Why We Play Whom We Play: On the Correlation between Player Personality and Choice of Avatar

Abstract

While there have been many studies dealing with the effects of human personality on general gameplay or character creation, most prominently so in role-playing games, few seem to focus on the choice of avatar when players are presented a roster of options. Addressing competitive action games especially, this paper aims to find an answer to the questions of how we choose to insert ourselves into digital media and what the decision might say about us, on a quest to better understand the player and enhance the experience. Therefore, a survey was conducted, concentrating on said interrelationship by means of using League of Legends, Overwatch and Super Smash Bros. Ultimate as example games, and Lewis Goldberg’s Big Five model as the psychological foundation for determining each participant’s personality traits. The results show that there is indeed a
connection to be found between the choice made and the respective psychological profile, which even differs from title to title – most notably the correlation between a player’s neuroticism score and their favored League of Legends agent. With more data to back up the findings, one could potentially predict preferences for future players and utilize them as a tool for decision guidance or even construct entire prototypes on the premise of providing character-specific gameplay additions through player personality and avatar choice correlation alone.

Full text

WHY WE PLAY WHOM WE PLAY: ON THE CORRELATION BETWEEN
PLAYER PERSONALITY AND CHOICE OF AVATAR
Lukas Keuter
Emmanuel Guardiola
Cologne Game Lab
TH Köln
51063 Cologne,
Germany
E-mail: l-keuter@live.de
KEYWORDS
Avatar choice, Big Five, personality, player research.
ABSTRACT
While there have been many studies dealing with the effects
of human personality on general gameplay or character
creation, most prominently so in role-playing games, few
seem to focus on the choice of avatar when players are
presented a roster of options. Addressing competitive action
games especially, this paper aims to find an answer to the
questions of how we choose to insert ourselves into digital
media and what the decision might say about us, on a quest
to better understand the player and enhance the experience.
Therefore, a survey was conducted, concentrating on said
interrelationship by means of using League of Legends,
Overwatch and Super Smash Bros. Ultimate as example
games, and Lewis Goldberg’s Big Five model as the
psychological foundation for determining each participant’s
personality traits. The results show that there is indeed a
connection to be found between the choice made and the
respective psychological profile, which even differs from
title to title – most notably the correlation between a player’s
neuroticism score and their favored League of Legends
agent. With more data to back up the findings, one could
potentially predict preferences for future players and utilize
them as a tool for decision guidance or even construct entire
prototypes on the premise of providing character-specific
gameplay additions through player personality and avatar
choice correlation alone.
INTRODUCTION
Somewhat recently, a curious trend arose on online forums
like Reddit and video platforms like YouTube: “What Your
Main Says About You” – an entirely new subgenre of
content humorously depicting the stereotypical player of
each avatar or class in numerous character-based
videogames. A quick search reveals a multitude of creators
focusing their efforts on the supposed science behind our
choices, to a high degree of success and oftentimes millions
of clicks (ScottFalco 2018).
At the same time, on the academic side of things, the
consensus appears to be that there is “not a significant
relationship between player personality traits and their class
or faction selection”, as Ian D. Mosley famously stated in his
2010 World of Warcraft study. But can that really be the case
when so many people seem to feel appealed by this kind of
topic? Even though the posts’ background is mostly a
comical one, the entertainment value standing front and
center, there has got to be some form of origin to the
stereotyping for this many consumers to be able to relate to
it in the first place.
This paper aims to examine the decision made right before
starting to play. The result might just help to understand the
most important element of each game: the player. Getting to
know the player greatly aids with designing better games,
designing good games means improving the experience, and
a well-presented experience will in return stick with the
players, enabling them to reflect and inspire positive change.
In fact, several studies do indeed hint at the possibility of a
deeper connection between player personality and avatar but
tend to either concentrate on character creation, and/or shift
their focus to the gameplay aspect. They establish a rather
clear connection between player personality and in-game
behavior and regularly imply trends towards a certain
similarity concerning players and their agents (Worth 2015),
without touching however on some of the most popular
games of the last ten-plus years that are roster-based
(Newzoo 2021; Ranker 2020).
Contrary to Mosley who clearly states that “character
selection choices such as class and faction are not related to
the player’s personality therefore would be a poor choice for
future researchers to use”, others remark that “it would be
interesting to assess the interactions between character role,
player’s identification, as well as personality in future
research” (Delhove and Greitemeyer 2018), while providing
an overview on a variety of games and characters within the
same game, including both, each individual’s most-played
and alternate agents (Worth 2015).
