occupational Research Agenda,” Am. J. Public Health, vol. 92, no. 9, pp.
1430–1440, Sep. 2002, doi: 10.2105/AJPH.92.9.1430.
[14] J. Messenger, “Working time and the future of work.,” ILO Futur. Work Res. Pap. Ser., 2018.
[15] D. Cook, “The Global Remote Work Revolution and the Future of Work.,” in The Business of
Pandemics: The COVID-19 Story, 2020.
[16] N. Salari et al., “Prevalence of stress, anxiety, depression among the general population during the
COVID-19 pandemic: a systematic review and meta-analysis,” Global. Health, vol. 16, no. 1, p.
57, Dec. 2020, doi: 10.1186/s12992-020-00589-w.
[17] G. Fink, “Stress: Concepts, Definition and History☆,” in Reference Module in Neuroscience and
Biobehavioral Psychology, Elsevier, 2017.

96
[18] American Psychological Association, “Stress in America: The state of our nation. Stress in
AmericaTM Survey,” 2017. [Online]. Available:
https://www.apa.org/news/press/releases/stress/2017/state-nation.pdf.
[19] A. M. Shahsavarani, E. Azad Marz Abadi, and M. Hakimi Kalkhoran, “Stress: Facts and theories
through literature review.,” Int. J. Med. Rev., 2015.
[20] H. Selye, “Confusion and controversy in the stress field.,” J. Human Stress, pp. 37–44, 1975.
[21] C. S. Kung and C. K. Chan, “Differential roles of positive and negative perfectionism in predicting
occupational eustress and distress.,” Pers. Individ. Dif., 2014.
[22] J. McGowan, D. Gardner, and R. Fletcher, “Positive and negative affective outcomes of
occupational stress.,” NZ. J. Psychol., 2006.
[23] H. Selye, “Forty years of stress research: principal remaining problems and misconceptions.,”
Can. Med. Assoc. J., 1976.
[24] R. Kupriyanov and R. Zhdanov, “The eustress concept: problems and outlooks.,” World J. Med.
Sci., vol. 11, no. 2, pp. 179–185, 2014, doi: http://dx.doi.org/10.5829/idosi.wjms.2014.11.2.8433.
[25] J. Kallio et al., “Assessment of perceived indoor environmental quality, stress and productivity
based on environmental sensor data and personality categorization,” Build. Environ., vol. 175, p.
106787, May 2020, doi: 10.1016/j.buildenv.2020.106787.
[26] T. Q. Thach et al., “Associations of perceived indoor environmental quality with stress in the
workplace.,” Indoor Air, vol. 30, no. 6, pp. 1166–1177, 2020.
[27] X. Zhang, Z. Lian, and Y. Wu, “Human physiological responses to wooden indoor environment,”
Physiol. Behav., vol. 174, pp. 27–34, 2017, doi: 10.1016/J.PHYSBEH.2017.02.043.
[28] A. Hassan et al., “Do plants affect brainwaves? Effect of indoor plants in work environment on
mental stress,” Eur. J. Hortic. Sci., vol. 85, no. 4, pp. 279–283, 2020, doi:
10.17660/eJHS.2020/85.4.9.
[29] M. Rashid and C. Zimring, “A Review of the Empirical Literature on the Relationships Between
Indoor Environment and Stress in Health Care and Office Settings,” Environ. Behav., vol. 40, no.
2, pp. 151–190, Mar. 2008, doi: 10.1177/0013916507311550.
[30] K. Gillis and B. Gatersleben, “A Review of Psychological Literature on the Health and Wellbeing
Benefits of Biophilic Design,” Buildings, vol. 5, no. 3, pp. 948–963, Aug. 2015, doi:
10.3390/buildings5030948.
[31] J. Kim and R. De Dear, “Workspace satisfaction: The privacy-communication trade-off in openplan offices.,” J. Environ. Psychol., vol. 36, pp. 18–26, 2013.
[32] V. Hongisto, J. Varjo, H. Leppämäki, D. Oliva, and J. Hyönä, “Work performance in private office
rooms: The effects of sound insulation and sound masking.,” Build. Environ., 2016.
[33] N. E. Klepeis et al., “The National Human Activity Pattern Survey (NHAPS): a resource for
assessing exposure to environmental pollutants.,” J. Expo. Sci. Environ. Epidemiol., pp. 231–252,
2001.
[34] C. Sun, Z. Lian, and L. Lan, “Work performance in relation to lighting environment in office
buildings,” Indoor Built Environ., vol. 28, no. 8, 2019, doi: 10.1177/1420326X18820089.
[35] B. Manav, “An experimental study on the appraisal of the visual environment at offices in relation

97
to colour temperature and illuminance,” Build. Environ., vol. 42, no. 2, pp. 979–983, 2007, doi:
10.1016/J.BUILDENV.2005.10.022.
[36] P. Leather, D. Beale, and L. Sullivan, “Noise, psychosocial stress and their interaction in the
workplace.,” J. Environ. Psychol., vol. 23, no. 2, pp. 213–222, 2003.
[37] Z. Lin and S. Deng, “A questionnaire survey on sleeping thermal environment and bedroom air
conditioning in high-rise residences in Hong Kong.,” Energy Build., pp. 1302–1307, 2006.
[38] S.-K. Wong, L. Wai-Chung Lai, D. C.-W. Ho, K.-W. Chau, C. Lo-Kuen Lam, and C. Hung-Fai
Ng, “Sick building syndrome and perceived indoor environmental quality: A survey of apartment
buildings in Hong Kong,” Habitat Int., vol. 33, no. 4, pp. 463–471, Oct. 2009, doi:
10.1016/j.habitatint.2009.03.001.
[39] M.-F. Marin et al., “Chronic stress, cognitive functioning and mental health,” Neurobiol. Learn.
Mem., vol. 96, no. 4, pp. 583–595, Nov. 2011, doi: 10.1016/j.nlm.2011.02.016.
[40] D. C. Ganster and J. Schaubroeck, “Work Stress and Employee Health,” J. Manage., vol. 17, no.
2, pp. 235–271, Jun. 1991, doi: 10.1177/014920639101700202.
[41] B. Brunner, I. Igic, A. C. Keller, and S. Wieser, “Who gains the most from improving working
conditions? Health-related absenteeism and presenteeism due to stress at work,” Eur. J. Heal.
Econ., vol. 20, no. 8, pp. 1165–1180, Nov. 2019, doi: 10.1007/s10198-019-01084-9.
[42] I. Rodríguez, M. W. Kozusznik, and J. M. Peiró, “Development and validation of the Valencia
Eustress-Distress Appraisal Scale.,” Int. J. Stress Manag., vol. 20, no. 4, pp. 279–308, Nov. 2013,
doi: 10.1037/a0034330.
[43] D. Watson and J. W. Pennebaker, “Health complaints, stress, and distress: exploring the central
role of negative affectivity.,” Psychol. Rev., vol. 96, no. 2, p. 234, 1989.
[44] G. Brulé and R. Morgan, “Working with stress: Can we turn distress into eustress.,” J.
Neuropsychol. Stress Manag., vol. 3, pp. 1–3, 2018.
[45] R. S. Ulrich, R. F. Simons, B. D. Losito, E. Fiorito, M. A. Miles, and M. Zelson, “Stress recovery
during exposure to natural and urban environments.,” J. Environ. Psychol., vol. 11, no. 3, pp. 201–
230, 1991.
[46] G. L. Engel, “The Clinical Application of the Biopsychosocial Model,” J. Med. Philos., vol. 6, no.
2, pp. 101–124, Jan. 1981, doi: 10.1093/jmp/6.2.101.
[47] Y. Lee, E. C. Nelson, M. J. Flynn, and J. S. Jackman, “Exploring soundscaping options for the
cognitive environment in an open-plan office.,” Build. Acoust., 2020.
[48] G. W. Evans and D. Johnson, “Stress and open-office noise.,” J. Appl. Psychol., 2000.
[49] L. Lan, S. Hadji, L. Xia, and Z. Lian, “The effects of light illuminance and correlated color
temperature on mood and creativity,” Build. Simul., vol. 14, no. 3, pp. 463–475, Jun. 2021, doi:
10.1007/s12273-020-0652-z.
[50] I. Knez and C. Kers, “Effects of Indoor Lighting, Gender, and Age on Mood and Cognitive
Performance,” Environ. Behav., vol. 32, no. 6, pp. 817–831, Nov. 2000, doi:
10.1177/0013916500326005.
[51] World Health Organization, “The constitution of the world health organization.,” 1947.
[52] World Health Organization., “Management of physical health conditions in adults with severe

