You are on page 1of 99

Exploring Tilt in Esports Gaming:

Combining Physiological and


Data-driven Approaches

Je Seok Lee
Ph.D. Thesis Defense
Tilt
What is tilt?
Player Experience of Esports
▪ Johnson et al. (2015):
▫ Esports games showed lower scores than action, adventure, or roleplaying
game (RPG) genres in player-needs satisfaction
▫ MOBA genre showed higher levels of frustration and challenge, but lower
immersion than all the other genres
▫ Overall, the play experience of esports includes several types of negative
emotions compared to games of other genres.
Player Experience of Esports
▪ Competition: esports games are designed to be competitive in nature,
which is one factor that engages players by stimulating their desire to win
(Lee & Schoenstedt, 2016)
▪ Overheated competition generates social and emotional challenges
▫ Excessive playtime
▫ Toxicity and violence
Fine line between engagement and addiction
▪ Esports requires certain level of engagement to
accomplish goals
▫ Grinding for a better performance and higher rank
▫ Studying to understand the game mechanics
▫ Group training for team coordination

▪ But players sometimes fall into “Tunnel vision”


▫ Losing self-control
▫ Pulled into excessive playtime

▪ Hard to draw a line between those but there are


certain triggers
Research Purpose
▪ To explore the experience of esports players during stressful moments of
gameplay
▪ To find possible triggers of tilt (stress events)
▪ Finding physiological evidence of tilt by measuring physiological
response of players to the stress triggers
▪ Compare other contexts of gameplay that would impact the stress while
playing
▫ i.e. player groups, event types, etc.
Research Questions
Research Questions
▪ RQ1: How do individual players’ physiological responses change before and
after stressful events while playing esports games?

Stress event
In-game
timeline

�� �� 🙂 Emotion
(Facial expression, ±3 seconds)
❤ Heart rate (HR and HRV)

-30s Event +30s


Research Questions
▪ RQ1: How do individual players’ physiological responses change before and after
stressful events while playing esports games?
▫ 1-1. How do changes in physiological responses vary across different levels
(game skill level and team cohesion level) of players?

Team cohesion level ▪ Team cohesion level: Team vs Solo


▪ Game skill level: Expert vs Novice
Team Expert Solo Expert
Game
skill
level
Team Novice Solo Novice
Research Questions
▪ RQ1: How do individual players’ physiological responses change before and after
stressful events while playing esports games?
▫ 1-2. How do changes in physiological responses vary across different types
of stress events?
▫ Event categories: Team, teammate, enemy, myself
Research Questions
▪ RQ1: How do individual players’ physiological responses change before and after
stressful events while playing esports games?
▫ 1-3. How do changes in physiological responses vary by different contexts of
the game?
▫ 1-3-1. different phases of the game?
▫ 1-3-2. the number of games played in a series of matches?
▫ 1-3-3. the result of the game?
▫ 1-3-4. the result of the previous game?
▫ 1-3-5. the closeness of the match result?

Research Questions
▪ RQ2: What is the pattern between a player’s physiological responses and gaming
performance?
▫ Performance variables: Kill/death/assist ratio, total gold earned, and total
damage dealt
▫ Summative data of an entire match
▫ 2-1. How do patterns vary across different levels (game skill level and team
cohesion level) of players?
▫ 2-2. How do patterns vary by the result of the game (won vs lost)?

Methods
Methods
▪ Target game: League of Legends (2011)
▫ 5v5 multiplayer esports game
▫ MOBA (Multiplayer Online Battle Arena) genre
▫ Biggest esports game for years (higher
viewership than NBA and NFL)
▫ Open data API and third party apps
▫ Many past studies in esports research
Participants
▪ Recruitment criteria and eligibility
▫ Participant pool: Students from UC wide campuses, high school students
from NASEF (North America Scholastic Esports Federation) league
▫ Emails and postings in Discord servers
▫ Skill level: Rank in League of Legends
▫ Experts: Gold 3 and above - (Top 38.9%) N=119 Team Solo
▫ Novice: Bronze ~ Gold 4 (Top 38.9% ~ 95.7%)
Expert 30 29
▫ Team cohesion level:
▫ Team: Team of 5 players, played at least 10 games together Novice 30 30
▫ Solo: Individual participants, playing with/against strangers
Participants
▪ Recruitment
▫ 10 participants per sessions: 5v5 team game
▫ Remote study: All participants play at their own place
▫ Shipped heart rate monitors to participants’ place
▫ Rewards: A commercial heart rate monitor (MSRP: $70) + Amazon gift card +
Extra incentive if participated as a team
▫ Matchmaking: Research team made matchmaking based on their rank and
position (OP.gg).