So why, then, do we play whom we play? How do we choose
to put ourselves into a game when faced with a choice of a
set cast of avatars, instead of the close to endless possibilities
of a fully fletched character creator? By limiting the number
of options to a few dozen per title, the idea for this paper was
to achieve a more quantifiable result while keeping every
individual outcome easily attributable due to the distinct
characterization of each agent within their respective
franchise, in contrast to the “blank slates” many role-playing
games (RPGs) tend to offer. A questionnaire was developed
to lay the groundwork for a comparison between players’
impersonations of choice and their personality based on a
corresponding psychological test. Conduction and
evaluation results were then used to provide an overview on
which connections between player personality and choice of
avatar can generally be expected.

METHODS
Please note: Unless explicitly stated otherwise, the terms
“character”, “player character (PC)”, “avatar” and “agent”
will be used synonymously, all referring to the player-
selected and -controlled character within a videogame.
Furthermore, this paper is part of a bigger unpublished series
of research on player personality and character choice that
not only covers the correlation central to this essay, but also
dives into age as well as gender differences, and specifically
the player-avatar-relationship, thus discussing in greater
detail which agent attributes might be the most influential
and what makes a player choose (them) in the first place.
Hence, some additional information originating from the
same survey but initially compiled for different purposes will
be mentioned occasionally when applicable.
A questionnaire was compiled in pursuance of comparing
each participant’s background, their specific choice of
avatar, and of course mental nature with each other, but also
among and with other participants. The Big Five or OCEAN
model was used for personality detection since it is one of
the most applied, well-known, and recognized approaches in
psychological trait theory, and considered to be mostly stable
over time (Leon et al. 1979). It divides the human psyche
into five distinct factors, those being: openness to
experience, conscientiousness, extraversion, agreeableness,
and neuroticism. To quickly summarize them individually,
openness to experience (inventive and curious versus
consistent and cautious) refers to an interest in the
unexplored and a sense of creativity, conscientiousness
(efficient and organized versus extravagant and careless) to
self-regulation and precision, extraversion (outgoing and
energetic versus solitary and reserved) to whether someone’s
motivation stems from within or from other people,
agreeableness (friendly and compassionate versus
challenging and callous) to altruism and positivity, and
finally neuroticism (sensitive and nervous versus resilient
and confident) referring to emotional lability and the way
negative feelings tend to be handled (Goldberg 2019). While
there are certainly other usable personality theories, most of
them, like the HEXACO model to name an example (Worth
2015), apt to be based on the Big Five, so it was considered
a good starting point that could potentially be expanded upon
during future research.
The most reasonable way to now incorporate the Big Five
assessment into the poll was to resort to the open-source
website bigfive-test.com, recommended by Lewis
Goldberg’s official International Personality Item Pool
(Goldberg 2019). Apart from being scientific and widely
available, it has the advantage of working with IDs, thus
making the results easily accessible and comparable for
everyone willing to share their scores via link (Rubynor
2021).
The three games presented were chosen mainly based on the
at the time most played and popular multiplayer titles with a
character roster (Newzoo 2021; Ranker 2020). Additionally,
League of Legends, Overwatch and Super Smash Bros.
Ultimate all offer a vast variety of diverse playable avatars
in all sorts of shapes and sizes with different playstyles and
lore behind them, while being well worth juxtaposing
through their contrasting genres and roster size, positioning
themselves on the “sweet spot” between heterogeneity and
comparability. League of Legends (LoL/League) is a 2009
PC and Mac multiplayer online battle arena (MOBA),
developed and published by Riot Games, with 148 (as of
June 2020) playable “champions” that are officially
subclassified into six different roles, namely Assassins,
Fighters, Mages, Marksmen, Supports and Tanks (Riot
Games 2021). Overwatch (OW) is a 2016 PC and console
multiplayer first-person hero shooter, developed and
published by Blizzard Entertainment, with 32 (as of June
2020) playable “heroes” that are also further subclassified
into three different roles this time around – Tank, Damage
and Support (Blizzard Entertainment 2021). Finally, there is
Super Smash Bros. Ultimate (SSBU/Smash), a 2018
crossover platform fighter for the Nintendo Switch,
developed by Bandai Namco Studios and Sora Ltd. and
published by Nintendo, with 80 (as of June 2020) playable
“fighters” in no official categories (Nintendo 2018).
Google Forms was chosen as the survey platform since it
was able to provide anonymity and would not allow for the
identification of individual responses while also being
powerful enough to handle a sizable number of entries and
multiple different types of questions and answers within the
same poll. Participants were presented with a short
explanation of the project and the questionnaire itself,
including average duration and the necessity to have played
at least one of the games listed (casually or competitively).
They were then asked about their background information as
well as mains and secondaries, and afterwards linked to the
Big Five evaluation.
The poll went online on May 1, 2020 and remained
accessible via link until May 14. Participants were recruited
via email, several Reddit posts in the corresponding
“subreddits” for scientific research as well as the three
games’ respective communities; the same goes for accordant
Facebook groups, dedicated Discord servers, group chats in
WhatsApp and Telegram as well as public Jodel posts plus
word of mouth.