98
mental disorders: WHO guidelines.,” 2018.
[53] World Health Organization., “Promoting mental health: Concepts, emerging evidence, practice:
Summary report.,” 2004.
[54] L. J. Waite, “Social well-being and health in the older population: moving beyond social
relationships.,” in Future Directions for the Demography of Aging: Proceedings of a Workshop.,
National Academies Press, 2018.
[55] World Health Organization, “Indoor environment: health aspects of air quality, thermal
environment, light and noise,” 1990.
[56] J. M. Samet and J. D. Spengler, “Indoor environments and health: moving into the 21st century.,”
Am. J. Public Health, pp. 1489–1493, 2003.
[57] E. S. Bernstein and S. Turban, “The impact of the ‘open’workspace on human collaboration.,”
Philos. Trans. R. Soc. B Biol. Sci., 2018.
[58] E. Oldham and H. Kim, “IEQ Field Investigation in High-Performance, Urban Elementary
Schools,” Atmosphere (Basel)., 2020.
[59] D. H. Kim and P. M. Bluyssen, “Clustering of office workers from the OFFICAIR study in The
Netherlands based on their self-reported health and comfort.,” Build. Environ., 2020.
[60] R. K. Raanaas, K. H. Evensen, D. Rich, G. Sjøstrøm, and G. Patil, “Benefits of indoor plants on
attention capacity in an office setting.,” J. Environ. Psychol., pp. 99–105, 2011.
[61] M. Arif, M. Katafygiotou, A. Mazroei, A. Kaushik, and E. Elsarrag, “Impact of indoor
environmental quality on occupant well-being and comfort: A review of the literature.,” Int. J.
Sustain. Built Environ., pp. 1–11, 2016.
[62] L. Rohde, T. S. Larsen, R. L. Jensen, and O. K. Larsen, “Framing holistic indoor environment:
Definitions of comfort, health and well-being.,” Indoor Built Environ., pp. 1118–1136, 2020.
[63] S. K. Wong, L. W. C. Lai, D. C. W. Ho, K. W. Chau, C. L. K. Lam, and C. H. F. Ng, “Sick
building syndrome and perceived indoor environmental quality: A survey of apartment buildings
in Hong Kong.,” Habitat Int., pp. 463–471, 2009.
[64] H. Belachew et al., “Sick building syndrome and associated risk factors among the population of
Gondar town, northwest Ethiopia.,” Environ. Health Prev. Med., pp. 1–9, 2018.
[65] P. MacNaughton, J. Spengler, J. Vallarino, S. Santanam, U. Satish, and J. Allen, “Environmental
perceptions and health before and after relocation to a green building.,” Build. Environ., pp. 138–
144, 2016.
[66] S. Hammond, “Using psychometric tests.,” in Research methods in psychology, 2006, pp. 182–
209.
[67] J. Pekkanen, A. Hyvärinen, U. Haverinen-Shaughnessy, M. Korppi, T. Putus, and A. Nevalainen,
“Moisture damage and childhood asthma: a population-based incident case–control study.,” Eur.
Respir. J., pp. 509–515, 2007.
[68] E. Simons, S. A. Hwang, E. F. Fitzgerald, C. Kielb, and S. Lin, “The impact of school building
conditions on student absenteeism in upstate New York.,” Am. J. Public Health, pp. 1679–1686,
2010.
[69] L. Sun and L. A. Wallace, “Residential cooking and use of kitchen ventilation: the impact on

99
exposure.,” J. Air Waste Manag. Assoc., 2020.
[70] M. S. A. Rodrigues, R. D. V. Leite, C. M. Lelis, and T. C. Chaves, “Differences in ergonomic and
workstation factors between computer office workers with and without reported musculoskeletal
pain.,” Work, pp. 563–572, 2017.
[71] K. Savelieva, T. Marttila, J. Lampi, S. Ung-Lanki, M. Elovainio, and J. Pekkanen, “Associations
between indoor environmental quality in schools and symptom reporting in pupil-administered
questionnaires.,” Environ. Heal., pp. 1–12, 2019.
[72] D. Lukcso, T. L. Guidotti, D. E. Franklin, and A. Burt, “Indoor environmental and air quality
characteristics, building-related health symptoms, and worker productivity in a federal government
building complex.,” Arch. Environ. Occup. Health, pp. 85–101, 2016.
[73] M. Donn, E. Braasch, M. Woodbury, E. Novak, and A. Banks, “Design Research: Optimising
Row-House Orientation.,” 2015.
[74] M. A. Mujeebu, “Introductory chapter: Indoor environmental quality.,” in Indoor Environmental
Quality., 2019.
[75] S. S. Kim, D. H. Kang, D. H. Choi, M. S. Yeo, and K. W. Kim, “Comparison of strategies to
improve indoor air quality at the pre-occupancy stage in new apartment buildings.,” Build.
Environ., pp. 320–328, 2008.
[76] J. Pašek and V. Sojková, “Facility management of smart buildings.,” Int. Rev. Appl. Sci. Eng., pp.
181–187, 2018.
[77] L. Carmichael et al., “Healthy buildings for a healthy city: Is the public health evidence base
informing current building policies?,” Sci. Total Environ., 2020.
[78] M. Santana and M. J. Cobo-Martín, “What is the future of work? A science mapping analysis.,”
Eur. Manag. J., 2020.
[79] B. A. Livingston, “Bargaining behind the scenes: Spousal negotiation, labor, and work–family
burnout.,” J. Manage., pp. 949–977, 2014.
[80] Z. Feng and K. Savani, “Covid-19 created a gender gap in perceived work productivity and job
satisfaction: implications for dual-career parents working from home.,” Gend. Manag. An Int. J.,
2020.
[81] K. M. Shockley, W. Shen, M. M. DeNunzio, M. L. Arvan, and E. A. Knudsen, “Disentangling the
relationship between gender and work–family conflict: An integration of theoretical perspectives
using meta-analytic methods.,” J. Appl. Psychol., 2017.
[82] W. Bönte and S. Krabel, “You can’t always get what you want: Gender differences in job
satisfaction of university graduates.,” Appl. Econ., pp. 2477–2487, 2014.
[83] J. Feyrer, “Aggregate evidence on the link between age structure and productivity.,” Popul. Dev.
Rev., pp. 78–99, 2008.
[84] L. Roosaar, J. Masso, and U. Varblane, “Age-related productivity decrease in high-waged and
low-waged employees.,” Int. J. Manpow., 2019.
[85] A. D. A. M. Gorlick, “The productivity pitfalls of working from home in the age of COVID-19,”
Stanford News, 2020.
[86] N. Bloom, J. Liang, J. Roberts, and Z. J. Ying, “Does working from home work? Evidence from a

100
Chinese experiment.,” Q. J. Econ., pp. 165–218, 2015.
[87] P. Choudhury, C. Foroughi, and B. Larson, “Work from Anywhere Or Co-locate?: Autonomy
Versus Learning Effects at the United States Patent Office,” 2018.
[88] G. Bryant, “Maintaining Young Lawyer Well-Being in a Pandemic,” 2020.
[89] D. Lerner et al., “Relationship of employee-reported work limitations to work productivity.,” Med.
Care, pp. 649–659, 2003.
[90] R. Z. Goetzel, S. R. Long, R. J. Ozminkowski, K. Hawkins, S. Wang, and W. Lynch, “Health,
absence, disability, and presenteeism cost estimates of certain physical and mental health
conditions affecting US employers.,” J. Occup. Environ. Med., pp. 398–412, 2004.
[91] K. Fassoulis and N. Alexopoulos, “The workplace as a factor of job satisfaction and
productivity.,” J. Facil. Manag., 2015.
[92] A. Soriano, M. W. Kozusznik, J. M. Peiró, and E. Demerouti, “Employees’ Work Patterns–Office
Type Fit and the Dynamic Relationship Between Flow and Performance.,” Appl. Psychol., 2020.
[93] A. Moretti, F. Menna, M. Aulicino, M. Paoletta, S. Liguori, and G. Iolascon, “Characterization of
Home Working Population during COVID-19 Emergency: A Cross-Sectional Analysis.,” Int. J.
Environ. Res. Public Health, 2020.
[94] S. Lopez-Leon, D. A. Forero, and P. Ruiz-Díaz, “Recommendations for working from home
during the pandemic (and Beyond),” Work, pp. 1–5, 2020.
[95] C. Suart, T. N. Suart, K. Graham, and R. Truant, “When the Labs Closed: Graduate Students’ and
Postdoctoral Fellows’ Experiences of Disrupted Research During the COVID-19 Pandemic,”
2020.
[96] J. Kim, C. Candido, L. Thomas, and R. de Dear, “Desk ownership in the workplace: The effect of
non-territorial working on employee workplace satisfaction, perceived productivity and health.,”
Build. Environ., pp. 203–214, 2016.
[97] A. Rudnicka, J. W. Newbold, D. Cook, M. E. Cecchinato, and S. JJ, “Eworklife: developing
effective strategies for remote working during the COVID-19 pandemic. In Eworklife: developing
effective strategies for remote working during the COVID-19 pandemic.,” New Futur. Work
Online Symp., 2020.
[98] M. Awada and I. Srour, “A genetic algorithm based framework to model the relationship between
building renovation decisions and occupants’ satisfaction with indoor environmental quality.,”
Build. Environ., pp. 247–257, 2018.
[99] S. Kang, D. Ou, and C. M. Mak, “The impact of indoor environmental quality on work
productivity in university open-plan research offices.,” Build. Environ., 2017.
[100] J. Kim and R. de Dear, “Nonlinear relationships between individual IEQ factors and overall
workspace satisfaction.,” Build. Environ., pp. 33–40, 2012.
[101] C. F. Chen et al., “The impacts of building characteristics, social psychological and cultural
factors on indoor environment quality productivity belief.,” Build. Environ., 2020.
[102] L. Heschong and D. Mahone, “Windows and offices: A study of office worker performance and
the indoor environment.,” 2003.
[103] A. Nawaz, A. Javed, and U. M. Raja, “Impact of workspace design on employee’s productivity: a