Participants
▪ Demographics
▫ Female: 12.6, Male: 84%, Non-binary: 3.4% - simliar ratio with our past study
Adults: 75.6%, Minors: 24.4%
Rank: Bronze ~ Gold 4: 50%, Gold 3 ~ Challenger: 50%
Study Procedure - Overview
Measures and Data Points

Stress event Performance Heart rates Facial expression


Measures and Data Points
Stress events Performance Heart rates Facial expression

CapturEsports
Measures and Data Points
Stress events Performance Heart rates Facial expression

Riot Games API

Match log data

Uploads data
Play screen upon participant’s
recording request
Data timeline sync
among participants
In-game
log data
Researcher
CapturEsports database
(on ICS server)
User PC
Measures and Data Points
Stress events Performance Heart rates Facial expression

● Event Log
○ In-game events
(player kills, ability usage)
● Match Summary Log
○ Summative stats from API
● Match Timeline Log
○ API Data Snapshot of each min
● Keystroke Log
CapturEsports

In-Game Event Log Snippet


Measures and Data Points
Stress events Performance Heart rates Facial expression

● Play screen recording - participant’s view

CapturEsports
Measures and Data Points
Stress events Performance Heart rates Facial expression

Finding stress events


{
{
"type": "BUILDING_KILL",
"type": "CHAMPION_KILL",
"timestamp": 823606,
"timestamp": 129014,
"position": {
● Event logs+Match timeline "position": {
"x": 981,
"x": 6146,
○ Player kill "y": 5820
"y": 10441
},
○ Building destroyed },
"killerId": 8,
"killerId": 9,
○ Monster stolen "victimId": 1,
"assistingParticipantIds": [
6
"assistingParticipantIds": [
],
8
"teamId": 100,
]
"buildingType": "TOWER_BUILDING",
}
"laneType": "TOP_LANE",
"towerType": "OUTER_TURRET"
}
Measures and Data Points
Stress events Performance Heart rates Facial expression

Finding stress events ● Teammate death


● My death
● Teamfight Loss
Data Player death ● Repeated teamfight loss
manipulation ● Gank
Log data ● Ambush
Video ● Solo kill
Raw event data
observation

Context of events
+
Communication Building ● Ally building destroyed
destroyed ● Enemy backdoor
Measures and Data Points
Stress events Performance Heart rates Facial expression

Categorizing stress events - brought from our research group’s past study result (Wu et al., 2021)

Team Teammate Enemy Myself

● Building destroyed ● Teammate death ● Backdoor ● My death


● Monster stolen ● Silence ● Ambush ● Solo-killed
● Teamfight loss ● Toxic Communication ● Instant kill
● Repeated teamfight loss ● Gank
Measures and Data Points
Stress events Performance Heart rates Facial expression

Measuring performance

● Summative data of an entire game


● Difficult to collect performance change
during specific periods of game
● From match summary logs
○ Total number of
■ Kill
■ Death
■ Assist
○ Total damage dealt
○ Total gold earned
Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Heart rates as a marker of mental stress?


Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Heart rates as a marker of stress: from public health and psychophysiology literatures
▫ Heart rate (HR) and heart rate variability (HRV) is widely used to measure stress level
in many fields, including clinical situations, and assumes that HRV is a noninvasive and
reliable index of stress (HG Kim et al., 2018)
▫ HRV evaluates the sympathovagal balance—the autonomic state resulting from
sympathetic and parasympathetic influences (Sztajzel, 2004)
▫ Even without active body movement from the user, extrinsic factors adjust HR and
HRV levels at rest and during times of stress (Anderson, 2020)
Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Heart rates as a marker of stress: from public health and psychophysiology literatures

Pollatos, O., Herbert, B. M., Matthias, E., &
Schandry, R. (2007). Heart rate response after
emotional picture presentation is modulated by
interoceptive awareness. International Journal
of Psychophysiology, 63(1), 117–124.
https://doi.org/10.1016/j.ijpsycho.2006.09.003
Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Heart rates as a marker of stress: from ‘less active’ sports


Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Heart rate variability: the fluctuation in the length of


heartbeat intervals
▫ Heart rate: Beats per minute, averaged over time
▫ HR does not reveal that the behavior of the heartbeat is
irregular when measured on a beat-to-beat basis
(McCraty & Shaffer, 2015)
▫ Lower HRV correlates with impaired regulatory and
homeostatic autonomic nervous system functions,
reducing the body’s ability to recover from internal and
external stress factors (HG Kim et al., 2018)
Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Heart rate monitors


▫ Commercial chest band heart rate monitors
(Garmin HRM Dual, Polar H10)
▫ More accurate than wrist band type monitors (e.g. Apple
watch)
▫ Validated/widely used for research purpose (Chow H, Yang
C, 2020)
▫ Connected to participant’s smartphone app to send data
Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Emotion analyzed from facial expressions


▫ Recorded face from Discord video chat
▫ Videos sliced into images
▫ OpenCV library in Python3
▫ Saved 4 frames per second
▫ Every 0.25 second
Measures and Data Points
Stress events Performance Heart rates Facial expression

▪ Microsoft Azure’s Face API *Identifiable images are masked

▫ AI service that recognizes emotion from


human faces
▫ Transmit images one by one to Azure
▫ 8 different categories of emotion
▫ Anger, Contempt, Disgust, Fear,
Happiness, Neutral, Sadness, Surprise
▫ Ratio of emotions, from 0 to 1.
Data cleaning
▪ Individual participant timeline
▫ Integrated all data (stress events,
HR, HRV, and emotion) into a
single timeline
▫ Time synchronization among
variables
▫ Standard time = In-game time
▫ Every 0.25 seconds
Data cleaning
▪ Stress event data sheet
▫ Collected all event data from individual
timelines
▫ Each row is a single event
▫ Event type, HR/HRV ±30s,
facial expression ±3s
▫ Included other context data
▫ Participant group (by skill/team
cohesion level)
▫ Game result, time marker
▫ Total 10718 cases
Data analysis
Stress
event

▪ RQ 1 - HR/HRV change caused by stress events


▫ Averaged the HR and HRV level by seconds
▫ To see overall trends before and after stress events
▫ Compared mean of inner period vs outer period
-30s +30s
▫ Compared mean of pre-event vs post-event period
▫ Compared between different groups