After those two weeks, the results were analyzed – first in
Google Forms’ own system, afterwards transferred in
Microsoft Excel and lastly in IBM’s SPSS Statistics for a
more detailed comparison. Since data was collected in two
different types – nominal for “main”, “main role”,
“secondary” and “secondary role” (everything that cannot
logically be described on a scale), and metric for the Big Five
(everything that can be directly assigned to a number), –
several distinctive methods had to be used, including the
bivariate correlation indicators Pearson, Kendall-Tau-b and
Spearman, as well as crosstabulation in Chi², the coefficient
of contingency, Phi and Cramer-V along with the eta
coefficient, all dependent on which combination of
categories were to be contrasted at each point in time. The
following Table 1 by Björn Walther (2020) was consulted as
reference:
Table 1: Methods for Calculating Correlation According to
the Type of Data
Nominal
Ordinal
Metric
Nominal
Coefficient of
contingency; Phi
and Cramer-V
Ordinal
Chi²
Kendall-Tau-
b; Spearman;
Gamma

Metric
Eta coefficient
Kendall-Tau-
b; Spearman;
Gamma
Bravais
Pearson
RESULTS
A total of 113 responses were collected, 99 of which were
complete and usable without circumstance, while the other
14 could only be considered partly and had to be disqualified
for everything personality-related, mostly due to issues with
the mobile version of the Big Five website. Starting off with
the broad, singular statistics:
League of Legends: The most-often selected main was Ashe
with 5 votes (7.7%), and while 9 (13.8%) out of the 65 LoL
players stated to not have a secondary in the game, Leona,
Miss Fortune and Vel’koz shared second place with 3 players
(4.6%) each.
Overwatch: The most popular character turned out to be
Mercy with 7 players (13.2%) “maining” her (playing her the
most). While 15 players (28.3%) stated not to have a
designated secondary PC, out of the ones that did, Mercy was
again the most popular choice with 4 participants (7.5%).
Super Smash Bros. Ultimate: Kirby was the most frequent
answer with a total of 9 players (17.6%). 22 participants
(43.1%) did not mention a secondary. The second biggest
fraction among alternative picks were 3 (5.9%) Roy players.
Early observations revealed that several clusters of arguably
like-minded people could immediately be identified simply
by sorting participants choosing the same avatar in groups
and determining their personality scores. For example, both
mains of the League of Legends character Miss Fortune
(“MF”) are very similar in all five personality aspects,
sporting a mere difference of 6 points in openness to
experience (100 vs. 106), 5 in conscientiousness (76 vs. 81),
2 in extraversion (63 vs. 65), 9 in agreeableness (102 vs. 93)
and 11 in neuroticism (56 vs. 67), which is a total variation
of 33 points out of the possible 600 (120 in each trait),
effectively indicating a 5.5% difference. Putting it into
perspective, their combined averages of 103 points in
openness, 78.5 in conscientiousness, 64 in extraversion, 97.5
in agreeableness and 61.5 in neuroticism are still a
consolidated 47.5 points shy of the overall averages
(counting every single player) of 86 in openness, 83 in
conscientiousness, 68.5 in extraversion, 87.5 in
agreeableness and 73 in neuroticism, resulting in a 7.9% total
difference in character traits. While not particularly notable
at first glance, considering the same calculations for other
pairings of participants playing the same avatar paints a
much clearer picture: Secondaries of the LoL champion
Kai’Sa score a 9 points difference in openness (71 vs. 80), 4
in conscientiousness (92 vs. 88), 11 in extraversion (56 vs.
67), 3 in agreeableness (75 vs. 72) and only 1 in neuroticism
(58 vs. 57), resulting in a total variation of 28 points or 4.6%
to each other and their combined averages of 75.5 in
openness, 90 in conscientiousness, 61.5 in extraversion, 73.5
in agreeableness and 57.5 in neuroticism accounting 54
points or a 9% collective difference to the average. However,
stacking those two examples against one another results in
an even bigger divergence of 69.5 points or 11.6%. Similar
findings can be observed with Falco and Link secondaries in
Super Smash Bros. Ultimate, just to name a couple examples
from another game: The two Falcos having a 15 points
difference in openness (83 vs. 98), no difference at all in
conscientiousness (both 77), 10 in extraversion (76 vs. 66),
2 in agreeableness (87 vs. 89) and 3 in neuroticism (78 vs.
81), resulting in a total of 30 points or 5% difference
compared to each other, and the two Links having a 2 points
difference in openness (86 vs. 88), again no difference at all
in conscientiousness (both 82), 9 in extraversion (69 vs. 60),
15 in agreeableness (95 vs. 80) and 11 in neuroticism (92 vs.