101
case study of public sector universities in Hazara division.,” Int. J. Sustain. Real Estate Constr.
Econ., pp. 201–214, 2019.
[104] L. Pastore and M. Andersen, “Building energy certification versus user satisfaction with the indoor
environment: Findings from a multi-site post-occupancy evaluation (POE) in Switzerland.,” Build.
Environ., pp. 60–74, 2019.
[105] E. M. Hunter, M. A. Clark, and D. S. Carlson, “Violating work-family boundaries: Reactions to
interruptions at work and home.,” J. Manage., pp. 1284–1308, 2019.
[106] M. A. Shaffer, J. R. Joplin, and Y. S. Hsu, “Expanding the boundaries of work—family research:
A review and agenda for future research.,” Int. J. Cross Cult. Manag., pp. 221–268, 2011.
[107] E. DeFilippis, S. M. Impink, M. Singell, J. T. Polzer, and R. Sadun, “Collaborating during
coronavirus: The impact of COVID-19 on the nature of work,” 2020.
[108] T. M. Alon, M. Doepke, J. Olmstead-Rumsey, and M. Tertilt, “The impact of COVID-19 on
gender equality,” 2020.
[109] J. Singh et al., “Sales profession and professionals in the age of digitization and artificial
intelligence technologies: concepts, priorities, and questions.,” J. Pers. Sell. Sales Manag., pp. 2–
22, 2019.
[110] M. C. Peasley, B. Hochstein, B. P. Britton, R. V. Srivastava, and G. T. Stewart, “Can’t Leave it at
Home? The Effects of Personal Stress on Burnout and Salesperson Performance.,” J. Bus. Res.,
2020.
[111] A. Balkeran, “Hustle Culture and the Implications for Our Workforce,” 2020.
[112] B. Obrenovic, J. Du, D. Godinic, D. Tsoy, M. A. S. Khan, and I. Jakhongirov, “Sustaining
enterprise operations and productivity during the COVID-19 pandemic:‘Enterprise Effectiveness
and Sustainability Model,’” Sustainability, 2020.
[113] . The National Institute for Occupational Safety and Health, “INDOOR ENVIRONMENTAL
QUALITY,” 2013. .
[114] T. Witterseh, D. P. Wyon, and G. Clausen, “The effects of moderate heat stress and open-plan
office noise distraction on SBS symptoms and on the performance of office work.,” Indoor Air,
2004.
[115] G. W. Evans, Environmental stress. Cambridge University Press., 1984.
[116] M. Spreng, “Possible health effects of noise induced cortisol increase.,” Noise Heal., 2000.
[117] S. Lamb and K. C. Kwok, “A longitudinal investigation of work environment stressors on the
performance and wellbeing of office workers.,” Appl. Ergon., pp. 104–111, 2016.
[118] H. Jahncke, S. Hygge, N. Halin, A. M. Green, and K. Dimberg, “Open-plan office noise:
Cognitive performance and restoration,” J. Environ. Psychol., vol. 31, no. 4, pp. 373–382, Dec.
2011, doi: 10.1016/j.jenvp.2011.07.002.
[119] M. Boubekri, I. N. Cheung, K. J. Reid, C. H. Wang, and P. C. Zee, “Impact of windows and
daylight exposure on overall health and sleep quality of office workers: a case-control pilot
study.,” J. Clin. sleep Med., 2014.
[120] J. A. Sanchez, T. Ikaga, and S. V. Sanchez, “Quantitative improvement in workplace performance
through biophilic design: A pilot experiment case study.,” Energy Build., 2018.

102
[121] A. Jamrozik et al., “Access to daylight and view in an office improves cognitive performance and
satisfaction and reduces eyestrain: A controlled crossover study.,” Build. Environ., 2019.
[122] I. Acosta, M. Á. Campano, R. Leslie, and L. Radetsky, “Daylighting design for healthy
environments: Analysis of educational spaces for optimal circadian stimulus.,” Sol. Energy, pp.
584–596, 2019.
[123] L. Tähkämö, T. Partonen, and A. K. Pesonen, “Systematic review of light exposure impact on
human circadian rhythm.,” Chronobiol. Int., pp. 151–170, 2019.
[124] W. A. Hofstra and A. W. de Weerd, “How to assess circadian rhythm in humans: a review of
literature.,” Epilepsy Behav., pp. 438–444, 2008.
[125] A. U. Viola, L. M. James, L. J. Schlangen, and D. J. Dijk, “Blue-enriched white light in the
workplace improves self-reported alertness, performance and sleep quality.,” Scand. J. Work.
Environ. Health, pp. 297–306, 2008.
[126] W. K. Osterhaus, “Discomfort glare assessment and prevention for daylight applications in office
environments.,” Sol. Energy, pp. 140–158, 2005.
[127] Z. Hamedani et al., “Visual discomfort and glare assessment in office environments: A review of
light-induced physiological and perceptual responses.,” Build. Environ., pp. 267–280, 2019.
[128] S. Altomonte et al., “Ten questions concerning well-being in the built environment.,” Build.
Environ., 2020.
[129] D. P. Wyon and P. Wargocki, “Room temperature effects on office work.,” in Creating the
productive workplace, 2nd ed., Taylor & Francis, 2006.
[130] D. Phung, P. K. Thai, Y. Guo, L. Morawska, S. Rutherford, and C. Chu, “Ambient temperature
and risk of cardiovascular hospitalization: An updated systematic review and meta-analysis.,” Sci.
Total Environ., pp. 1084–1102, 2016.
[131] S. Barabino, L. Shen, L. Chen, S. Rashid, M. Rolando, and M. R. Dana, “The controlledenvironment chamber: a new mouse model of dry eye.,” Invest. Ophthalmol. Vis. Sci., pp. 2766–
2771, 2005.
[132] P. Wolkoff, “Indoor air humidity, air quality, and health–An overview.,” Int. J. Hyg. Environ.
Health, pp. 376–390, 2018.
[133] R. Jones, G. M. Recer, S. A. Hwang, and S. Lin, “Association between indoor mold and asthma
among children in Buffalo, New York,” Indoor Air, pp. 156–164, 2011.
[134] J. Sundell, “On the history of indoor air quality and health.,” Indoor Air, pp. 51–58, 2004.
[135] C. Mandin et al., “Assessment of indoor air quality in office buildings across Europe–The
OFFICAIR study.,” Sci. Total Environ., pp. 169–178, 2017.
[136] P. Wargocki, Ventilation, indoor air quality, health, and productivity. Boca Raton: CRC Press,
2016.
[137] K. W. Tham, “Indoor air quality and its effects on humans—A review of challenges and
developments in the last 30 years.,” Energy Build., pp. 637–650, 2016.
[138] G. W. Evans, “The built environment and mental health,” J. urban Heal., pp. 536–555, 2003.
[139] H. Shin, W. Park, B. Kim, K. Ji, and K. T. Kim, “Indoor Air Quality and Human Health Risk

103
Assessment for Un-regulated Small-sized Sensitive Population Facilities.,” J. Environ. Heal. Sci.,
pp. 397–407, 2018.
[140] J. G. Allen, P. MacNaughton, U. Satish, S. Santanam, J. Vallarino, and J. D. Spengler,
“Associations of cognitive function scores with carbon dioxide, ventilation, and volatile organic
compound exposures in office workers: a controlled exposure study of green and conventional
office environments.,” Environ. Health Perspect., pp. 805–812, 2016.
[141] P. MacNaughton et al., “The impact of working in a green certified building on cognitive function
and health.,” Build. Environ., pp. 178–186, 2017.
[142] P. MacNaughton, J. Spengler, J. Vallarino, S. Santanam, U. Satish, and J. Allen, “Environmental
perceptions and health before and after relocation to a green building,” Build. Environ., vol. 104,
pp. 138–144, Aug. 2016, doi: 10.1016/j.buildenv.2016.05.011.
[143] X. Zhang, P. Wargocki, Z. Lian, and C. Thyregod, “Effects of exposure to carbon dioxide and
bioeffluents on perceived air quality, self‐assessed acute health symptoms, and cognitive
performance.,” Indoor Air, pp. 47–64, 2017.
[144] M. Kohlhuber, A. Mielck, S. K. Weiland, and G. Bolte, “Social inequality in perceived
environmental exposures in relation to housing conditions in Germany.,” Environ. Res., pp. 246–
255, 2006.
[145] G. W. Evans, “The environment of childhood poverty.,” Am. Psychol., p. 77, 2004.
[146] S. H. Hong, J. Gilbertson, T. Oreszczyn, G. Green, and I. Ridley, “A field study of thermal
comfort in low-income dwellings in England before and after energy efficient refurbishment.,”
Build. Environ., pp. 1228–1236, 2009.
[147] A. E. Gold, “No home for justice: How eviction perpetuates health inequity among low-income
and minority tenants.,” Geo. J. Poverty L. Pol’y, 2016.
[148] S. N. Shah et al., “Housing quality and mental health: The association between pest infestation and
depressive symptoms among public housing residents.,” J. Urban Heal., pp. 691–702, 2018.
[149] J. Krieger and D. L. Higgins, “Housing and health: time again for public health action.,” Am. J.
Public Health, pp. 758–768, 2002.
[150] G. W. Evans, N. M. Wells, H. Y. E. Chan, and H. Saltzman, “Housing quality and mental health.,”
J. Consult. Clin. Psychol., 2000.
[151] S. F. Suglia, C. S. Duarte, and M. T. Sandel, “Housing quality, housing instability, and maternal
mental health.,” J. Urban Heal., pp. 1105–1116, 2011.
[152] R. Gifford and C. Lacombe, “Housing quality and childrens socioemotional health.,” J. Hous.
Built Environ., pp. 177–189, 2006.
[153] S. Bae, A. O. Asojo, and C. S. Martin, “Impact of occupants’ demographics on indoor
environmental quality satisfaction in the workplace.,” Build. Res. Inf., 2020.
[154] I. A. Sakellaris et al., “Perceived indoor environment and occupants’ comfort in European
‘modern’ office buildings: The OFFICAIR study.,” Int. J. Environ. Res. Public Health, 2016.
[155] J. Kim, R. de Dear, C. Candido, H. Zhang, and E. Arens, “Gender differences in office occupant
perception of indoor environmental quality (IEQ),” Build. Environ., 2013.
[156] J. K. Maykot, R. F. Rupp, and E. Ghisi, “A field study about gender and thermal comfort