-30s +30s
Data analysis
▪ RQ 1 - HR/HRV change caused by stress events Outer period Inner period Outer period

▫ Averaged the HR and HRV level by seconds


▫ Compared mean of inner period vs outer period
▫ To find the impact of the stress event
▫ F-test to check equal variances.
▫ T-tests for two-sample comparison. Welch’s two-sample t-test
was used for non-parametric tests.
-30s +30s
▫ Compared mean of pre-event vs post-event period
▫ Compared between different groups
Data analysis
▪ RQ 1 - HR/HRV change caused by stress events Pre-event Post-event

▫ Averaged the HR and HRV level by seconds


▫ Compared mean of inner period vs outer period
▫ Compared mean of pre-event vs post-event period
▫ To find how participants recovered from stress after
the event
▫ F-test to check equal variances. -30s +30s

▫ T-tests for two-sample comparison. Welch’s two-sample t-test


was used for non-parametric tests.

▫ Compared between different groups


Data analysis
▪ RQ 1 - HR/HRV change caused by stress events
Group A
▫ Averaged the HR and HRV level by seconds
▫ Compared mean of inner period vs outer period
Group B
▫ Compared mean of pre-event vs post-event period
-30s +30s
▫ Compared between different groups
▫ To see differences between user groups, event categories, and game contexts
▫ Same T-tests on each group between inner-outer period and pre-post period.
▫ F-test to check equal variances.
▫ T-tests for two-sample comparison. Welch’s two-sample t-test was used for
non-parametric tests.
▫ ANOVA for multiple comparisons. Tukey HSD for post hoc test.

Data analysis
▪ RQ 1 - Facial expression change caused by stress events

-3 ~ -2 s

-3s -2.75s -2.50s -2.25s

Neutral Happiness Neutral Neutral

Neutral: 75% x 6 seconds


Happiness: 25% (3 seconds before and after events)
Data analysis
▪ RQ 2 - Correlation between performance and stress level
▫ Performance/stress level over an entire match (not time series data)
▫ A Shapiro-Wilk test and a Q-Q plot were used to determine the equal
variance.
▫ Spearman’s rank correlation coefficient was used for all variables (non-equal
variances).

▪ All statistical analysis used R 4.1.0 on RStudio 1.4.1717


▪ Plots were drawn with the ggplot2 and ggpubr library in RStudio.
Results
RQ1: Emotional Change before and after Stress Events

▪ All HR and HRV before vs after Event

HR HRV

N Mean SD N Mean SD

30 seconds before 10717 99.08 17.06 10717 20.24 20.98


and after stress events

In-game average 318 97.81 15.30 315 22.78 15.31

Baseline 106 81.53 12.94 105 34.49 34.75


RQ1: Emotional Change before and after Stress Events

▪ All HR and HRV before vs after Event

Table 3. Result of one sample t-test between stress event and baseline, and event and in-game average.

HR HRV
t df P-value t df P-value
Event ~ Baseline 387.5 59 <0.001** 27.95 59 <0.001**

Event ~ In-game -355.13 59 <0.001** -63.68 59 <0.001**


average

Mean HR and HRV are significantly different from


the baseline and in-game average HR and HRV
=> During the stress period, players showed higher stress
level than their baseline status and overall gaming status
RQ1: Emotional Change before and after Stress Events

▪ All HR and HRV before vs after Event

Table 4. Result of two-sample t-test of HR and HRV between inner period and outer period

F df P-value Test chosen t df P-value Mean of Mean of


Outer Inner
period period
HR 1.1951 28 0.6312 Two sample -5.888 58 <0.001** 98.85 99.28
T-test
HRV 0.67303 28 0.2954 Two sample 4.1754 58 <0.001** 20.39 20.10
T-test

Mean HR was significantly higher, and mean HRV was


significantly lower during the inner period of stress.
=> Players were more stressed during the stress event
period
RQ1: Emotional Change before and after Stress Events

▪ All HR and HRV before vs after Event

Table 4. Result of two-sample t-test of HR and HRV between inner period and outer period

F df P-value Test chosen t df P-value Mean of Mean of


Pre-event Post-event
HR 0.49469 29 0.062 Welch 7.848 58 <0.001** 98.82 99.32
Two sample
T-test
HRV 0.58054 29 0.149 Two sample 3.047 58 0.003* 20.13 20.35
Both mean HR and HRV T-test
were significantly higher after the event
=> After the stress event ended, arousal level remained higher (HR)
but they could recover from stress (higher HRV)
=> They cannot play after death for a certain amount of time until
they respawn, which gives them a break to breathe and take a rest
RQ1: Emotional Change before and after Stress Events

▪ Emotional change before vs after Event


(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 80.8 81.1 80.8 79.8 78.8 78.3
Happiness 12.4 12.5 12.7 13.4 14.5 15.6
Surprise 4.5 4.3 4.3 4.5 4.1 3.7
Sadness 1.5 1.5 1.6 1.7 1.8 1.7
Contempt 0.5 0.4 0.4 0.4 0.5 0.5
Disgust 0.1 0.1 0.1 0.1 0.1 0.1
Anger 0.2 0.1 0.1 0.1 0.2 0.1
Fear 0 0.0 0.0 0.0 0.0 0.0
N of cases = 10717 Log-scaled graph of emotions
Table 6. The ratio of emotion during 3 seconds before and after all stress events.
Emotions were observed 4 times per second.
RQ1: Emotional Change before and after Stress Events

▪ Emotional change before vs after Event


▫ Neutral is a dominating emotion but drops (-2.5%)
▫ Happiness constantly increases (+3.2%)
▫ Inappropriate to interpret happiness literally as
happiness
▫ AI recognizes smiles and laughters as happiness
▫ But it many cases, players' laughter shows mixed
emotions of embarrassment, dismay, and
astonishment
▫ Closer to a higher arousal level, or more actively Log-scaled graph of emotions
showing emotions
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Solo players vs Team players

HR HRV
Group N Mean Min Max SD N Mean Min Max SD
Solo 5152 100.23 99.66 100.86 0.35 5152 18.16 17.49 18.77 0.31
Team 5565 98.01 97.35 98.48 0.34 5565 22.33 21.65 23.20 0.37

F df P-value Test chosen t df P-value Mean of Mean of


Team Solo
Players Players
HR 0.941 59 0.819 Two sample -34.50 118 <0.001** 98.00 100.22
T-test
HRV 1.397 59 0.201 Two sample 66.43 118 <0.001** 22.32 18.16
T-test

Solo players showed significantly higher HR and lower HRV


=> Solo players’ stress level was higher than team players’
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Solo players vs Team players

F df P-value Test Chosen t df P-value Mean of Mean of


Outer Inner
Period Period
Solo HR 0.985 28 0.9727 Two sample -5.778 58 <0.001** 100.00 100.43
T-test
Solo HRV 0.587 28 0.1609 Two sample 2.616 58 0.011* 18.26 18.06
T-test
Team HR 1.442 28 0.3265 Two sample -5.884 58 <0.001** 97.78 98.20
T-test
Team HRV 0.994 28 0.9915 Two sample 4.681 58 <0.001** 22.52 22.14
T-test

Both groups showed the higher stress level during the inner period
=> Stress level peaked during the event
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Solo players vs Team players

F df P-value Test chosen t df P-value Mean of Mean of


Pre-event Post-event
Solo 1.851 29 0.102 Two sample -7.744 58 <0.001** 99.97 100.47
HR T-test
Solo 1.000 29 0.999 Two sample -3.457 58 0.001* 18.26 18.06
HRV T-test
Team 2.099 29 0.050 Two sample -7.829 58 <0.001** 97.75 98.25
HR T-test
Team 1.831 29 0.109 Two sample -2.139 58 0.036* 22.22 22.42
HRV T-test

In the post-event period, Solo’s HRV is lower and Team’s HRV is higher
=> Team players could recover from the stress easily, while Solo
players could not.
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Solo players vs Team players


Solo (Case N=5152)
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 81.6 81.8 81.6 81.1 80.0 79.0
Happiness 11.2 11.1 11.2 11.6 12.8 14.5

Team (Case N=5565)


(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 80.0 80.4 80.0 78.5 77.7 77.6
Happiness 13.5 13.7 14.1 15.0 16.1 16.5

Team players showed higher happiness overall


=> They were more open to show their emotion to teammates, and able
to ‘laugh it off’ when a stressful situation happens.
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Expert players vs Novice players


HR HRV
Group N Mean Min Max SD N Mean Min Max SD
Expert 5202 99.70 98.88 100.26 0.425 5202 18.99 18.34 19.70 0.352
Novice 5515 98.44 98.03 98.97 0.284 5515 21.52 20.87 22.55 0.371

F df P-value Test chosen t df P-value Mean of Mean of


Experts Novices
HR 2.237 59 0.002 Welch Two 19.02 102. <0.001** 99.69 98.43
Sample t-test 96
HRV 0.900 59 0.689 Two Sample -38.27 118 <0.001** 18.98 21.52
t-test

Expert players showed significantly higher HR and lower HRV


=> Expert players’ stress level was higher than team players’
=> Experts know much better about the game, more things to take care
about, and become angrier because they exactly know what happened
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Expert players vs Novice players


F df P-value Test chosen t df P-value Mean of Mean of
Outer Inner
period period
Expert HR 1.763 28 0.130 Two sample -4.898 58 <0.001** 99.46 99.91
T-test 7
Expert 0.901 28 0.786 Two sample 4.3966 58 <0.001** 19.16 18.81
HRV T-test
Novice HR 0.621 28 0.209 Two sample -7.396 58 <0.001** 98.23 98.62
T-test 2
Novice HRV 1.767 28 0.128 Two sample 2.5646 58 0.012* 21.64 21.40
T-test

Both groups showed the higher stress level during the inner period
=> Stress level peaked during the event
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Expert players vs Novice players


F df P-value Test chosen t df P-value Mean of Mean of
Pre-event Post-event
Expert HR 3.9552 29 <0.001 Welch Two -9.048 42.78 <0.001** 99.37 100.02
Sample t-test
Expert 3.8442 29 <0.001 Welch Two -0.085 43.13 0.932 18.98 18.99
HRV Sample t-test
Novice 0.9791 29 0.955 Two Sample -5.942 57.99 <0.001** 98.26 98.62
HR t-test
Novice HRV 0.6316 29 0.222 Two Sample -5.887 58 <0.001** 21.29 21.74
t-test

Novice’s HRV significantly increased in the post-event period,


while expert’s did not.
=> Experts could not recover as much as novices did.
1-1: How Do Changes in Physiological Responses Vary across Different
Levels (Team Cohesion Level and Game Skill Level) of Players?

▪ Expert players vs Novice players


Expert (Case N=5202) Experts show more various
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s emotion, especially higher
Neutral 79.7 80.0 79.8 78.7 77.8 77.2 happiness and surprise
Happiness 13.2 13.2 13.6 14.1 15.4 16.4
Surprise 5.7 5.4 5.1 5.5 5.1 4.7 Novices show more sadness
Sadness 0.8 0.9 1.0 1.1 1.0 1.1
than experts (dull faces)

Novice (Case N=5515) => Experts understood what


(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s had happened and could react
Neutral 81.9 82.2 81.7 80.8 79.8 79.4 with more emotions while
Happiness 11.7 11.7 11.9 12.6 13.7 14.7 novices did not perfectly
Surprise 3.5 3.3 3.4 3.6 3.2 2.8 perceive what was happening,
Sadness 2.1 2.1 2.3 2.2 2.5 2.4 resulting in less variability in
their emotions.
RQ 1-2: How do Changes in Physiological Responses Vary across Different
Types of Stress Events?

▪ Compare different stress event categories

ANOVA result shows HR and HRV are different from each other (p<0.05), except the HRV of Enemy
category and Myself Category (p=0.971)
RQ 1-2: How do Changes in Physiological Responses Vary across Different
Types of Stress Events?

▪ Compare different stress event categories


Mean of Mean of
F df P-value t df P-value Outer Inner
Period Period
Team HR 0.660 28 0.273 -10.367 58 <0.001** 99.66 99.99
Team HRV 1.275 28 0.512 0.491 58 0.625 19.65 19.60

Teammate HR 2.401 28 0.021 -2.84 47.36 0.007* 98.14 98.41

Teammate HRV 0.908 28 0.802 4.459 58 <0.001** 20.72 20.41

Enemy HR 0.35 28 0.006 -5.307 49.27 <0.001** 99.64 100.96

Enemy HRV 0.60 28 0.18 4.09 58 <0.001** 21.64 20.50


Myself HR 0.936 28 0.864 -4.29 58 <0.001** 98.53 99.87

Myself HRV 0.42 28 0.022 4.553 51.66 <0.001** 21.62 20.64

T-test shows peak in HR and valley in HRV during the inner period,
except the HRV of Team category
=> Events in Team category did not significantly impact the stress level
RQ 1-2: How do Changes in Physiological Responses Vary across Different
Types of Stress Events?

▪ Compare different stress event categories


F df P-value t df P-value Pre-event Post-event

Team HR 0.758 29 0.461 2.325 58.00 0.023* 99.89 99.77

Team HRV 0.924 29 0.833 -5.996 58.00 <0.001** 19.39 19.86


Teammate HR 24.857 29 <0.001 -13.645 31.33 <0.001** 97.94 98.60
Teammate HRV 1.460 29 0.313 1.672 58 0.099 20.62 20.49
Enemy HR 0.635 29 0.227 -7.948 58 <0.001** 99.48 101.15
Enemy HRV 1.039 29 0.9177 -4.4013 58 <0.001** 20.44 21.65
Myself HR 1.259 29 0.538 -11.526 58 <0.001** 98.07 100.36
Myself HRV 2.849 29 0.006 -2.240 47.12 0.029* 20.84 21.38

All variables showed the same direction except the Team HR and Teammate HRV
=> Like the Team HRV did not significantly decrease during the inner period,
Team HR did not increase after the stress events
=> Teammate HRV did not increase after the stress events, which implies that teammates
are the stress factor that players feel hard to recover
=> Or, they did not have a chance to cool down because teammate events are not linked to the player’s death and respawn.
RQ 1-2: How do Changes in Physiological Responses Vary across Different
Types of Stress Events?

▪ Compare different stress event categories

After enemy and myself events, which showed higher


peaks in HR and HRV, the ratio of happiness increased
dramatically (12.3%p and 18.6%p, respectively)

=> Events in the enemy and myself categories are more


directly linked to a player’s death, which causes a more
dynamic emotional expression.
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Time phase
▫ 1-3-1. By different phases of the game?
▫ 1-3-2. By the number of games played in a series of matches?

▪ Match result
▫ 1-3-3. By the result of the game?
▫ 1-3-4. By the result of the previous game?
▫ 1-3-5. By the closeness of the match result?

=> Addresses if length of playtime and result of the game would impact tilt.
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ RQ 1-3-1. Events in the first 10 minutes vs Later period of the game


▫ 10 minute is
▫ when the early lane fights wrap up (first tower kill)
▫ when the important neutral monsters generates (at 8 min)
▫ the threshold that Riot API uses to distinguish between the early and late game
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Events in the first 10 minutes vs Later period

Mean of Mean of
F df P-value Test chosen t df P-value Early Late
Phase Phase
Two sample
HR 0.809 59 0.418 23.78 118.00 <0.001** 99.34 97.68
T-test
Welch
HRV 0.384 59 <0.001 Two-sample -24.06 98.53 <0.001** 19.95 21.70
T-test

Mean HR was significantly higher and mean HRV was


significantly lower in the early phase of the game
=> Higher stress level in the late game phase
=> More intense events are in the late phase (teamfights,
objective fights)
=> Arousal level naturally goes up constantly over the match
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Events in the first 10 minutes vs Later period


Mean of Mean of
F df P-value Test chosen t df P-value Outer Inner
period period
Two sample
Late HR 1.757 28 0.133 -4.72 58 <0.001** 99.15 99.53
T-test
Two sample
Late HRV 0.912 28 0.811 3.568 58 0.001* 20.08 19.83
T-test
Welch
Early HR 0.095 28 <0.001 Two-sample -10.098 36.02 <0.001** 97.36 97.99
T-test
Two sample
Early HRV 0.593 28 0.169 4.525 58 <0.001* 21.95 21.46
T-test

HR was higher and HRV was lower during the inner periods.
Players’ stress levels were significantly higher during stress
events compared to the peripheral periods.
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Events in the first 10 minutes vs Later period


Mean of Mean of
F df P-value Test chosen t df P-value
Pre-event Post-event
Welch
Late HR 3.7505 29 <0.001 Two-sample -9.9111 43.438 <0.001** 99.06 99.63
T-test
Two sample
Late HRV 1.7842 29 0.125 -2.5477 58 0.014* 19.86 20.05
T-test
Two sample
Early HR 0.51409 29 0.078 -1.9951 58 0.051 97.58 97.78
T-test
Two sample
Early HRV 0.84314 29 0.649 -3.2742 58 0.002* 21.51 21.88
T-test

The mean HR level of early events was not significantly higher


in the post-event period
=> Players come back to the normal arousal level more quickly
in the early phase of the game
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Events in the first 10 minutes vs Later period Players remained calm in the
early game phase (higher neutral
Late (Case N=8984)
and lower happiness)
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 79.8 80.1 79.8 78.8 77.8 77.2 In the later phase, players were
Happiness 13.2 13.2 13.5 14.1 15.3 16.4 bolder in expressing their
Surprise 4.6 4.4 4.3 4.6 4.2 3.8 emotions (overall higher level in
Sadness 1.5 1.6 1.7 1.8 2.0 1.9 happiness)

Early (Case N=1733)


(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 85.9 86.1 85.8 84.7 83.7 83.4
Happiness 8.2 8.4 8.9 9.5 10.7 11.4
Surprise 4.2 3.9 3.8 4.4 3.9 3.5
Sadness 1.2 1.2 1.1 0.9 1.0 1.1
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ RQ 1-3-2. Number of Games Played in a Series of Games


▫ Participants were asked to play three matches in a session
▫ Would they show higher stress levels in the repeated play?
▫ 1st match vs 2nd match vs 3rd match
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?
▪ Number of Games Played in a Series of Games
▫ Highest stress level in the first match, and slightly higher HR
level in the second match than third match.
=> Due to the competition setting of the study session, they
were nervous in the first match, and then got relaxed in the
later matches
=> And/or, players’ threshold of stimulation goes down after
matches.
=> Similar to the behavior pattern found in gambling addiction
▫ When compared 2nd and 3rd matches, HR level was lower in
the 3rd match but HRV level was also lower (p=0.036).
▫ Players became slower in adjusting their stress level in the 3rd
match.
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Number of Games Played in a Series of Games


F df P-value t df P-value Pre-event Post-event

1st HR 0.888 29.00 0.750 -5.091 58 <0.001** 101.88 102.24

HRV 0.854 29.00 0.673 -3.325 58 0.002* 18.54 18.81

2nd HR 2.322 29.00 0.027 -8.157 50.07 <0.001** 97.33 97.86

HRV 0.843 29.00 0.649 -3.274 58 0.002* 21.51 21.88

3rd HR 6.311 29.00 <0.001 -10.846 37.97 <0.001** 97.02 97.65

HRV 1.401 29.00 0.369 -1.256 58 0.214 21.44 21.58

It’s again found that players were not able to recover in the
third match as much as in the other matches.
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Number of Games Played in a Series of Games


The happiness scale increased
1st (Case N=3817)
4.2% in the first match, while it
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
increased 2.3% and 2.8% in the
Neutral 81.0 81.2 81.2 79.8 78.4 77.7
following matches.
Happiness 11.9 11.9 12.0 13.2 14.9 16.1

Players showed greater


2nd (Case N=3189)
emotional change in the first
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s match compared to the other
Neutral 79.9 80.2 79.8 78.8 78.5 78.7 two matches.
Happiness 12.8 13.2 13.6 14.1 14.6 15.1
Could be related to the lowered
3rd (Case N=3711)
threshold of stimulation
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 81.5 81.8 81.3 80.7 79.5 78.4
Happiness 12.6 12.3 12.6 12.9 14.1 15.4
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ RQ 1-3-3. Result of the Game (Won vs Lost)


Mean of Mean of
F df P-value Test chosen t df P-value Won Lost
Game Game
Welch
HR 4.7431 59 <0.001 Two-sample -40.29 82.82 <0.001 96.98 100.17
T-test
Welch
HRV 2.7894 59 <0.001 Two-sample 53.23 96.49 <0.001 22.76 18.95
T-test

Significantly higher HR and lower HRV in lost matches


=> Higher stress level while losing.

Same trends when compared inner vs outer period (higher stress


during the inner period)
and pre-event vs post-event (higher HR/higher HRV in the post period)
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Result of the Game (Won vs Lost)

Win (Case N=3693) Winning players showed higher


(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s happiness, and more increase in
Neutral 80.8 81.2 81.2 79.6 78.3 78.0 happiness during the stress event.
Happiness 11.9 11.8 12.0 13.6 15.1 16.0
Surprise 5.0 4.8 4.6 4.6 4.3 3.7
Sadness 1.4 1.3 1.5 1.6 1.4 1.3

Lost (Case N=7024)


(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 80.8 81.0 80.5 79.9 79.1 78.4
Happiness 12.8 12.8 13.2 13.2 14.2 15.3
Surprise 4.3 4.0 4.0 4.5 4.0 3.7
Sadness 1.5 1.6 1.7 1.8 2.0 2.0
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ RQ 1-3-4 Result of the Previous Game (Won vs Lost)

Mean: Mean:
F df P-value Test chosen t df P-value Won in the Lost in the
last game last game
Two sample
HR 1.388 59 0.21 -48.39 118 <0.001** 95.45 98.85
T-test
Two sample
HRV 1.151 59 0.59 15.27 118 <0.001** 21.84 20.62
T-test

Players showed higher stress level when lost in the previous


match
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Result of the Previous Game (Won vs Lost)


Mean of Mean of
F df P-value Test chosen t df P-value Outer Inner
period period
Welch
Last Win
2.110 28 0.048 Two-sample -3.814 49.29 <0.001** 95.27 95.64
HR
T-test
Welch
Last Win
0.401 28 0.017 Two-sample 5.365 51.14 <0.001** 22.10 21.59
HRV
T-test
Last Lost Two sample
1.375 28 0.393 -5.331 58 <0.001** 98.65 99.05
HR T-test
Last Lost Two sample
0.889 28 0.756 1.408 58 0.16 20.71 20.56
HRV T-test

Players could not recover well from the stress event if they lost
in the previous match
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?

▪ Result of the Previous Game (Won vs Lost)


Last Win (Case N=2888)
Similar to the previous sub-question,
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
winning players displayed higher
Neutral 78.5 78.7 78.1 77.6 76.8 76.3
happiness level overall, and higher
Happiness 14.5 14.5 15.1 15.4 16.1 17.4 change in happiness after a stress
Surprise 4.6 4.5 4.3 4.3 4.1 3.4 event when they had won the
Sadness 1.7 1.7 1.8 2.0 2.2 2.1 previous game.
Last Lost (Case N=4012)
(%) -3~-2s -2~-1s -1~0s 0~1s 1~2s 2~3s
Neutral 82.4 82.7 82.3 81.4 80.6 80.3
Happiness 11.3 11.4 11.6 12.0 13.0 13.7
Surprise 4.0 3.9 3.9 4.2 3.8 3.5
Sadness 1.5 1.5 1.7 1.8 1.8 1.8
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?
▪ RQ 1-3-5: Closeness of the match result
▫ Close game vs One-sided game vs Normal difference
▫ Delta number of towers destroyed:
▫ 0-5: close game
▫ 6-8: normal game
▫ 9-12: one-sided game
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?
▪ Closeness of the match result
▫ HR level: Normal>One-sided>Close (p<0.05)
▫ HRV level: Normal<Close<One-sided (p<0.05)
▪ Similar trends in peaks and valleys during the inner
period
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?
▪ Closeness of the match result
P-valu
F df t df P-value Pre-event Post-event
e
Close HR 2.060 29 0.056 -10.15 58 <0.001** 96.07 96.88
HR
1.759 29 0.134 -0.191 58 0.850 20.25 20.28
V
Normal HR 1.508 29 0.275 -7.566 58 <0.001** 100.52 100.98
HR
0.558 29 0.122 -1.910 58 0.061 19.16 19.29
V
One-
HR 2.290 29 0.029 -6.918 50.27 <0.001** 96.84 97.31
sided
HR
2.084 29 0.052 -3.771 58 <0.001** 21.64 22.07
V
Players’ HRV levels didnot increase significantly after stress events in
close and normal games
=> Players could not recover well from stress in close or normal games
=> Players lose tension when they are defeated by a huge difference
RQ 1-3: How Do Changes in Physiological Responses Vary by the
Context of a Game?
▪ Closeness of the match result

Players in one-sided games stayed


calm or insensitive after stress events,
showing the highest score and the
smallest decrease in neutral emotion
(-1.4%p)

In close games, players showed the


largest change in happiness (+6.0%p),
which reflects their immersion in the
game.
RQ 2: What is the pattern between a player’s physiological responses
and gaming performance?
▪ 2-1. How do patterns vary across different levels (game skill level and team cohesion
level) of players?
▪ 2-2. How do patterns vary by the result of the game (won vs lost)?

▪ Shapiro-Wilk normality test and Q-Q plots to determine the equal variance
▪ All variables did not show equal variances, so used Spearman’s rank correlation
coefficient.
▪ Delta HR and HRV levels: Average in-game HR/HRV - baseline HR/HRV level
RQ 2: What is the pattern between a player’s physiological responses
and gaming performance?
▪ Overall correlation between HR/HRV and Performance
Correlation Method Correlation
P-value
Efficient

KDA Delta HR Spearman -0.022 0.77

Total Gold Delta HR Spearman -0.11 0.13

Total Damage Delta HR Spearman -0.16 0.032*

KDA Delta HRV Spearman -0.016 0.82

Total Gold Delta HRV Spearman 0.097 0.18

Total Damage Delta HRV Spearman 0.13 0.067

HR levels are weakly but negatively correlated with the total amount of damage
RQ 2: What is the pattern between a player’s physiological responses
and gaming performance? Correlation Efficient P-value

▪ 2-1. How do patterns vary across different Expert KDA Delta HR -0.073 0.5

levels of players - Expert vs Novice Total Gold Delta HR -0.057 0.6

Total Damage Delta HR -0.0031 0.98

KDA Delta HRV 0.18 0.096

Total Gold Delta HRV 0.18 0.11

Total Damage Delta HRV 0.13 0.23

Novice KDA Delta HR 0.032 0.75

Total Gold Delta HR -0.17 0.092

Total Damage Delta HR -0.3 0.0021*

KDA Delta HRV -0.18 0.071

Total Gold Delta HRV 0.021 0.83


(Novice group)
Total Damage Delta HRV 0.15 0.14

A stronger correlation between total damage and HR is found in the Novice group.
Novice players would affected more by stress in terms of the amount of damage they deal.
RQ 2: What is the pattern between a player’s physiological responses
and gaming performance?
Correlation
▪ 2-1. How do patterns vary across different levels of players - Efficient
P-value

Team vs solo Team KDA Delta HR -0.32 0.011*


(Team group)
Total Gold Delta HR -0.26 0.035*

Total Damage Delta HR -0.24 0.053

KDA Delta HRV 0.12 0.36

Total Gold Delta HRV 0.069 0.59

Total Damage Delta HRV 0.033 0.8

Solo KDA Delta HR 0.13 0.16

(Solo group) Total Gold Delta HR -0.029 0.75

Total Damage Delta HR -0.1 0.25

KDA Delta HRV -0.09 0.32

Total Gold Delta HRV 0.16 0.071

Total Damage Delta HRV 0.23 0.01*


RQ 2: What is the pattern between a player’s physiological responses
and gaming performance?
Correlation
▪ 2-2. How do patterns vary by the match result - Won vs Lost Efficient
P-value

Won KDA Delta HR -0.029 0.78


(Won group)
Total Gold Delta HR -0.21 0.047*

Total Damage Delta HR -0.3 0.004*

KDA Delta HRV 0.023 0.83

Total Gold Delta HRV 0.21 0.051

Total Damage Delta HRV 0.21 0.046*

Lost KDA Delta HR 0.022 0.83

Total Gold Delta HR -0.022 0.98

Total Damage Delta HR -0.0033 0.97

KDA Delta HRV 0.0033 0.75

Total Gold Delta HRV 0.043 0.67

Total Damage Delta HRV 0.077 0.44


Discussion
Reflection of Past Studies
▪ Novel method incorporating diverse data types
▪ Verifies past studies that used survey, interview, or self-reported index
▫ From Shores et al. (2014): “players in the more competitive ranked matches were
associated with higher toxicity indexes than players in normal matches.”
▫ Higher stress level of heart rates in expert players
▫ More dramatic changes in facial expression from expert players
▫ From Kou and Gui’s (2020): “personal performances lower than expectation could
trigger negative emotions such as frustration and anger”
▫ Replicated the highest stress response during Myself events
▫ From Wu et al. (2021): Triggers of tilt (teammate, myself, enemy, and other players)
▫ Demonstrated different physiological responses from different triggers
Validity of Emotion Measurement
▪ Heart rate (HR) - serves as an arousal level
▪ Heart rate variability (HRV) - more direct marker of stress level
▪ Facial expression - supplement measure for variability of emotion
▪ HR/HRV and Performance
▫ The correlations were less clear
▫ HR/HRV are better suited for measuring emotion during a shorter period of
time, not the entire match
Effect of the Gameplay Length
▪ Repetitive play: prominent factor of tilt proposed by Shores et al. (2014)
▪ Players were stressed most during the first match
▫ Nervous in the first match, due to the competition setting of the study
procedure
▪ Arousal level went down in the second and third match
▫ Relaxed after the first match
▫ Lowered threshold of stimulus
▫ Needs further investigation in a natural setting (playing normal ranked
games, no premade teams, etc.)
For Healthier Gaming Behavior
▪ Playing conditions that ameliorate tilt
▪ Team players > Solo players
▫ Having a regular team/friends to play together
▫ Structured play (i.e. school clubs)
▫ Socio-emotional learning
▪ Result of the (previous) match
▫ Different stress response by the result of the match
▫ Affect the ability to stop playing (Kou et al., 2020)
For Healthier Gaming Behavior
▪ Getting out of tunnel vision - possible solution of game addiction
▫ Raising understanding of the reasons for excessive playtime
▫ Wu et al. (2021): “tilt is an important construct to understand because it
represents the esports version of self-management.”
▫ Possible causes of tilt are introduced and tested with physiological measures
▫ Also demonstrates that higher stress levels would negatively affect the
performance of the player
▫ Insights into fostering a healthier gaming environment
Limitation and Future Work
Incomplete controls over participants
▪ Remote study
▪ Not in a perfect control for accurate physiological measurements
▫ Temperature/food intake/body posture, etc.
▪ Unstable network condition/lack of system resource for some users
▪ Resembles the real-world settings of playing
▫ Not sitting together with the teammates and opponents
▫ Familiar gaming gears and system settings
Emotion Analysis
▪ AI’s classification of emotion
▪ Interpreted in diverse ways according to the context
▪ Needs additional human observation and interpretation
Participants’ Attitude
▪ May not be serious as they play ranked games (no additional incentives)
▪ Participants acted too kind for strangers
▪ Needed more number of matches to observe streakiness of play
Future Works
▪ Other aspects of gaming
▫ i.e. positive events can be explored for flow in gaming
▪ Closer observation and interview to integrate qualitative analysis
▫ More detailed mental models while being tilted
▪ Design study
▫ Design implementation to be used while playing to help self-regulation of
players
▫ For professional training or self-regulation of general players
Q&A

You might also like