81) – all in all 37 points and 6.1% variation between them.
Figure 1 showcases just how similar these pairs are in and of
themselves, yet how distinct they can look next to the players
of different characters. Each chart depicts the Big Five scores
of two participants playing the same character; player 1
indicated by vertical, player 2 by horizontal stripes:
Figure 1: Four Comparisons between Players of the Same
Character’s Big Five Scores in Pairs
0
20
40
60
80
100
120
O C E A N
"Falco" Player #1 "Falco" Player #2
0
20
40
60
80
100
120
O C E A N
"Kai'Sa" Player #1 "Kai'Sa" Player #2
0
20
40
60
80
100
120
O C E A N
"Link" Player #1 "Link" Player #2
0
20
40
60
80
100
120
O C E A N
"MF" Player #1 "MF" Player #2

There are two issues hiding in those numbers, however: For
one, not all groups playing the same character feature those
rather striking similarities, and secondly – directly connected
to that – the sample size per singular character (counting
main and secondary separately, not combined) is extremely
small and only ranges from 1 to 7, meaning that the data
above can merely be counted as a hint into a direction, rather
than any kind of rule.
To achieve a larger sample size, there are again two feasible
contact points to be tackled. For one, the mean value of all
players maining a certain character combined, compared to
the same numbers of other avatars; and secondly, the official
grouping of agents into classes or categories. Unfortunately,
this is not possible for Super Smash Bros. Ultimate since
there simply are no official allocations within the game itself
or the media surrounding it. League of Legends and
Overwatch, however, both feature such a system, an
overview of which can be found on their respective websites.
League of Legends fosters an assortment of six roles –
Assassins, Fighters, Mages, Marksmen, Supports and Tanks
(Riot Games 2021) – while Overwatch features three – Tank,
Damage and Support (Blizzard Entertainment 2021). The
fact that there is an overlap between two of the roles (namely
Support and Tank) even makes a cross-title comparison
possible later down the line.
Starting off with the combined personality trait averages for
each character’s mains, it makes the most sense to look at the
game with the smallest number of playable avatars, ergo
Overwatch, to generate the biggest mean number of players
per PC. As can be seen in Figure 2, the results show that no
two hero’s averages are consistently close to each other all
the way through. In fact, every single character
(distinguished here by differing bar patterns) with a
reasonable sum of participants behind them has a relatively
clearly detectable profile to them.
Figure 2: Comparison between Average Big Five Profiles
of Different Heroes’ Players
With one exception, that is. Mercy and Moira are never more
than 6 points apart from each other, while scoring identically
in openness and conscientiousness. One could now argue
that both heroes are classified as Supports, thus play
similarly, and are ultimately selected by like-minded people;
however, so is Ana. She counts as just as much of a Support,
but places quite differently in neuroticism, extraversion, and
openness, while agreeableness and conscientiousness are,
again, similar.
Curiously, there are once more certain characteristics
asserting themselves as pivotal, even when examining one
game only. League of Legends, which is used as an example
for the sake of it having the highest number of players within
this study, shows quite a few noticeable dissimilarities
between its champions’ roles with up to 10 points (8.3%)
difference within one category; Fighters, for example,
scoring an average of 78 in conscientiousness, while
Assassins score an 88. This might not seem like a huge deal
at first, but given the generous size of each category’s player
base even within this survey alone, it is rather telling that a
notable distinction can be found at all, especially given the
almost nonexistent difference between the total averages of
all participants and those of the League players by
themselves, which are 86 vs. 85 in openness, 83 vs. 84 in
conscientiousness, 68 vs. 69 in extraversion, 87 for both in
agreeableness, and 73 vs. 71 in neuroticism. Figure 3 depicts
this phenomenon by assigning specific patterns
correspondent to each role:
Figure 3: Comparing Each of LoL’s Roles’ Average
Player’s Big Five Scores
However, it does not stop there. Overwatch’s players’ Big
Five averages with 86 for openness, 81 for
conscientiousness, 69 for extraversion, 86 for agreeableness
and 75 for neuroticism are once more very close to the
overall mean value, as can be seen below in Figure 4, where
the horizontally striped total average (LoL, OW and SSBU
players’ combined average Big Five scores) bar barely
differs from neither League’s, nor Overwatch’s mean values:

Figure 4: Comparison Between LoL and OW Players’
Mean Big Five Scores as well as the Total Average
With next to no significant discrepancies between the
average League of Legends player and the average
Overwatch player, one might now assume to find nothing of
use when comparing certain shared roles in more than one
game. However, when looking at the shared roles of Support
and Tank in both games featuring them, especially
considering neuroticism, extraversion and agreeableness as
the most decisive factors for these titles, one can not only
clearly make out a difference between “LoL Support mains”
and “LoL Tank mains” on one hand, and “OW Support
mains” and “OW Tank mains” on the other, which is to be
expected based on the previous diagrams, but also spot
unmistakable similarities between “OW-“ and “LoL Support
mains” as well as “OW-“ and “LoL Tank mains”, effectively
proving the affiliation of personality traits and preferred in-
game roles even across separate titles and genres. As an
outline, compare the first and the third set of bars in Figure
5, as well as the second and the fourth ones with each other
(different patterns in this case correspond with different Big
Five traits):
Figure 5: Comparison between the Average “OW Support”
and “-Tank” Players’ Scores in Neuroticism, Extraversion
and Agreeableness in Contrast to Those of “LoL Support”
and “-Tank” Players
Finally, moving on to the heart of this analysis. When using
any type of mathematical correlation, a specific formula is
chosen dependent on the type of data analyzed. This formula
then outputs a number between -1 and 1, the correlation
coefficient. The closer it is to 0, the weaker the correlation.
Positive correlation basically means that the higher figure A
is, the higher figure B is in return. Negative correlation on
the other hand means the higher figure A is, the lower figure
B is, and vice versa. Additionally, another score is shown
that signals the significance of each correlation – this time
however, it is only between 0 and 1 and the closer it gets to
0, the more significant the correlation. The most common
classification defines everything below 0.05 as “significant”
and everything below 0.01 as “highly significant”, which is
the system that will be used and referred to from this point
onwards.
When dealing with the correlation between personality and
the choice of avatar itself, a proper value can only be
calculated via the so-called eta coefficient within a
crosstabulation, meaning that the result will only ever
display the significance of the correlation, but no correlation
coefficient itself. This is because metric numbers (the score
of each Big Five category) are being compared to nominal
ones (the players’ agents). Each champion, hero and fighter
can be expressed via number; however, this number has no
exact value apart from identifying the PC – it does not exist
on any scale and only works for enumeration. Nevertheless,
there is in fact a highly significant correlation between a
player’s neuroticism score and their League of Legends
main. Furthermore, there is another significant correlation to
be found between a player’s conscientiousness score and
their Super Smash Bros. Ultimate secondary, albeit slightly
less substantial. Both can be retraced in Table 2 and Table 3,
respectively.
Table 2: Neuroticism Score and “LoL Main” Are Highly
Significantly Correlated
Value
df
Asymptotic
significance
(bilateral)
Chi² (Pearson)
1538.050
1406
0.008
Likelihood-
ratio
351.225
1406
1.000
Correlation
(linear-linear)
2.448
1
0.118
Number of
valid cases
57
Table 3: Conscientiousness Score and “SSBU Secondary”
Are Significantly Correlated
Value
df
Asymptotic
significance
(bilateral)
Chi² (Pearson)
376.444
323
0.022
Likelihood-ratio
140.149
232
1.000
Correlation
(linear-linear)
2.163
1
0.141
Number of valid
cases
28

In fact, quite a few correlations in this category almost reach
significance (“OW main” vs. conscientiousness, “OW
secondary” vs. openness, “SSBU main” vs. neuroticism,
“SSBU secondary” vs. agreeableness and especially “LoL
secondary” vs. extraversion just fall short by only a few
fractions), but again, the relatively small sample size per PC
must be acknowledged.
DISCUSSION
In general, mains were quite evenly distributed, which
speaks for all three games featuring a wide variety of well-
designed and balanced characters. Some outliers include
Ashe for LoL, Mercy for OW and Kirby for SSBU, all of
which accentuate a beginner-friendly playstyle (Ashe for
example being one of the tutorial characters when you first
start the game and Kirby being the only choice at the
beginning of SSBU’s campaign) and are indeed mostly
chosen by less experienced players within the respective
titles.
Apart from Mercy, secondaries were just as evenly
distributed as the mains, if not more. Mercy is again an
understandable exception as Overwatch’s team-based
structure requires more on the spot adaptation than both
League and Smash. Switching during one and the same
round leads to a dynamic calling for quick changes,
especially when it comes to supporting roles, in order not to
endanger the damage-dealing core of the team. Mercy shines
as easy to pick up on one hand, but also as useful to most
team compositions imaginable, granting players a lot of
variety due to her being the purest form of a classic healer
within Overwatch on the other.
What is hinted at by the small number of direct comparisons
possible between players’ of the same character’s personality
traits, is that the likelihood of two Big Five-wise similar
people might indeed be more likely to play assimilable or
even one and the same avatar. However, this of course does
not mean that all players of the same agent are
psychologically identical. As can be deduced from the results
of the parallel player-avatar-relationship survey, one’s
reasons for picking a PC are much more dependent on
personality than the choice itself, and those reasons are
manifold. Since with the Big Five alone there is no system to
concretely group personality types into clusters, direct
comparisons are limited by nature.
This whole issue is further underlined by the profiling of
Overwatch averages, where every playable character with a
reasonably sized player base within this study can be
identified with relative ease. However, in what way an avatar
might correlate with their respective player bases can only be
speculated on so far. Someone scoring low in neuroticism
might be more in tune with themselves ergo choose someone
who is less like them, challenging different perspectives;
support-heavy heroes could demand and therefore attract
players scoring higher in conscientiousness, as mastering
Mercy, Moira, Ana and Lúcio takes a considerable amount
of awareness of one’s surroundings and other people alike;
while damage-dealing and “tanky” classes arguably tend to
be more suitable for less careful approaches in playstyle.
Technically, all those thoughts so far come down to a
comparison of roles, which makes sense as it is the most
efficient and especially the most official way of arrangement
that there is without getting into more subjective territory.
On that note, when comparing League of Legends’ roles’
average Big Five scores, there are in fact small differences
to be seen immediately, such as the role Fighter scoring
lowest in conscientiousness, which could again be attributed
to a more spontaneous and less considerate playstyle being
required. Something similar could be said for their
comparably low scores in agreeableness and openness –
possibly explained by the fact that Fighters function more
independently from their team than any other class and the
least amount of communication is required to kill and stay
alive. Tanks, for example, usually do not inflict enough
damage on their own, Mages and Assassins are too frail, and
Marksmen only really prosper with a Support by their side,
both of which very fittingly score highest in extraversion, as
communication is key when embodying this position, even
more so than per usual since these are the two classes
working together the closest (Fandom 2021).
This is further reinforced by the fact that Overwatch’s roles
score similarly, while there is next to no difference at all to
be found anywhere between the average personalities of
LoL, OW and SSBU players, essentially meaning that
personality traits and preferred roles correlate across
different games, with “LoL Tanks“ being more akin to “OW
Tanks“ than to “LoL Supports”, and the other way around.
This, then, solidifies the argument that human personality
might indeed determine the likeliness to take on a specific
role within a videogame, independently even from the type
of videogame. In addition to LoL, OW and SSBU’s average
players featuring only very few unshared characteristics, the
independent roles’ averages all stray rather far away from
that total average, strengthening the claim. In short: all three
games’ average players display a lot of similarities in their
psychological profile, and so do all roles across titles – to
each other, that is. However, looking at different roles
reveals all their contrasting characteristics. Using this logic,
it makes more sense to group players by role and not by
game, at least concerning the study at hand.
Why is that the case, though? One possible explanation
might have to do with our reasons for picking a certain
character in the first place. As described by the previously
mentioned parallel study, the top five most important factors
to the average participant by far were the avatar’s playstyle,
design, role, appearance and finally their personal success
with them. In simpler terms, this means that what is most
relevant for our choice is firstly what the agent plays like and
how they look, which can be attributed to the fact that despite
all genre conventions, the same role will feel at least
somewhat similar in every game when it comes to objectives
and goals – a tank will always be the one to jump in front of
others to protect and take the damage for them, no matter if
it is Braum from League of Legends or Reinhardt from
Overwatch.
Additionally, talking about their outward appearance, they
seem to follow a pattern as well, at least to a certain degree,
which makes sense considering designers and artists will
almost always try to make their characters look like what
they do, with strong silhouettes and a game plan that is
crystal clear to everyone as soon as they see them (Isbister
2016).
With all of that naturally comes a specific expectation from
the player’s side and the choice itself might even turn into a

self-fulfilling prophecy, which then ties into the remaining
factor success. Because they want to have fun, because they
want to be successful, people might just tend to stick with
what they know, even in different titles. When they start a
new game, perhaps they seek familiarity in the unfamiliar
because success feels most probable this way. All in all,
players might tend to look for a similar experience according
to their personality and could find it in inheriting a similar
role within a different framework.
Moving on to the foundation of this whole analysis, the
correlation between the Big Five and the players’ choices of
mains and secondaries, two unmistakable connections could
be identified, one of them “significant”, the other one even
“highly significant”. Starting off with the latter: neuroticism
and League of Legends mains. This confirms the train of
thought previously theorized when simply comparing like-
minded people and their most-played characters. While there
were already quite a few clusters to be identified with the
naked eye, such as Ashe mains leaning more towards a lower
score in neuroticism, Teemos existing mostly towards the
middle of the spectrum, and Morganas, for example, always
scoring relatively high, the actual scientific proof confirms
these earlier speculations.
So why does League of Legends stand out in this regard
then? Possibly because it has the most players of the survey
and by far the largest number of playable characters. The
sample size is more extensive by nature and the chance of
finding an avatar catering to one’s specific demands much
higher.