104
temperatures in office buildings.,” Energy Build., pp. 254–264, 2018.
[157] H. Selye, The stress of life. New York, 1956.
[158] H. Selye, “Stress and the general adaptation syndrome.,” Br. Med. J., 1950.
[159] A. Y. Shalev, “Acute stress reactions in adults.,” Biol. Psychiatry, pp. 532–543, 2002.
[160] S. Elzeiny and M. Qaraqe, “Blueprint to Workplace Stress Detection Approaches,” in 2018
International Conference on Computer and Applications (ICCA), Aug. 2018, pp. 407–412, doi:
10.1109/COMAPP.2018.8460293.
[161] S. V. Madhu, A. Siddiqui, N. G. Desai, S. B. Sharma, and A. K. Bansal, “Chronic stress, sense of
coherence and risk of type 2 diabetes mellitus.,” Diabetes Metab. Syndr. Clin. Res. Rev., vol. 13,
no. 1, pp. 18–23, 2019.
[162] T. Cox, “Stress, coping and problem solving.,” Work Stress, pp. 5–14, 1987.
[163] R. S. Lazarus, Psychological stress and the coping process. McGraw-Hill., 1966.
[164] Cleveland Clinic, “Stress,” 2021. https://my.clevelandclinic.org/health/articles/11874-stress.
[165] F. Edition, Diagnostic and statistical manual of mental disorders. Am Psychiatric Assoc, 2013.
[166] G. Fink, “Stress science: neuroendocrinology,” in Stress: definition and history., 2010, pp. 3–9.
[167] R. A. Karasek Jr, “Job demands, job decision latitude, and mental strain: Implications for job
redesign.,” Adm. Sci. Q., pp. 285–308, 1979.
[168] J. Siegrist, “Adverse health effects of high-effort/low-reward conditions.,” J. Occup. Health
Psychol., vol. 1, no. 1, p. 27, 1996.
[169] J. V. Johnson and E. M. Hall, “Job strain, work place social support, and cardiovascular disease: a
cross-sectional study of a random sample of the Swedish working population.,” Am. J. Public
Health, vol. 78, no. 10, pp. 1336–1342, 1988.
[170] S. B. Oswalt and C. C. Riddock, “What to do about being overwhelmed: Graduate students, stress
and university services.,” Coll. Student Aff. J., vol. 27, no. 1, pp. 24–44, 2007.
[171] Mental Health America, “Mind the Workplace,” 2017.
https://www.mhanational.org/sites/default/files/Mind the Workplace - MHA Workplace Health
Survey 2017 FINAL.pdf (accessed May 07, 2021).
[172] American Psychological Association, “STRESS AND HEALTH DISPARITIES,” 2017. [Online].
Available: https://www.apa.org/pi/health-disparities/resources/stress-report.pdf.
[173] American Institute of Stress, “Workplace stress,” 2013. http://www.stress.org/workplace-stress/.
[174] J. Hassard et al., “Calculating the cost of work-related stress and psychosocial risks.,” 2014. doi:
https://eprints.bbk.ac.uk/id/eprint/20923.
[175] M. Le Fevre, J. Matheny, and G. S. Kolt, “Eustress, distress, and interpretation in occupational
stress,” J. Manag. Psychol., vol. 18, no. 7, pp. 726–744, Nov. 2003, doi:
10.1108/02683940310502412.
[176] M. B. Hargrove, W. S. Becker, and D. F. Hargrove, “The HRD eustress model: Generating
positive stress with challenging work.,” Hum. Resour. Dev. Rev., vol. 14, no. 3, pp. 279–298,
2015.

105
[177] B. L. Simmons and D. L. Nelson, “Eustress at work: Extending the holistic stress model.,” in
Positive organizational behavior, 2007, pp. 40–53.
[178] S. Lamb and K. C. S. Kwok, “A longitudinal investigation of work environment stressors on the
performance and wellbeing of office workers,” Appl. Ergon., vol. 52, pp. 104–111, Jan. 2016, doi:
10.1016/j.apergo.2015.07.010.
[179] A. Singh, M. Syal, S. Korkmaz, and S. Grady, “Costs and Benefits of IEQ Improvements in LEED
Office Buildings,” J. Infrastruct. Syst., vol. 17, no. 2, pp. 86–94, Jun. 2011, doi:
10.1061/(ASCE)IS.1943-555X.0000046.
[180] M. Awada et al., “Ten questions concerning occupant health in buildings during normal operations
and extreme events including the COVID-19 pandemic.,” Build. Environ., 2021.
[181] E. Largo-Wight, W. W. Chen, V. Dodd, and R. Weiler, “Healthy Workplaces: The Effects of
Nature Contact at Work on Employee Stress and Health,” Public Health Rep., vol. 126, no. 1, pp.
124–130, 2011, doi: 10.1177/00333549111260S116.
[182] H. Ohly et al., “Attention Restoration Theory: A systematic review of the attention restoration
potential of exposure to natural environments,” J. Toxicol. Environ. Heal. Part B, vol. 19, no. 7,
pp. 305–343, Oct. 2016, doi: 10.1080/10937404.2016.1196155.
[183] U.S. Bureau of Labor Statistics ., “2018 Standard Occupational Classification System,” 2020. .
[184] A. R. Cardoso, P. Guimarães, and J. Varejão, “Are older workers worthy of their pay? An
empirical investigation of age-productivity and age-wage nexuses.,” Economist (Leiden)., 2011.
[185] M. P. Pérez, M. P. de Luis Carnicer, and A. M. Sánchez, “Differential effects of gender on
perceptions of teleworking by human resources managers.,” Women Manag. Rev., 2002.
[186] N. Madan, D. Mani, and M. Pillutla, “The Impact of Job Need for Human Proximity and
Communication Technologies on Remote Work Efficacy,” SSRN, 2020.
[187] A. J. Miller and D. L. Carlson, “Great expectations? Working‐and middle‐class cohabitors’
expected and actual divisions of housework.,” J. Marriage Fam., pp. 346–363, 2016.
[188] J. Cerrato and E. Cifre, “Gender inequality in household chores and work-family conflict.,” Front.
Psychol., 2018.
[189] L. Colley and C. D. S. Williamson, “Working during the Pandemic: From resistance to
revolution?,” 2020.
[190] C. Leonardi, C. Mennecozzi, E. Not, F. Pianesi, and M. Zancanaro, “Designing a familiar
technology for elderly people.,” in 6th International Conference of the International Society for
Gerontechnology, ISG08, 2008, pp. 67–72.
[191] C. Conforti et al., “Impact of the COVID-19 Pandemic on Dermatology Practice Worldwide:
Results of a Survey Promoted by the International Dermoscopy Society (IDS).,” Dermatol. Pract.
Concept., 2021.
[192] M. S. Abbasi et al., “E-Learning perception and satisfaction among health sciences students amid
the COVID-19 pandemic.,” Work, pp. 1–8, 2020.
[193] C. A. Krueger et al., “COVID Will End But Telemedicine May be Here to Stay.,” J. Arthroplasty,
2021.
[194] W. Ali, “Online and remote learning in higher education institutes: A necessity in light of COVID-

106
19 pandemic.,” High. Educ. Stud., pp. 16–25, 2020.
[195] A. Singh, M. Syal, S. C. Grady, and S. Korkmaz, “Effects of green buildings on employee health
and productivity.,” Am. J. Public Health, pp. 1665–1668, 2010.
[196] Y. Al Horr, M. Arif, A. Kaushik, A. Mazroei, M. Katafygiotou, and E. Elsarrag, “Occupant
productivity and office indoor environment quality: A review of the literature.,” Build. Environ.,
2016.
[197] Y. Xiao, B. Becerik-Gerber, G. Lucas, and S. C. Roll, “Impacts of working from home during
COVID-19 pandemic on physical and mental well-being of office workstation users.,” J. Occup.
Environ. Med., vol. 63, no. 3, 2021.
[198] P. Songsangyos and S. Iamamporn, “Remote Working with Work-life Balance.,” Int. J. Appl.
Comput. Technol. Inf. Syst., 2020.
[199] N. Donnelly and S. B. Proctor‐Thomson, “Disrupted work: home‐based teleworking (HbTW) in
the aftermath of a natural disaster.,” New Technol. Work Employ., pp. 47–61, 2015.
[200] B. P. Haynes, “The impact of office layout on productivity.,” J. Facil. Manag., 2008.
[201] M. Toniolo-Barrios and L. Pitt, “Mindfulness and the challenges of working from home in times
of crisis.,” Bus. Horizons., 2020.
[202] Y. Geng, W. Ji, B. Lin, and Y. Zhu, “The impact of thermal environment on occupant IEQ
perception and productivity.,” Build. Environ., pp. 158–167, 2017.
[203] H. Jahncke, S. Hygge, N. Halin, A. M. Green, and K. Dimberg, “Open-plan office noise:
Cognitive performance and restoration.,” J. Environ. Psychol., 2011.
[204] M. Robertson et al., “The effects of an office ergonomics training and chair intervention on
worker knowledge, behavior and musculoskeletal risk.,” Appl. Ergon., vol. 40, no. 1, pp. 124–135,
2009.
[205] D. Russo, P. H. Hanel, S. Altnickel, and N. van Berkel, “Predictors of Well-being and Productivity
among Software Professionals during the COVID-19 Pandemic--A Longitudinal Study.,” arXiv
Prepr. arXiv2007.12580, 2020.
[206] S. Schieman, P. J. Badawy, M. Milkie, and A. Bierman, “Work-Life Conflict During the COVID19 Pandemic.” 2020, doi: 10.13140/RG.2.2.32315.44329.
[207] J. Morris and J. Hassard, “Home Working? The Present and Future of How and Where We Work
in the Context of COVID-19. Cardiff Business School COVID-19 and Work WP 2020,” 2020.
[208] L. Waizenegger, B. McKenna, W. Cai, and T. Bendz, “An affordance perspective of team
collaboration and enforced working from home during COVID-19.,” Eur. J. Inf. Syst., pp. 1–14,
2020.
[209] L. Wattis, K. Standing, and M. A. Yerkes, “Mothers and work–life balance: Exploring the
contradictions and complexities involved in work–family negotiation.,” Community. Work Fam.,
pp. 1–19, 2013.
[210] B. N. Burk, A. Pechenik Mausolf, and L. Oakleaf, “Pandemic motherhood and the academy: A
critical examination of the leisure-work dichotomy.,” Leis. Sci., pp. 1–7, 2020.
[211] A. B. Sakpere, I. Olanipekun, and W. Sakpere, “Work Productivity in the Period of COVID-19
Pandemic and Lockdown: A Developing World Perspective.,” 2020.

107
[212] Pew Research Center, “How the Coronavirus Outbreak Has – and Hasn’t – Changed the Way
Americans Work,” 2020.
[213] S. C. O. T. T. Keeter, “People financially affected by COVID-19 outbreak are experiencing more
psychological distress than others.,” 2020.
[214] J. D. Baker, “Online survey software.,” 2013.
[215] C. Hogue, “Government organization summary report: 2012.,” 2013.
[216] A. Bee, “Household income inequality within US counties: 2006-2010.,” 2012.
[217] R. Torre and M. Myrskylä, “Income inequality and population health: an analysis of panel data for
21 developed countries, 1975–2006.,” Popul. Stud. (NY)., pp. 1–13, 2014.
[218] A. Joshi, S. Kale, S. Chandel, and D. K. Pal, “Likert scale: Explored and explained.,” Br. J. Appl.
Sci. Technol., 2015.
[219] J. W. Tukey, “Comparing individual means in the analysis of variance.,” Biometrics, pp. 99–114,
1949.
[220] R. E. Wright, “Logistic regression.,” in Reading and understanding multivariate statistics,
American Psychological Association, 1995, pp. 217–244.
[221] J. E. González and M. Krarti, “Reflecting on Impacts of COVID19 on Sustainable Buildings and
Cities,” J. Eng. Sustain. Bldgs. Cities, 2021.
[222] O. B. Oluesdryi, “Consequences of residential conditions on psychological well-being of
communities in Kwara state, Nigeria.,” Quantum J. Eng. Sci. Technol., pp. 21–32, 2020.
[223] D. Godinic, B. Obrenovic, and A. Khudaykulov, “Effects of economic uncertainty on mental
health in the COVID-19 pandemic context: social identity disturbance, job uncertainty and
psychological well-being model.,” Int. J. Innov. Econ. Dev, pp. 61–74, 2020.
[224] M. Indraganti and K. D. Rao, “Effect of age, gender, economic group and tenure on thermal
comfort: A field study in residential buildings in hot and dry climate with seasonal variations.,”
Energy Build., pp. 273–281, 2010.
[225] S. Leder, G. R. Newsham, J. A. Veitch, S. Mancini, and K. E. Charles, “Effects of office
environment on employee satisfaction: a new analysis.,” Build. Res. Inf., vol. 44, no. 1, pp. 34–50,
2016.
[226] J. H. Choi, V. Loftness, D. Nou, B. Tinianov, and D. Yeom, “Multi-Season Assessment of
Occupant Responses to Manual Shading and Dynamic Glass in a Workplace Environment.,”
Energies, 2020.
[227] D. S. Shin, B. Y. Jeong, and M. H. Park, “Structural equation modeling of office environment
quality, sick building syndrome, and musculoskeletal complaints on aggregate satisfaction of
office workers.,” Hum. Factors Ergon. Manuf. Serv. Ind., 2018.
[228] C. Blehm, S. Vishnu, A. Khattak, S. Mitra, and R. W. Yee, “Computer vision syndrome: a
review.,” Surv. Ophthalmol., 2005.
[229] R. Mork, H. K. Falkenberg, K. I. Fostervold, and H. M. S. Thorud, “Discomfort glare and
psychological stress during computer work: subjective responses and associations between neck
pain and trapezius muscle blood flow.,” Int. Arch. Occup. Environ. Health, pp. 29–42, 2020.

108
[230] M. I. ALI, P. NASIRI, M. Azkhosh, and M. HOSSEINI, “Factors affecting road traffic noise
annoyance among white-collar employees working in Tehran.,” 2010.
[231] M. J. Brown and D. E. Jacobs, “Residential light and risk for depression and falls: results from the
LARES study of eight European cities.,” Public Health Rep., pp. 131–140, 2011.
[232] C. Liao, “Association between Indoor Air Quality and Sleep Quality.,” 2019.
[233] P. Strøm‐Tejsen, D. Zukowska, P. Wargocki, and D. P. Wyon, “The effects of bedroom air quality
on sleep and next‐day performance,” Indoor Air, pp. 679–686, 2016.
[234] S. Di Blasio, L. Shtrepi, G. E. Puglisi, and A. Astolfi, “A cross-sectional survey on the impact of
irrelevant speech noise on annoyance, mental health and well-being, performance and occupants’
behavior in shared and open-plan offices.,” Int. J. Environ. Res. Public Health, 2019.
[235] R. Monteiro, D. Tomé, P. Neves, D. Silva, and M. A. Rodrigues, “The interactive effect of
occupational noise on attention and short-term memory: A pilot study.,” Noise Health, 2018.
[236] A. Lipczynska, S. Schiavon, and L. T. Graham, “Thermal comfort and self-reported productivity
in an office with ceiling fans in the tropics.,” Build. Environ., pp. 202–212, 2018.
[237] T. Hoyt, K. H. Lee, H. Zhang, E. Arens, and T. Webster, “Energy savings from extended air
temperature setpoints and reductions in room air mixing.,” 2005.
[238] & A. N. S. I. American Society of Heating, Refrigerating, Air-Conditioning Engineers, “Thermal
environmental conditions for human occupancy,” 2004.
[239] S. A. Samani, “The impact of personal control over office workspace on environmental
satisfaction and performance.,” J. Soc. Sci. Humanit., pp. 163–175, 2015.
[240] J. M. Barrero, N. Bloom, and S. J. Davis, “Why Working From Home Will Stick.,” Univ.
Chicago, Becker Friedman Inst. Econ. Work. Pap., 2020.
[241] M. C. Baechler, T. L. Gilbride, P. C. Cole, M. G. Hefty, and K. Ruiz, “High-Performance Home
Technologies: Guide to Determining Climate Regions by County,” 2015. [Online]. Available:
chromeextension://efaidnbmnnnibpcajpcglclefindmkaj/viewer.html?pdfurl=https%3A%2F%2Fwww.ener
gy.gov%2Fsites%2Fprod%2Ffiles%2F2015%2F10%2Ff27%2Fba_climate_region_guide_7.3.pdf
&clen=6356555&chunk=true.
[242] C. McCarthy, N. Pradhan, C. Redpath, and A. Adler, “Validation of the Empatica E4 wristband,”
in 2016 IEEE EMBS International Student Conference (ISC), May 2016, pp. 1–4, doi:
10.1109/EMBSISC.2016.7508621.
[243] “Polar Electro,” 2023. https://www.polar.com/us-en/sensors/h10-heart-rate-sensor/ (accessed Aug.
20, 2023).
[244] J. A. Albert, V. Owolabi, A. Gebel, C. M. Brahms, U. Granacher, and B. Arnrich, “Evaluation of
the Pose Tracking Performance of the Azure Kinect and Kinect v2 for Gait Analysis in
Comparison with a Gold Standard: A Pilot Study,” Sensors, vol. 20, no. 18, p. 5104, Sep. 2020,
doi: 10.3390/s20185104.
[245] “Mini Mouse Macro,” 2023. https://sourceforge.net/projects/minimousemacro/reviews/.
[246] M. Kozusznik, J. M. Peiró, S. Lloret, and I. Rodriguez, “Hierarchy of Eustress and Distress: Rasch
Calibration of the Valencia Eustress-Distress Appraisal Scale,” Cent. Eur. J. Manag., vol. 2, no.
1,2, Jun. 2016, doi: 10.5817/CEJM2015-1-2-5.

109
[247] A. Di Fabio, J. M. Peiró, I. Rodríguez, and M. W. Kozusznik, “The Valencia Eustress-Distress
Appraisal Scale (VEDAS): Validation of the Italian Version,” Sustainability, vol. 10, no. 11, p.
3903, Oct. 2018, doi: 10.3390/su10113903.
[248] A. Bernardes et al., “How Reliable Are Ultra-Short-Term HRV Measurements during Cognitively
Demanding Tasks?,” Sensors, vol. 22, no. 17, p. 6528, Aug. 2022, doi: 10.3390/s22176528.
[249] M. P. Tarvainen, J.-P. Niskanen, J. A. Lipponen, P. O. Ranta-aho, and P. A. Karjalainen, “Kubios
HRV – Heart rate variability analysis software,” Comput. Methods Programs Biomed., vol. 113,
no. 1, pp. 210–220, Jan. 2014, doi: 10.1016/j.cmpb.2013.07.024.
[250] M. Gjoreski, M. Luštrek, M. Gams, and H. Gjoreski, “Monitoring stress with a wrist device using
context,” J. Biomed. Inform., vol. 73, pp. 159–170, Sep. 2017, doi: 10.1016/j.jbi.2017.08.006.
[251] M. Wu, “Trimmed and winsorized estimators,” Michigan State University., 2006.
[252] M. Benedek and C. Kaernbach, “A continuous measure of phasic electrodermal activity.,” J.
Neurosci. Methods, vol. 190, no. 1, pp. 80–91, 2010.
[253] A. Alberdi, A. Aztiria, and A. Basarab, “Towards an automatic early stress recognition system for
office environments based on multimodal measurements: A review.,” J. Biomed. Inform., vol. 59,
pp. 49–75, 2016.
[254] T. Baltrusaitis, P. Robinson, and L.-P. Morency, “OpenFace: An open source facial behavior
analysis toolkit,” in 2016 IEEE Winter Conference on Applications of Computer Vision (WACV),
Mar. 2016, pp. 1–10, doi: 10.1109/WACV.2016.7477553.
[255] S. Koldijk, M. A. Neerincx, and W. Kraaij, “Detecting Work Stress in Offices by Combining
Unobtrusive Sensors,” IEEE Trans. Affect. Comput., vol. 9, no. 2, pp. 227–239, Apr. 2018, doi:
10.1109/TAFFC.2016.2610975.
[256] Y. M. Lim, A. Ayesh, and M. Stacey, “Detecting cognitive stress from keyboard and mouse
dynamics during mental arithmetic,” in 2014 Science and Information Conference, Aug. 2014, pp.
146–152, doi: 10.1109/SAI.2014.6918183.
[257] E. A. Sağbaş, S. Korukoglu, and S. Balli, “Stress Detection via Keyboard Typing Behaviors by
Using Smartphone Sensors and Machine Learning Techniques,” J. Med. Syst., vol. 44, no. 4, p. 68,
Apr. 2020, doi: 10.1007/s10916-020-1530-z.
[258] H. Kang, “The prevention and handling of the missing data,” Korean J. Anesthesiol., vol. 64, no.
5, p. 402, 2013, doi: 10.4097/kjae.2013.64.5.402.
[259] scikit learn, “sklearn.preprocessing.RobustScaler.” https://scikitlearn.org/stable/modules/generated/sklearn.preprocessing.RobustScaler.html.
[260] N. V. Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “SMOTE: Synthetic Minority
Over-sampling Technique,” J. Artif. Intell. Res., vol. 16, pp. 321–357, Jun. 2002, doi:
10.1613/jair.953.
[261] W. Wang and D. Sun, “The improved AdaBoost algorithms for imbalanced data classification,”
Inf. Sci. (Ny)., vol. 563, pp. 358–374, Jul. 2021, doi: 10.1016/j.ins.2021.03.042.
[262] I. Rish, “An empirical study of the naive Bayes classifier.,” in IJCAI 2001 workshop on empirical
methods in artificial intelligence, 2001, pp. 41–46.
[263] A. J. Valkenburg, S. P. Niehof, M. van Dijk, E. J. M. Verhaar, and D. Tibboel, “Skin conductance
peaks could result from changes in vital parameters unrelated to pain,” Pediatr. Res., vol. 71, no.