And why does neuroticism? Firstly, because of the nature of
the neuroticism score itself. One could make an argument for
it being the primary source for wanting to play videogames
at all because it fulfills certain emotional needs no other
media or activity does. (Hemenover and Bowman 2018)
Secondly, neuroticism is by far the most telling of the bunch
for the purposes of this paper; as once more ratified by the
parallel player-avatar-relationship questionnaire:
Neuroticism correlates by far the most and with the most
reasons for picking a character. A perfect example for this
phenomenon are the positive correlations found between
neuroticism score and someone’s reasons for maining a
champion in League, most notably the emotional connection
as well as feelings of similarity and identification towards
the avatar, how accessible those are to explore a different
personality or promote escapism altogether, and last but not
least – and statistically speaking highly significant – how
much the character is seen as an idealized version of the
player.
When combining the approaches, there might be a pattern to
be pointed out: One could argue that Morgana players are on
average indeed more focused on the fantastical, getting away
from real life and portraying a different or idealized persona.
This higher rate of emotionality would even fit quite well
with what Morgana stands for within the League of Legends
universe, as the deeply conflicted fallen angel type (Riot
Games 2021). On the opposite end of the spectrum, there is
Ashe, whose way more stoic manner is metaphorically
reflected all the way down to her status as “the frost archer”.
Thus, in theory, even those who play her could be rather
pragmatic, realistic, and satisfied with themselves. In not
actively seeking any similarities, they might end up being
more like her.
The second significant correlation to be found is
conscientiousness versus Super Smash Bros. Ultimate
secondaries, and although puzzling at first, similar clusters
of Falcos at low conscientiousness, Links in the middle and
Pits at high conscientiousness are extremely pronounced,
even while looking at the raw data alone. When arranging all
participants by their conscientiousness score, both Falco
secondaries are right below each other, both Link secondaries
are right below each other, and both Pit secondaries are right
below each other, without a single space in between each of
the pairs. This could be taken as a hint for the possibility that
more than just League of Legends and neuroticism can and
have to be considered. Those little clues are unlikely to
merely be coincidence, otherwise the probability of them
being so manifold would be very low.
The biggest problem is that the numbers are too small; under
different circumstances, significant correlations would have
presumably been found in plenty more places. As a matter of
fact, there are so many more correlations between
participants’ Big Five scores and their mains and secondaries
in all three games that almost reach statistical significance
that it becomes close to impossible to assume there would
not be any kind of relationship between the two. Only 200
participants would have likely been enough to irrevocably
prove the point. However, to be able to recommend
characters to players solely based on their personality traits,
much more data would be needed.
Firstly, every avatar must be represented to start with, which
was not the case. Secondly, every avatar must be represented
multiple times to be able to calculate a reliable average,
which was only the case for some of the PCs. That means
that even in the most optimal scenario possible, at least 296
participants would have to fill out the questionnaire, at least
when League of Legends is being considered. For Overwatch
on its own, however, 64 would technically be enough. The
more, the better, obviously, but even a number like 296 is far
from impossible given more time and resources. It would be
amazing to now be able to easily predict people’s character
choices; and for a small pool of avatars that is not even
unthinkable, but for a waterproof and comprehensive
overview, it is simply not enough yet.
Tiny, restricted prognoses are conceivable though, such as
people that scored high in neuroticism being more likely to
pick Morgana than Ashe in LoL and the other way around,
whilst someone scoring high in conscientiousness would be
likely to pick Pit over Falco and the other way around in
SSBU. What this tells us, is the fact that videogames in
general can not only be designed with elements catering to
different personalities and player types in mind beforehand,
but also incorporate highly adaptive systems that alter
gameplay and customization possibilities while someone is
playing, based on their choice of avatar, profiting both
audience and developers alike.
Although character creators do have their advantages, they
have also been thoroughly analyzed countless times before
already. Furthermore, they make it quite difficult to group
the results together due to the vast number of combinations
usually possible on top of simply never offering more than a
small backstory for each creation. They are designed to be
filled in by the player, therefore making the connection
easier to grasp, but at the same time less clashing and
arguably less interesting. A character that is fully fledged out

from the start does not go hand in hand with the player’s
personality as easily, which is why it was important to
conduct research in this very particular area. Dealing with
these quite offbeat games (in comparison to the many RPGs
previously explored) inevitably calls for other ways of
expressing the wish for customization and individuality.
Self-expression and interactivity itself are being handled
quite differently, and the player versus player competition-
based environments focus much more on aspects like
performance and emotionality.
CONCLUSION
While many studies of the past have focused their efforts on
investigating the broad topic of the player-avatar-
relationship before, considering the comparison between our
personality and the character of choice, they all either
covered gameplay first and foremost and contrasted behavior
in videogames to the human psyche, or else preferred
examining avatar creation tools as their test objects. This
paper, however, analyzed whom players select to represent
them in a virtual environment when faced with the choice of
a roster of pre-defined characters, and why they might do so.
To better understand the most important part of each game –
the player – a questionnaire was developed, and a survey
conducted with 113 participants based on the Big Five model
of personality and the three competitive action games
League of Legends, Overwatch and Super Smash Bros.
Ultimate, culminating in an argument for a statistically
significant correlation between the participants’ Big Five
scores and their most-played characters. Plenty of those
characters had a distinct psychological profile when their
players’ average Big Five scores were stacked against those
of other characters, meaning that several avatars could be
identified by these personality-related numbers alone. At the
same time, roles like Tank and Support were consistent
across titles and genres due to their players’ surprisingly
similar average personality scores irrespective of
corresponding games.
On the topic of what could further add to these results, other
psychological or perhaps entirely games-related personality
models such as Nick Yee’s Gamer Motivation Profiling
(2016) and Richard Bartle’s taxonomy of player types (1996)
might be well worth considering. The Big Five worked well
for making out correlations with each individual factor;
however, the lack of a proper way of categorization hurts the
explanatory power of the conclusion. Nevertheless, a niche
was filled that was long overdue when it comes to
understanding how and why someone chooses to put part of
themselves into a game, which can be used to improve key
design features or develop completely new ones. With more
data, even just a bigger sample size, it could soon become
feasible to predict role and character choices and offer
character-specific gameplay additions based on personality
profile correlations alone.
REFERENCES
Bandai Namco Studios and Sora Ltd. 2018. Super Smash Bros.
Ultimate.
Bartle, R. 1996. "Hearts, Clubs, Diamonds, Spades: Players Who
suit MUDs". https://mud.co.uk/richard/hcds.htm
Blizzard Entertainment. 2004. World of Warcraft.
Blizzard Entertainment. 2016. Overwatch.
Delhove, M. and T. Greitemeyer. 2018. “The Relationship Between
Video Game Character Preferences and Aggressive and
Prosocial Personality Traits”. University of Innsbruck,
Austria.
Hemenover, S.H. and N.D. Bowman. 2018. “Video games,
emotion, and emotion regulation: expanding the scope”.
Annals of the International Communication Association.
Isbister, K. 2016. How Games Move Us: Emotion by Design. The
MIT Press.
Leon, G.R.; B. Gillum; R. Gillum; and M. Gouze. 1979.
“Personality Stability and Change Over a 30-Year
Period–Middle Age to Old Age”. Journal of Consulting
and Clinical Psychology, 47, 517-524.
Mosley, I.D. and S. Patrick. 2010. “Personality and Character
Selection in World of Warcraft”. McNair Scholars
Research Journal, Vol. 6: Iss. 1, Article 12.
Riot Games. 2009. League of Legends.
Walther, B. 2020. “Bivariate Korrelation in SPSS Rechnen“.
Bjoernwalther.com.
Worth, N. 2015. “Players and Avatars: The Connections between
Player Personality, Avatar Personality, and Behavior in
Video Games”. Brock University St. Catharines, Ontario.
Yee, N. 2016. “The Gamer Motivation Profile: What We Learned
From 250,000 Gamers”. In Proceedings of the 2016
Annual Symposium on Computer-Human Interaction in
Play. CHI PLAY.
WEB REFERENCES
https://bigfive-test.com/compare
https://ipip.ori.org/index.htm
https://leagueoflegends.fandom.com/wiki/Champion_classes
(retrieved June 26, 2020)
https://na.leagueoflegends.com/en-us/champions/ (retrieved June
26, 2020)
https://newzoo.com/insights/rankings/top-20-core-pc-games/
(retrieved April 1, 2020)
https://overwatch.gamepedia.com/Competitive_Play
https://playoverwatch.com/en-us/heroes (retrieved June 26, 2020)
https://www.google.com/forms/about/
https://www.ranker.com/list/most-popular-video-games-
today/ranker-games (retrieved April 1, 2020)
https://www.smashbros.com/en_US/fighter/index.html (retrieved
June 26, 2020)
https://youtu.be/lvsjAX0JWFc (retrieved April 1, 2020)
BIOGRAPHY
LUKAS KEUTER was born in Geilenkirchen, Germany
and has been studying Digital Games at TH Köln’s Cologne
Game Lab following 2016, obtaining his bachelor’s degree
in 2020 and pursuing the subsequent master’s program in the
tracks of game design and -arts with a focus on children’s
mental health. In 2021, he co-founded the independent game
development studio Spoonful Games, where he has been
acting as game designer ever since, recently working on
relationship-based character interaction as well as narrative
AI integration.