110
1–4, pp. 375–379, Apr. 2012, doi: 10.1038/pr.2011.72.
[264] R. Todeschini, “k-nearest neighbour method: The influence of data transformations and metrics,”
Chemom. Intell. Lab. Syst., vol. 6, no. 3, pp. 213–220, Sep. 1989, doi: 10.1016/0169-
7439(89)80086-3.
[265] C. Zhang and Y. Ma, Ensemble machine learning: methods and applications. Springer Science &
Business Media., 2012.
[266] T. Chen and C. Guestrin, “XGBoost,” in Proceedings of the 22nd ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining, Aug. 2016, pp. 785–794, doi:
10.1145/2939672.2939785.
[267] D. R. Seshadri et al., “Wearable Sensors for COVID-19: A Call to Action to Harness Our Digital
Infrastructure for Remote Patient Monitoring and Virtual Assessments,” Front. Digit. Heal., vol.
2, Jun. 2020, doi: 10.3389/fdgth.2020.00008.
[268] C. M. Stoney, M. C. Davis, and K. A. Matthews, “Sex Differences in Physiological Responses to
Stress and in Coronary Heart Disease: A Causal Link?,” Psychophysiology, vol. 24, no. 2, pp.
127–131, Mar. 1987, doi: 10.1111/j.1469-8986.1987.tb00264.x.
[269] M. A. Rosales, A. A. Bandala, R. R. Vicerra, and E. P. Dadios, “Physiological-Based Smart Stress
Detector using Machine Learning Algorithms,” in 2019 IEEE 11th International Conference on
Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment,
and Management ( HNICEM ), Nov. 2019, pp. 1–6, doi: 10.1109/HNICEM48295.2019.9073355.
[270] C. Setz, B. Arnrich, J. Schumm, R. La Marca, G. Troster, and U. Ehlert, “Discriminating Stress
From Cognitive Load Using a Wearable EDA Device,” IEEE Trans. Inf. Technol. Biomed., vol.
14, no. 2, pp. 410–417, Mar. 2010, doi: 10.1109/TITB.2009.2036164.
[271] G. R. Davis, D. Bellar, D. Scott, and M. Lavergne, “Effects of core temperature, skin temperature,
and inter-beat interval on resting metabolic rate measurements in thermoneutral conditions,” J.
Therm. Biol., vol. 85, p. 102399, Oct. 2019, doi: 10.1016/j.jtherbio.2019.102399.
[272] T. Iqbal et al., “A Sensitivity Analysis of Biophysiological Responses of Stress for Wearable
Sensors in Connected Health,” IEEE Access, vol. 9, pp. 93567–93579, 2021, doi:
10.1109/ACCESS.2021.3082423.
[273] L. Potter, J. Scallon, D. Swegle, T. Gould, and G. E. Okudan Kremer, “Establishing a Link
between Electrodermal Activity and Classroom Engagement.,” in IIE Annual Conference.
Proceedings, 2019, pp. 988–993.
[274] R. Holder, R. K. Sah, M. Cleveland, and H. Ghasemzadeh, “Comparing the Predictability of
Sensor Modalities to Detect Stress from Wearable Sensor Data,” in 2022 IEEE 19th Annual
Consumer Communications & Networking Conference (CCNC), Jan. 2022, pp. 557–562, doi:
10.1109/CCNC49033.2022.9700682.
[275] M. B. Morshed et al., “Advancing the Understanding and Measurement of Workplace Stress in
Remote Information Workers from Passive Sensors and Behavioral Data.,” 2022, [Online].
Available: https://www.researchgate.net/profile/Mehrab-Bin-Morshed2/publication/363885542_Advancing_the_Understanding_and_Measurement_of_Workplace_Stre
ss_in_Remote_Information_Workers_from_Passive_Sensors_and_Behavioral_Data/links/63336c8
213096c2907d43f18/Advancing.
[276] G. Giannakakis, M. R. Koujan, A. Roussos, and K. Marias, “Automatic stress detection evaluating
models of facial action units,” in 2020 15th IEEE International Conference on Automatic Face

111
and Gesture Recognition (FG 2020), Nov. 2020, pp. 728–733, doi: 10.1109/FG47880.2020.00129.
[277] P. Goldberg et al., “Attentive or Not? Toward a Machine Learning Approach to Assessing
Students’ Visible Engagement in Classroom Instruction,” Educ. Psychol. Rev., vol. 33, no. 1, pp.
27–49, Mar. 2021, doi: 10.1007/s10648-019-09514-z.
[278] B. De Carolis, F. D’Errico, N. Macchiarulo, and G. Palestra, “‘Engaged Faces’: Measuring and
Monitoring Student Engagement from Face and Gaze Behavior,” in IEEE/WIC/ACM International
Conference on Web Intelligence - Companion Volume, Oct. 2019, pp. 80–85, doi:
10.1145/3358695.3361748.
[279] J. Yang, K. Wang, X. Peng, and Y. Qiao, “Deep Recurrent Multi-instance Learning with Spatiotemporal Features for Engagement Intensity Prediction,” in Proceedings of the 20th ACM
International Conference on Multimodal Interaction, Oct. 2018, pp. 594–598, doi:
10.1145/3242969.3264981.
[280] Azure - Microsoft, “Azure Kinect DK,” 2023. https://azure.microsoft.com/en-us/products/kinectdk (accessed Jul. 09, 2023).
[281] H. Yu, E. B. Klerman, R. W. Picard, and A. Sano, “Personalized Wellbeing Prediction using
Behavioral, Physiological and Weather Data,” in 2019 IEEE EMBS International Conference on
Biomedical & Health Informatics (BHI), May 2019, pp. 1–4, doi: 10.1109/BHI.2019.8834456.
[282] C.-T. Li, J. Cao, and T. M. H. Li, “Eustress or distress: An empirical study of perceived stress in
everyday college life,” in Proceedings of the 2016 ACM International Joint Conference on
Pervasive and Ubiquitous Computing: Adjunct, Sep. 2016, pp. 1209–1217, doi:
10.1145/2968219.2968309.
[283] F. Albertetti, A. Simalastar, and A. Rizzotti-Kaddouri, “Stress Detection with Deep Learning
Approaches Using Physiological Signals,” 2021, pp. 95–111.
[284] S. K. Vosburg, “Mood and the Quantity and Quality of Ideas,” Creat. Res. J., vol. 11, no. 4, pp.
315–324, Oct. 1998, doi: 10.1207/s15326934crj1104_5.
[285] M. Cleary, J. Sayers, V. Lopez, and C. Hungerford, “Boredom in the Workplace: Reasons, Impact,
and Solutions,” Issues Ment. Health Nurs., vol. 37, no. 2, pp. 83–89, Feb. 2016, doi:
10.3109/01612840.2015.1084554.
[286] W. L. Romine et al., “Using Machine Learning to Train a Wearable Device for Measuring
Students’ Cognitive Load during Problem-Solving Activities Based on Electrodermal Activity,
Body Temperature, and Heart Rate: Development of a Cognitive Load Tracker for Both Personal
and Cla,” Sensors, vol. 20, no. 17, p. 4833, Aug. 2020, doi: 10.3390/s20174833.
[287] S. H. Cha, J. Seo, S. H. Baek, and C. Koo, “Towards a well-planned, activity-based work
environment: Automated recognition of office activities using accelerometers,” Build. Environ.,
vol. 144, pp. 86–93, Oct. 2018, doi: 10.1016/j.buildenv.2018.07.051.
[288] M. Bin Morshed et al., “Advancing the Understanding and Measurement of Workplace Stress in
Remote Information Workers from Passive Sensors and Behavioral Data,” in 2022 10th
International Conference on Affective Computing and Intelligent Interaction (ACII), Oct. 2022,
pp. 1–8, doi: 10.1109/ACII55700.2022.9953824.
[289] Empatica E4, “EMPATICA S.R.L.”
https://box.empatica.com/documentation/20141119_E4_TechSpecs.pdf (accessed Oct. 10, 2023).
[290] M. Schaffarczyk, B. Rogers, R. Reer, and T. Gronwald, “Validity of the Polar H10 Sensor for

112
Heart Rate Variability Analysis during Resting State and Incremental Exercise in Recreational
Men and Women,” Sensors, vol. 22, no. 17, p. 6536, Aug. 2022, doi: 10.3390/s22176536.
[291] M. Awada et al., “Ten questions concerning the impact of environmental stress on office
workers,” Build. Environ., vol. 229, p. 109964, Feb. 2023, doi: 10.1016/j.buildenv.2022.109964.
[292] K. Shobe, “Productivity Driven by Job Satisfaction, Physical Work Environment, Management
Support and Job Autonomy,” Bus. Econ. J., vol. 09, no. 02, 2018, doi: 10.4172/2151-
6219.1000351.
[293] F. Faul, E. Erdfelder, A. Buchner, and A. G. Lang, “Statistical power analyses using G* Power
3.1: Tests for correlation and regression analyses.,” Behav. Res. Methods, vol. 41, no. 4, pp. 1149–
1160, 2009.
[294] HearingLife, “Online Hearing Test.” https://www.hearinglife.com/hearing-loss/your-hearing/testyour-hearing.
[295] Center for Disease Control and Prevention, “What Noises Cause Hearing Loss?,” Loud Noise Can
Cause Hearing Loss, 2019.
https://www.cdc.gov/nceh/hearing_loss/what_noises_cause_hearing_loss.html#:~:text=Noise
above 70 dB over,immediate harm to your ears. (accessed Apr. 02, 2022).
[296] United States. Office of Noise Abatement Control., Information on levels of environmental noise
requisite to protect public health and welfare with an adequate margin of safety. US Government
Printing Office, 1974.
[297] A. J. Angwin, W. J. Wilson, W. L. Arnott, A. Signorini, R. J. Barry, and D. A. Copland, “White
noise enhances new-word learning in healthy adults,” Sci. Rep., 2017.
[298] A. J. Angwin et al., “White noise facilitates new-word learning from context.,” Brain Lang., 2019.
[299] E. Othman et al., “Low intensity white noise improves performance in auditory working memory
task: An fMRI study.,” Heliyon, vol. 5, no. 9, 2019.
[300] B. Products, “Advanced - Decibel Meter / Sound Level Reader - W/ Battery!”
https://bafxpro.com/products/bafx-products-decibel-meter-sound-level-reader-w-batteryadvanced-sound-meter.
[301] J. W. Peirce, “PsychoPy—psychophysics software in Python.,” J. Neurosci. Methods, 2007.
[302] N. Shalev, G. Humphreys, and N. Demeyere, “Manipulating perceptual parameters in a continuous
performance task,” Behav. Res. Methods, 2018.
[303] S. Raz, Y. Bar-Haim, A. Sadeh, and O. Dan, “Reliability and Validity of the Online Continuous
Performance Test Among Young Adults,” Assessment, vol. 21, no. 1, pp. 108–118, Feb. 2014, doi:
10.1177/1073191112443409.
[304] F. Scarpina and S. Tagini, “The stroop color and word test.,” Front. Psychol., 2017.
[305] A. M. Owen, K. M. McMillan, A. R. Laird, and E. Bullmore, “N‐back working memory paradigm:
A meta‐analysis of normative functional neuroimaging studies.,” Hum. Brain Mapp., 2005.
[306] A. HOCKEY and G. GEFFEN, “The concurrent validity and test?retest reliability of a visuospatial
working memory task,” Intelligence, vol. 32, no. 6, pp. 591–605, Dec. 2004, doi:
10.1016/j.intell.2004.07.009.
[307] M. T. Mednick, S. A. Mednick, and E. V. Mednick, “Incubation of creative performance and

113
specific associative priming.,” J. Abnorm. Soc. Psychol., 1964.
[308] S. A. MEDNICK, “The Remote Associates Test*,” J. Creat. Behav., vol. 2, no. 3, pp. 213–214,
Jul. 1968, doi: 10.1002/j.2162-6057.1968.tb00104.x.
[309] “Remote Associates Test,” A collection of tasks from the Remote Associates Test of Creativity.
https://www.remote-associates-test.com/.
[310] L. Lan, P. Wargocki, and Z. Lian, “Thermal effects on human performance in office environment
measured by integrating task speed and accuracy.,” Appl. Ergon., 2014.
[311] I. Teachnology, “Teachnology.” https://www.teach-nology.com/.
[312] W. Prokasy, Electrodermal activity in psychological research. New York: Elsevier, 2012.
[313] H. F. Posada-Quintero and K. H. Chon, “Innovations in electrodermal activity data collection and
signal processing: A systematic review.,” Sensors, vol. 20, no. 2, p. 479, 2020.
[314] Empatica, “E4 wristband,” 2021.
https://www.empatica.com/research/e4/?gclid=CjwKCAjwtpGGBhBJEiwAyRZX2mR4HYW67O
3yXtIcz7eDYJw21RWNNg0pBCJogFM1008kEjhg-3rMvhoCaWoQAvD_BwE.
[315] M. Benedek and C. Kaernbach, “Decomposition of skin conductance data by means of
nonnegative deconvolution,” Psychophysiology, Mar. 2010, doi: 10.1111/j.1469-
8986.2009.00972.x.
[316] D. M. Alexander, C. Trengove, P. Johnston, T. Cooper, J. P. August, and E. Gordon, “Separating
individual skin conductance responses in a short interstimulus-interval paradigm,” J. Neurosci.
Methods, vol. 146, no. 1, pp. 116–123, Jul. 2005, doi: 10.1016/j.jneumeth.2005.02.001.
[317] J. V. Bradley, “Complete Counterbalancing of Immediate Sequential Effects in a Latin Square
Design,” J. Am. Stat. Assoc., vol. 53, no. 282, pp. 525–528, Jun. 1958, doi:
10.1080/01621459.1958.10501456.
[318] D. George and P. Mallery, IBM SPSS Statistics 26 Step by Step. Routledge, 2019.
[319] V. H. Rausch, E. M. Bauch, and N. Bunzeck, “White noise improves learning by modulating
activity in dopaminergic midbrain regions and right superior temporal sulcus.,” J. Cogn.
Neurosci., vol. 26, no. 7, pp. 1469–1480, 2014.
[320] M. Usher and M. Feingold, “Stochastic resonance in the speed of memory retrieval,” Biol.
Cybern., vol. 83, no. 6, pp. L011–L016, Nov. 2000, doi: 10.1007/PL00007974.
[321] N. K. Ilkkaya et al., “The effects of music, white noise, and ambient noise on sedation and anxiety
in patients under spinal anesthesia during surgery.,” J. Perianesthesia Nurs., vol. 29, no. 5, pp.
418–426, 2014.
[322] P. J. Jastreboff, “Tinnitus habituation therapy (THI) and tinnitus retraining therapy (THI),” in
Tinnitus handbook, 2000, pp. 357-376.
[323] R. Mehta, R. Zhu, and A. Cheema, “Is noise always bad? Exploring the effects of ambient noise
on creative cognition.,” J. Consum. Res., vol. 39, no. 4, pp. 784–799, 2012.
[324] C. Martindale and J. Greenough, “The differential effect of increased arousal on creative and
intellectual performance.,” J. Genet. Psychol., vol. 123, no. 2, pp. 329–335, 1973.
[325] A. Hillier, J. K. Alexander, and D. Q. Beversdorf, “The effect of auditory stressors on cognitive

114
flexibility.,” Neurocase, 2006.
[326] J. Yin, N. Arfaei, P. MacNaughton, P. J. Catalano, J. G. Allen, and J. D. Spengler, “Effects of
biophilic interventions in office on stress reaction and cognitive function: A randomized crossover
study in virtual reality,” Indoor Air, vol. 29, no. 6, pp. 1028–1039, Nov. 2019, doi:
10.1111/ina.12593.
[327] D. Li and W. C. Sullivan, “Impact of views to school landscapes on recovery from stress and
mental fatigue.,” Landsc. Urban Plan., vol. 148, pp. 149–158, 2016.
[328] X. Liu, K. Iwanaga, Y. Shimomura, and T. Katsuura, “Comparison of stress responses between
mental tasks and white noise exposure.,” J. Physiol. Anthropol., vol. 26, no. 2, pp. 165–171, 2007.
[329] U. Kraus et al., “Individual Daytime Noise Exposure during Routine Activities and Heart Rate
Variability in Adults: A Repeated Measures Study,” Environ. Health Perspect., vol. 121, no. 5, pp.
607–612, May 2013, doi: 10.1289/ehp.1205606.
[330] L. J. Loewen and P. Suedfeld, “Cognitive and Arousal Effects of Masking Office Noise,” Environ.
Behav., vol. 24, no. 3, pp. 381–395, May 1992, doi: 10.1177/0013916592243006.
[331] R. Cools and M. D’Esposito, “Inverted-U–shaped dopamine actions on human working memory
and cognitive control.,” Biol. Psychiatry, vol. 69, no. 12, pp. 113–125, 2011.
[332] A. Azizi and P. Ghafoorpoor Yazdi, “Introduction to noise and its applications.,” in ComputerBased Analysis of the Stochastic Stability of Mechanical Structures Driven by White and Colored
Noise, 2019, pp. 13–23.
[333] R. D. Badgaiyan, “Detection of dopamine neurotransmission in ‘real time’.,” Front. Neurosci.,
2013.
[334] Awair inc., “Awair Omni,” 2022. https://www.getawair.com/products/omni.
[335] J. Yin, S. Zhu, P. MacNaughton, J. G. Allen, and J. D. Spengler, “Physiological and cognitive
performance of exposure to biophilic indoor environment,” Build. Environ., vol. 132, pp. 255–262,
Mar. 2018, doi: 10.1016/j.buildenv.2018.01.006.
[336] M. Hedblom, B. Gunnarsson, M. Schaefer, I. Knez, P. Thorsson, and J. N. Lundström, “Sounds of
Nature in the City: No Evidence of Bird Song Improving Stress Recovery,” Int. J. Environ. Res.
Public Health, vol. 16, no. 8, p. 1390, Apr. 2019, doi: 10.3390/ijerph16081390.
[337] Illuminating Engineering Society, “The IES Lighting Library Standards Collection,” 2023.
https://www.ies.org/standards/lighting-library/ (accessed Oct. 02, 2024).
[338] Amazon, “Torkase Smart Light Bulbs Work with Alexa and Google Home - No Hub Required,
WiFi LED Bulb 2700K Warm White to 6500K Daylight Dimmable, 7W(60W Equivalent), APP
Remote Control, 2.4Ghz Only, 1 Pack,” 2023. https://www.amazon.com/Torkase-Smart-LightBulbs-Google/dp/B08Q825LQN (accessed Nov. 08, 2023).
[339] SPER SCIENTIFIC DIRECT, “Color LED Chroma Light Meter,” 2023.
https://sperdirect.com/products/chroma-lightmeter?currency=USD&variant=37811395395741&utm_medium=cpc&utm_source=google&utm_
campaign=Google Shopping&gclid=CjwKCAjwge2iBhBBEiwAfXDBRwysMQy_g1T4NhvB0t7-gtL-Fdlu-PD5jPUYzjsvJj65F1JwSW-CBoCIXAQAvD_BwE&gad=1 (accessed
Nov. 08, 2023).
[340] H.-G. Kim, E.-J. Cheon, D.-S. Bai, Y. H. Lee, and B.-H. Koo, “Stress and Heart Rate Variability:

115
A Meta-Analysis and Review of the Literature,” Psychiatry Investig., vol. 15, no. 3, pp. 235–245,
Mar. 2018, doi: 10.30773/pi.2017.08.17.
[341] A. Bali and A. S. Jaggi, “Clinical experimental stress studies: methods and assessment,” Rev.
Neurosci., vol. 26, no. 5, pp. 555–579, Oct. 2015, doi: 10.1515/revneuro-2015-0004.
[342] E. M. Penna et al., “Mental Fatigue Impairs Physical Performance in Young Swimmers,” Pediatr.
Exerc. Sci., vol. 30, no. 2, pp. 208–215, May 2018, doi: 10.1123/pes.2017-0128.
[343] J. J. Alvarsson, S. Wiens, and M. E. Nilsson, “Stress Recovery during Exposure to Nature Sound
and Environmental Noise,” Int. J. Environ. Res. Public Health, vol. 7, no. 3, pp. 1036–1046, Mar.
2010, doi: 10.3390/ijerph7031036.
[344] A. G. Emfield and M. B. Neider, “Evaluating visual and auditory contributions to the cognitive
restoration effect,” Front. Psychol., vol. 5, Jun. 2014, doi: 10.3389/fpsyg.2014.00548.
[345] G. Barbiero, R. Berto, A. Venturella, and N. Maculan, “Bracing Biophilia: When biophilic design
promotes pupil’s attentional performance, perceived restorativeness and affiliation with Nature,”
Environ. Dev. Sustain., Nov. 2021, doi: 10.1007/s10668-021-01903-1.
[346] M. G. Berman, J. Jonides, and S. Kaplan, “The Cognitive Benefits of Interacting With Nature,”
Psychol. Sci., vol. 19, no. 12, pp. 1207–1212, Dec. 2008, doi: 10.1111/j.1467-9280.2008.02225.x.
[347] J. Yin, J. Yuan, N. Arfaei, P. J. Catalano, J. G. Allen, and J. D. Spengler, “Effects of biophilic
indoor environment on stress and anxiety recovery: A between-subjects experiment in virtual
reality,” Environ. Int., vol. 136, p. 105427, Mar. 2020, doi: 10.1016/j.envint.2019.105427.
[348] T. M. Marteau and H. Bekker, “The development of a six‐item short‐form of the state scale of the
Spielberger State—Trait Anxiety Inventory (STAI),” Br. J. Clin. Psychol., vol. 31, no. 3, pp. 301–
306, Sep. 1992, doi: 10.1111/j.2044-8260.1992.tb00997.x.
[349] E. Åhsberg, F. Garnberale, and A. Kjellberg, “Perceived quality of fatigue during different
occupational tasks Development of a questionnaire,” Int. J. Ind. Ergon., vol. 20, no. 2, pp. 121–
135, Aug. 1997, doi: 10.1016/S0169-8141(96)00044-3.
[350] A. Kuijsters, J. Redi, B. de Ruyter, and I. Heynderickx, “Lighting to Make You Feel Better:
Improving the Mood of Elderly People with Affective Ambiences,” PLoS One, vol. 10, no. 7, p.
e0132732, Jul. 2015, doi: 10.1371/journal.pone.0132732.
[351] Y.-M. Mu et al., “Alerting effects of light in healthy individuals: a systematic review and metaanalysis,” Neural Regen. Res., vol. 17, no. 9, p. 1929, 2022, doi: 10.4103/1673-5374.335141.
[352] S. Westland, Q. Pan, and S. Lee, “A review of the effects of colour and light on non‐image
function in humans,” Color. Technol., vol. 133, no. 5, pp. 349–361, Oct. 2017, doi:
10.1111/cote.12289.
[353] I. Provencio and Warthen, “The role of intrinsically photosensitive retinal ganglion cells in
nonimage-forming responses to light,” Eye Brain, p. 43, Sep. 2012, doi: 10.2147/EB.S27839.
[354] L. K. Grant, B. A. Kent, M. D. Mayer, R. Stickgold, S. W. Lockley, and S. A. Rahman, “Daytime
Exposure to Short Wavelength-Enriched Light Improves Cognitive Performance in SleepRestricted College-Aged Adults,” Front. Neurol., vol. 12, Feb. 2021, doi:
10.3389/fneur.2021.624217.
[355] S. Lehrl et al., “Blue light improves cognitive performance,” J. Neural Transm., vol. 114, no. 4,
pp. 457–460, Apr. 2007, doi: 10.1007/s00702-006-0621-4.

116
[356] S. L. Chellappa, R. Steiner, P. Blattner, P. Oelhafen, T. Götz, and C. Cajochen, “Non-Visual
Effects of Light on Melatonin, Alertness and Cognitive Performance: Can Blue-Enriched Light
Keep Us Alert?,” PLoS One, vol. 6, no. 1, p. e16429, Jan. 2011, doi:
10.1371/journal.pone.0016429.
[357] S. Kaplan, “The restorative benefits of nature: Toward an integrative framework,” J. Environ.
Psychol., vol. 15, no. 3, pp. 169–182, Sep. 1995, doi: 10.1016/0272-4944(95)90001-2.
[358] S. Higuchi et al., “Is the use of high correlated color temperature light at night related to delay of
sleep timing in university students? A cross-country study in Japan and China,” J. Physiol.
Anthropol., vol. 40, no. 1, p. 7, Dec. 2021, doi: 10.1186/s40101-021-00257-x.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Understanding human-building-emergency interactions in the built environment

PDF

Understanding human-building interactions through perceptual decision-making processes

PDF

Enabling human-building communication to promote pro-environmental behavior in office buildings

PDF

In-situ quality assessment of scan data for as-built models using building-specific geometric features

PDF

A radio frequency based indoor localization framework for supporting building emergency response operations

PDF

Impact of virtual reality (VR)-based training on construction robotics remote-operation

PDF

Point cloud data fusion of RGB and thermal information for advanced building envelope modeling in support of energy audits for large districts

PDF

Indoor air quality for human health in residential buildings

PDF

Impacts of building performance on occupants' work productivity: a post occupancy evaluation study

PDF

A framework for comprehensive assessment of resilience and other dimensions of asset management in metropolis-scale transport systems

PDF

Behavioral form finding using multi-agent systems: a computational methodology for combining generative design with environmental and structural analysis in architectural design

PDF

Building occupancy modeling and occupancy-loads relationships for building heating/cooling energy efficiency

PDF

Semantic modeling of outdoor scenes for the creation of virtual environments and simulations

PDF

User-centric smart sensing for non-intrusive electricity consumption disaggregation in buildings

PDF

Learning personal thermal comfort and integrating personal comfort requirements into HVAC system control loop

PDF

Applications of mindfulness toward promoting psychosocial health in diverse communities

PDF

Investigation of health system performance: effects of integrated triple element method of high reliability, patient safety, and care coordination

PDF

Whole day workload, activity engagement, and psychological state

PDF

Human-environmental interaction: potential use of pupil size for office lighting controls

$Chemical and toxicological characteristics and historical trends of size-fractioned particulate matter from traffic-related emissions in Los Angeles$

PDF

Chemical and toxicological characteristics and historical trends of size-fractioned particulate matter from traffic-related emissions in Los Angeles

Asset Metadata

Creator Awada, Mohamad (author)

Core Title Towards health-conscious spaces: building for human well-being and performance

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Civil Engineering

Degree Conferral Date 2024-08

Publication Date 07/12/2024

Defense Date 07/11/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag built environments,health,indoor environmental quality,OAI-PMH Harvest,Physiology,Psychology,Stress

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Becerik-Gerber, Burcin (committee chair), Lucas, Gale (committee member), Roll, Shawn (committee member), Soibelman, Lucio (committee member)

Creator Email awadam@usc.edu,mohamadawada14@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113997NB4

Unique identifier UC113997NB4

Identifier etd-AwadaMoham-13223.pdf (filename)

Legacy Identifier etd-AwadaMoham-13223

Document Type Thesis

Format theses (aat)

Rights Awada, Mohamad

Internet Media Type application/pdf

Type texts

Source 20240712-usctheses-batch-1181 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu