MSc Artificial Intelligence
Track: Gaming
Master Thesis
Investigating the Relationship between FFM, Game
Literacy, Content Generation and Game-play Preference
by
Norbert Heijne
10357769
24th October 2016
42 EC
October 2014 - October 2016
Supervisor:
Dr.ing. S.C.J. (Sander) Bakkes
Assessor:
Dr. F.M. (Frank) Nack
Faculteit der Natuurwetenschappen, Wiskunde en Informatica
Abstract
Game mechanic generation has a large search space that contains
many possible solutions that result in a viable game, but the game
mechanics might not be preferred by the player. To guide the generation process we would have to be able to predict the game-play
preference of the player. Game-play preference categorization is a relatively underexplored topic of investigation within the domain of game
artificial intelligence. To see whether we can categorize game-play preference with user profiling, we investigated the relationships between
personality, game literacy, experienced difficulty level and game-play
preference. We have created a game based on ”the Legend of Zelda: A
link to the past” which contains content generation for multiple gameplay elements. We gathered data via questionnaires and game metrics
to investigate the relationships. Our data shows that personality and
game literacy are likely indirectly linked to preference. However, experienced difficulty could not be linked to preference.
1
Investigating the Relationship between FFM, Game
Literacy, Content Generation and Game-play
Preference
Norbert Heijne
October 2016
Abstract
Game mechanic generation has a large search space that contains
many possible solutions that result in a viable game, but the game
mechanics might not be preferred by the player. To guide the generation process we would have to be able to predict the game-play
preference of the player. Game-play preference categorization is a relatively underexplored topic of investigation within the domain of game
artificial intelligence. To see whether we can categorize game-play preference with user profiling, we investigated the relationships between
personality, game literacy, experienced difficulty level and game-play
preference. We have created a game based on ”the Legend of Zelda: A
link to the past” which contains content generation for multiple gameplay elements. We gathered data via questionnaires and game metrics
to investigate the relationships. Our data shows that personality and
game literacy are likely indirectly linked to preference. However, experienced difficulty could not be linked to preference.
1
Introduction
Procedurally generated content (algorithmically created content while playing) has been used in many games to make the development process more
efficient by reducing design costs (design of a game’s environment and the
level content within those environments). In the genre rogue or rogue-light
game genre a player is often confronted with the same environments because
the player is forced to start over when the player fails to complete the game
during that try. In these kinds of games procedural generation of the environment is often mandatory to keep the game interesting and increase the
replay value.
2
Personalisation of games has also become more prevalent, this usually
consist of influencing the game’s difficulty settings based on the performance
of the player, which allows many different type of players to play the game at
their own level. Some games offer settings with which the player can tweak
the generation of content, examples of this are found in level generators
for strategy games where the player can set the frequency with which some
elements of the game occur.
Another layer of generation is the game mechanics generation where a
game’s set of rules are determined by an algorithm. Currently however this
type of generation has been mainly used to discover new types of games or
more interesting variations on existing types of games by manipulating the
set of rules which defines the game, an example is the board game Yavalath
[6] which came into existence through game generation techniques.
Game design formalization [27, 18] has added to the development of
tools for game mechanic generation but using game mechanics generation
during game-play still has a large underlying problem. The large, sparse
and chaotic search space for valid game mechanic rule combinations presents
numerous challenges. Many rules are valid but not all are interesting or
easily understood, since the player’s understanding of the game mechanics
and subsequent appreciation are the determining factors on whether the
game is actually fun.
Combining the above techniques could result in more interesting gameplay, where not only the environment changes but also the underlying elements change slightly with each play-through. For this we would need a
formal definition of what is fun and interesting about a game. Through
common sense we already know that different people favor different types of
games. There are few games that everyone will like.
Whether a game lies within a player’s preference depends on the required
time investment to get to a level of enjoyability that the player expects from
the game, the steepness of the learning curve, the amount of actions required
within a certain time frame and the amount of game conventions that it
builds on [19]. A too high of a requirement on any of the above factors could
prevent a player from enjoying the game [17]. This is why personalisation for
games is important. If a game would be flexible enough to allow all kinds of
players to get their preferred type of game-play, it would likely benefit from
it.
It thus becomes apparent that some form of player profile should be
constructed for the game mechanics generation to determine which game
mechanics should be used as the starting point and which parts can or should
be changed. Without a direction the game mechanics generation would be
3
no better than randomly changing the mechanics of a game’s element.
Even in the case of procedural content generation player profiling could
prove useful in directing the parameters of a game’s generated elements
into a direction that is more preferable to its player. Even with many
games implementing some form of procedural content generation, a way of
evaluating the generation process should be very welcome [13].
To investigate whether we can apply player profiling to direct game mechanic generation we will first have to try this in a more limited setting. Instead
of going directly to the game mechanic generation we still need a platform
on which we can gather data and which is limited enough in scope such
that we can complete our research goal without adding too many technical
challenges. Therefore there will be a focus on personalised procedural content generation and dynamic difficulty adjustment instead of game mechanic
generation.
1.1
Research Questions
To see if we can categorize game-play preference with user profiling we would
have to investigate a broad set of metrics which relate to the user profile of
a gamer. To see if this could also be used in an algorithm that would not
require any non-game metric input then the preference categorization would
also have to be retrievable with in-game metrics. To complement our in-game
metrics we would also have to implement a form of adaptable difficulty to
see if the amount of challenge a game element brings has influence on the
preference for such a game element.
To investigate this we would have to include the following types of metrics:
• Game literacy metrics, which are composed of game preference metrics
to represent the history and preference toward certain types of gameplay and gamer type metrics to indicate the amount of accumulated
practice and game knowledge.
• Personality scores to see if behavioural patterns transfer over to ingame behaviour.
• Game metrics, which are composed of behavioural metrics and performance metrics.
• Preference metrics, for each game-play element we would have to measure the preference to be able to investigate any possible categorization.
4
By cross checking these metrics we will investigate the influence between
each of the metrics and their influence on preference. To answer our question
if we can categorize game-play preference with user profiling the following
questions would have to be answered:
Problem statement: Can we categorize game-play preference with user
profiling?
RQ 1: Can we link performance to preference?
RQ 2: Can we link personality to preference?
RQ 3: Can we link difficulty preference and game knowledge to preference?
RQ 4: Should a game generation algorithm also focus on dynamic
difficulty adjustment?
RQ 5: Can we create a game that enables us to gather data to test the
possibility of these links and the dynamic difficulty adjustment?
RQ 6: If any links can be found is there a viable player profiling
solution among them?
2
Key Definitions
Procedural content generation: PCG is a well applied practice in many games
to date, this includes the generation of assets (textures, meshes, sound etc.),
environments (levels, lay-outs) and animations (sprites, inverse kinematics).
Non-player character : An NPC is a character in the game which can be
spoken or interacted with and that is not controlled by any person.
Dynamic difficulty adjustment: DDA is the application of a system that
adapts the difficulty while playing.
Game engine: A game engine is a piece of software that determines how
a game should be programmed and what the possibilities are. The engine
communicates with underlying processes of the computer that runs the game.
This allows the programmer using the game engine to focus on implementing
the actual game.
5
Game assets: Building blocks of a game, for example music, sprites (image sets which depict multiple frames of a 2D animation), tile sets (images
that contain multiple smaller images which can be used to create the environment), sound effects, enemies / NPCs / the player character (a combination
of sprites, sound effects and programmed behaviour).
Hardcore gamer : A hardcore gamer is for our intents and purposes a video
game hobbyist, one which spends or has spent a lot of time per week playing
video games and which prefers more challenging games. Hardcore gamers
are more likely than casual gamers to have a higher competence level for
playing games in general.
Casual gamer : A casual gamer is for our intents and purposes a video
game enthusiast which does not have a lot of time to play games or does not
want to invest too much time into playing games, as a result a casual gamer
is more likely to have a lower competence level for playing non-casual types
of video games and would be more drawn to casual types of games (mobile
games, online mini-games).
Action Role Playing Game (ARPG): A video game genre that entails that
the player takes the role of a character in a story with real-time combat
mechanics. RPGs in general usually have some form of quantification of the
characters strengths and weaknesses, where the character gets stronger the
further the story progresses.
Player profiling: Creating a model based on the player’s characteristics
and/or behaviour. This is used in making predictions about players or to
categorically classify them.
Trait based profiling: The Five Factor Model is an example of a trait based
profile, each trait is considered independent of each each other and the traits
are expressed in scales as opposed to categories.
Type based profiling: The Meyer-Briggs Type indicator is an example of
a type based profile, the measurements are translated into a type, a type
being a categorical classification that represents the measurements.
6
3
Related Work
In this section we will describe relevant papers for the three most important
aspects of our research, player profiling by using personality traits or types,
the reasoning behind and uses of certain types of content generation and the
reasoning and possible implementations of dynamic difficulty adjustment.
3.1
Personality player profiling
There exist many personality profiling tools and each tool has its uses. We
will consider three types, the Five Factor Model (FFM) with facets, the First
Demographic Game Design model (DGD1) which is strongly coupled with
the Myers-Briggs Type Indicator (MBTI) and the BrainHex model which is
based on the Second Demographic Game Design model (DGD2).
The Five Factor Model (FFM) consist of multiple personality dimensions:
openness to experience, conscientiousness, extraversion, agreeableness and
neuroticism [29]. Each dimensions consist of multiple facets which zoom in
on a particular type of behaviour or thought pattern within the dimension.
Other research has also tried to find relationships between in-game behaviour and personality profiling. In this particular research [28] their correlation analysis showed relationships between the FFM traits and video game
data, they concluded that a video game can be used to create an adequate
personality profile of a player. The research has shown that the relationship
between behaviour and personality is present in games which strengthens the
use of personality profiling as a metric to explore the game-play preference.
We see another instance of player modeling and relating game-play behaviour to FFM personality traits along with the facets of each FFM trait in
other research [1]. By using trait facets and segregation of behaviour by location they greatly increased the power of personality to explain behaviour.
An important lesson to learn from this research is to keep our situations
clean from too much interference from other game-play elements to reduce
noise in our data.
Personality profiling has also been done in a broad setting using game
genres [29]. The research showed that 2.6% to 7.5% of the game genre
preference can be accounted for due to personality factors. Even with a
low percentage like this the research points out that there are relationships
between personality and gaming preferences, but that there are still unidentified aspects that impact the choice of games. As such this should focus the
preference analysis on game elements instead of the game as part of a genre.
There is other research which states that there are significant relations
7
between FFM dimensions and game genres [23]. It states that people who
enjoy casual, music and party games tend to be more extraverted and that
people who enjoy role-playing games, MMORPGs, action role-playing games,
turnbased strategy and real-time strategy games tend to be less extraverted.
Additionally, people who like sport, racing, flight simulation, simulation and
fighting games tend to be more conscientious. Finally, people who like action
adventure and platform games tend to be more open to experience.
FFM dimensions have also been linked to game genres with estimations
of underlying motivation based on the Player Experience of Need Satisfaction (PENS) measure. Relations were found between agreeableness and the
feeling of competence from playing games. Openness to experience and the
need for autonomy, which sees games as providing interesting choices and
activities.
The DGD1 model identifies four categories of players’ in game preferences and behaviours [20, 8]. These player types are the Conqueror, Manager, Wanderer and Participant types. Each player type is associated with
particular personality traits using the MBTI. The Myers-Briggs types can
also be transformed into FFM traits reliably [11].
The BrainHex model identifies seven categories, each with their own underlying reason for enjoying a game, the type of sensations that they pursue
and consequently how they respond to those sensations [22]. The player
types are Seeker, Survivor, Daredevil, Mastermind, Conqueror, Socialiser
and achiever.
BrainHex [5] is based on other research about type and trait based models, in particular about Myers-Briggs, FFM and their own development of
a neurological based hybrid model in effort to create a more robust instrument. The underlying research mentions that game literacy and Openness
to Experience (an FFM trait) are related to whether a player has the willingness to persevere in pursuit of victory, i.e. a gamer hobbyist (Hardcore
gamer) instead of the mass market player (Casual gamer).
The BrainHex survey [22] tells us more about the BrainHex model itself.
Where they did a preliminary analysis of the Myers-Briggs types and how
they relate to the types in the brainhex model. From this we learn that the
research regarding this model is still ongoing.
3.2
Procedural Content Generation
Procedural Content generation that would be relevant to the scope of personalised game mechanics generation are things such as generating interact-able
objects such as story generation (narrative and conversations), characters,
8
items or enemy generation (comprised of texture, model, animation and behaviour generation), the environment generation (comprised of a mission and
space generator) and generation of the tasks themselves such as puzzles or
combat [3, 13].
Stories are also very hard to generate, multiple solutions are available
such as using partial-order plans (POP) to represent stories. Or taking
existing stories and representing the story as character actions. This is then
coupled with natural language processing to create game-like stories [13]. It
also states that currently this is still experimental and mostly used in an
offline scenario. As such we will have to handcraft our own story to reduce
our technical challenges regarding narrative and story.
An example of the item generation is weapon generation, one way of
personalising and bringing diversity into weapons is by adjusting the output
pattern of an available weapon [12]. In this case the weapons were all ranged
weapons that fired rounds in a distinct pattern, that pattern was evolved
using evolutionary algorithms based on the behaviour of the player.
Task generation such as puzzle generation is usually done through search
algorithms. An initial state is generated, this state is tested if it is solvable,
if it is not then a new state is generated by applying changes. This usually
also needs to adhere to some difficulty setting which creates multiple layers
of searching through the solution space. An example of this is the Lunar
Lockout generator POGGI [15] or the sokoban problem generator by Murase
et al. [21].
Search algorithms can be expensive to add to a game, especially when it
is unknown if a viable solution will be found that adheres to the constraints
such as complexity, difficulty and space requirements. This adds significantly
to the technical challenges and as such we will most likely implement a more
simple version.
Our mission generator is based on research by Dormans and Bakkes
[9] which described a method of using a replacement grammar on a graph
structure and by doing this ensuring validity of the graph with each transformation. We have taken the basics of this system, the grammar and graph
structure to produce our own level representations.
Although the initial intent was to have a procedural setup for the space
generation of the levels, it became clear that we would get results with less
noise by creating a more static but predictable generation of the layout.
9
3.3
Dynamic Difficulty Adjustment
A system to adapt the difficulty while playing is called Dynamic Difficulty
Adjustment (DDA). Usually limited to a specific parts of the game (combat
difficulty, puzzles used or resource parameters such as available life or ammunition). For example in the game Resident Evil 5, the game’s difficulty
is set beforehand via the difficulty setting, and then the player performance
determines an offset from the initial setting. Personalised difficulty systems
are often preferred over static systems when done correctly [3, 4].
The initial difficulty should be set to a global safe setting, a setting that
almost everyone can play and finish [4]. This ensures that the game can
adjust accordingly without the player abandoning the game because of the
difficulty.
The Flow model [14], which tries to balance the difficulty based on the
skill of the player is applied in our fight generator. The type of difficulty
adaptation action that we take is ”Proactive”, we adjust the oncoming fights
and puzzles based on performance of previous fights and puzzles along with
our assumption model on how difficulty should fluctuate. We opted out of
”Reactive” adjustments to difficulty, such as changing an enemy’s health
during a fight, because this does not make sense in our game’s conventions
and we assume this would likely confuse players.
In research there is also support for the hypothesis that a player is forced
into a cognitive state that approximates the flow state during game-play [8].
The game would require a cyclical balance of skill and challenge. The player’s
skill and resulting challenge relates to the overall complexity that is related
to the information that the player has to his or her disposal and the amount
of new information coming in. New players that have no concept of Zelda or
ARPG games will most likely not go into a flow state because the amount
of new information is too much of a challenge.
Another form of DDA with regards to combat is dynamic scripting [24],
where the internal decision making of the enemies is generated using a rulebase and this rule-base uses online learning to update the rules’ weights
based on the actions of the player. This grants an advantage over increasing
the number of enemies or changing the statistics of the enemies. When a
player is sufficiently skilled, often the enemies become trivial because their
behaviour does not change. When a player is a novice the game can be less
overwhelming because the way the enemies interact can remain simple.
To make the DDA for combat complete the enemies would have to work
together to combat the player, often by having the individuals perform
better they would still not take advantage of each others strengths or cover
10
their allies weaknesses. Learning your role within the team can improve the
performance of the AI as a team and as such create more varied challenges
for the player [2].
4
Research Focus and Limitations
To make this research project feasible we will have to place limitations on
multiple parts of our research. Because game mechanics generation for a
specific genre is a research subject by itself. However, our research is aimed
at underlying principles of user profiling and preference, as such we will
limit ourselves to procedural content generation within a narrow genre of
games. Any elements of the game that we cannot easily generate will have
to be designed by hand and emulate content generation to a certain degree.
With these limitations we can focus on player profiling in a more controlled
environment without having to deal with additional technical challenges.
4.1
Research Focus and Limitations overview
In the following sections we will describe how we will direct our design such
that we are left with a task which is feasible. We limit our task by employing
the following:
• We restrict our research to content generation instead of game mechanic generation.
• We will employ a form of dynamic difficulty adjustment for all puzzles
and fights.
• Reuse of the assets and conventions that are available.
• Using game design instead of complex generation algorithms.
4.2
Content Generation
In this research we apply content generation techniques to emulate the game
mechanics generation possibilities as much as possible. Such that the data
and results that we acquire can be applied to the game generation case albeit
in a limited scope.
Our content generation will focus on levels and its content, fights (enemy spawns and placement), room contents (props and walkable space) and
puzzle generation (chosen from three puzzle types).
11
4.3
Dynamic Difficulty Adjustment
Our implementation of a DDA system will focus on the difficulty of the
combat by determining the type and amount of enemies spawned for a fight.
Puzzles have difficulty levels with predetermined settings for each type of
puzzle. We also try to keep the increase in difficulty by other factors stable
within a level.
4.4
Legend of Zelda Assets and Conventions
Figure 1: The Legend of Zelda: A link to the past.
The game The Legend of Zelda: A link to the past, a type of Action RPG,
takes the player on an adventure that entails rescuing a princess from a
dungeon and subsequently going on a quest to save the kingdom from ruin.
The player is confronted with increasing challenges throughout the game
requiring either reflexes and hand-eye coordination or solving a puzzle with
logic and wit.
The game carries conventions that have been very successful within the
Action RPG genre of which the most prevalent are these:
• Combat is done with some kind of weapon (often close range), with the
player having only a rudimentary control over the weapon (a button
causes a specific attack animation). The player then complements the
rudimentary combat with tools and items found in the world.
12
• The hero has statistics with which the player can track the current
strength of the character (life points displayed as hearts and magic
power displayed as a green bar, see figure 2).
• When the hero has zero life points left it is game over.
• The hero gets stronger as the game progresses, either through an increase in statistics or available tools.
• The world contains secrets to find which are often unlocked through
the use of an item, tool, solving of a puzzle or defeat of a stronger than
usual enemy.
• The story often sends the hero to a dungeon (an enclosed area that is
often hostile) which is riddled with enemies and traps and ends in a
climactic boss fight (a much stronger enemy which often requires the
use of the item found in the dungeon to defeat it).
• Characters within the story often give hints on how to play the game
or how to proceed with the story.
The game that we made is based on the actual game The Legend of
Zelda: A link to the past, it uses roughly the same game mechanics and
conventions, albeit that we will generate our levels algorithmically and as
such will contain less depth than levels with thoroughly thought out game
design.
The thought behind this is that we are more likely to improve player
experience by using assets (e.g. characters, tile sets, enemies ) that are
already deemed successful and carry many conventions that are known. In
contrast to starting from scratch and emulating these conventions. On top
of that the Action RPG genre uses many different kinds of elements, which
widens our scope and potential data.
Because the assets are taken from a well known game, nostalgia could
influence the player’s rating of parts of the game and knowledge on the mechanics of the game could influence performance. To get better insight into
which players might be influenced we added a question to the demographic
questionnaire that covers if they have played The Legend of Zelda: A link to
the past before.
13
Figure 2: Screen-shots from the game The Legend of Zelda: A link to the
past, from top-left to bottom-right: fighting on a bridge, walking through
the village, receiving an important item, fighting a boss.
4.5
Game Design
Many parts of our game could not be generated because they are either too
hard to generate or there was no research or algorithm available that could
run efficiently with our chosen game engine. In this case, game design is
used to hand craft scenarios (e.g. the village) or to craft smaller elements
(e.g. enemies, hazards or props). These elements will then be used within
the generation process.
The village was handcrafted, this includes the items that are attainable,
the NPC conversations, NPC placement and the story that is being told.
We attempted to create an enjoyable story, such that the players who are
14
more interested in such a story would not be disappointed. The village also
enables us to properly measure differences between players in regards to
story decisions and NPC interaction.
Most of the assets were already supplied with the engine and where the
supplied assets fell short, new ones were created. This required manual
adjustment to get the right feel and impact on for example difficulty with
enemies or hazards.
The effects that we can measure in our research is limited to the quality
of our design. For example, the assumption of the linearity of the difficulty scaling depends on a correct implementation of the enemy and hazard
difficulty settings.
5
The Game: Dynamic Zelda
Since the focus of this research is on PCG and the player’s preferences, it
was in our best interest to make sure that all other aspects of a proper game
were accounted for. It would be detrimental to this research if the game was
considered to be just experimental. Surely a game with no sound, horrible
graphics and only game-play elements that have not been tried and tested
before would detract from any kind of flow experience.
The decision to then use an already existing game becomes a logical
next step. When this research began we opted for a game engine that
contained most of these assets and game-play elements from the start. We
stumbled upon the Solarus Engine [25], an open-source game engine that
had implemented its own version of The Legend of Zelda: A Link to the
Past [26]. Compared to other platform that is often used, Infinite Mario,
it had more to offer in terms of game-play variations. Zelda games often
contains combat, puzzles, story elements, exploration and is an iconic title
that most gamers would recognize. The engine has a relatively low cost
to run the game in terms of processing power and has many elements and
conventions already implemented that many players would be familiar with.
As such we opted to turn this into something that would implement PCG.
15
Figure 3: Flow of the game.
5.1
The Game Flow
The game’s structure is set up as followed:
• The village: A place where the player is introduced to the quest, has
the opportunity to learn more about the background of the characters
and prepare for the journey ahead.
• Tutorial Level: A level with no exploration options or extra resources,
it contains four combat rooms and one of each puzzle at the lowest
difficulty.
• Shop: The shop houses a choice of equipment pieces that could simplify
certain tasks, primarily tasks that involve getting hurt such as the
mazes, the moving floor rooms or the fights. These pieces can be
bought with money found in the levels. Exiting the shop transports
the player to the next level.
• Levels: A themed segment is made up out two levels with a generated
layout in a particular theme that contains rooms that either generate
combat, puzzles or a treasure chest. This is the main part of the game.
After completing a themed segment, the player is transported to the
shop. There are three segments in total.
• Boss fight: A typical boss fight for a Zelda game, this is handled by
our combat generator.
• The village again: The player rounds up the quest with a choice on
how to fulfil the quest requirements and completes the game.
16
5.2
Village
Figure 4: The Village.
The Village (figure 4) is where the player starts his adventure, the player
is spawned near A2 with sinister music playing in the background. The
player is now motivated by the player character’s father (A2) to go and seek
help from the witch (C1) to cure the player character’s brother (A1) from
his illness and is given the instruction to head east to ask her for help after
picking up the Sword and Shield from the shed.
Some of the NPCs are placed specifically to allow players to prepare for
the oncoming journey. These include the shopkeeper (B1) which sells apples
to the player, which restores 1 heart of the players health pool. The boy who
lost his milk bottle (B8) reveals that he lost his bottle and asks the player to
find it among the bushes. The player gets to keep the bottle when he finds
it, which can later be used to fill it with potion from the witch to attain
background information on the witch and to contain fairies which revive the
player on death.
To complete the game after acquiring the ”Cure Flower” quest item, the
player has to talk to the brewer (B6) or the witch (C1) to make the cure.
The brewer will do this for free, and create a strong cure. The witch will
ask the player for 50 rupees and creates a diluted cure. Both will cure the
17
illness of the brother, thus the choice is there for flavor. When the player
presents the cure to A1 the game’s credits roll and the game is complete.
All other NPCs are non-beneficial and only add flavor and background
information to the game.
The Village is built in such a way that little interaction is required and
that conversational content can be ignored if the player wishes to follow the
instructions as fast as possible.
5.3
Level Generation
In this section we will present all the parameters used when generating a
level and how a level is generated.
The levels are built in four steps:
• A level’s content is first generated as a graph representation which is
called a mission. The mission only has information on the connectivity
of rooms, their type and the content of the chests. The generation
process is based on the mission generator in [9].
• Spatial planning is laid out using the maze generator that prioritizes
placing rooms such that the level always stretches outward. Pathways
inside the rooms are also laid out using the maze generator.
• Assets are places, for the forest and swamp levels the tree background
is generated, for the cave the walls, floors and doorways are generated.
Afterwards any space left that was designated as non-walkable is filled
with static props, throw-able props (bushes or rocks) or hazardous tiles
(such as water or holes).
• Barriers are placed, such as bushes and rocks that block specific entrances. Treasure chests are placed.
5.3.1
Level length settings
The parameters used in generating a level can be found in table 1.
A themed segment (Forest / Swamp / Caves) consists of two levels. A
level has a main length of rooms which are either a puzzle or fight, added to
that is a start and an end room and the first level of a segment has a room
with a treasure chest halfway through the main path that holds an item with
which the player can destroy the new type of barriers in that segment. Each
themed segment has a new type of barrier and also spawns old barrier types.
18
Figure 5: Level layout and use of the level length parameters for generating
a level
An optional path entrance is placed just before the treasure room or the
halfway point that is blocked by a new type barrier, requiring the player
to go back in the first level of a segment if they want to explore. Each
room in the optional path opens up a path to a branch that has a specific
number of rooms before entering a room with a treasure chest that holds
100 rupees. In the second level one of the optional treasure chests contains
a heart container.
Level name
Tutorial level
Forest 1st level
Forest 2nd level
Swamp 1st level
Swamp 2nd level
Caves 1st level
Caves 2nd level
branch
length
N/A
1
2
3
4
5
6
optional
path length
0
2
2
2
2
2
2
main path
length
7
4
4
4
4
4
4
Barrier
type
bush
rock
rock
large skulls
large skulls
bomb-able blocks
bomb-able blocks
Table 1: The length parameters (amount of rooms) for generating a level.
The purpose of the linearly increasing branch lengths is to increase the
time between the payout of the extrinsic rewards and to avoid a ceiling effect
19
when it comes to the number of explored optional rooms.
5.3.2
Mission generator
The underlying mission generator is used primarily to generate a graph that
can be used within our generation process. It uses the parameters and
generation process described in the previous section. During development
the generic implementation of the mission generation found in the paper [9]
was changed to generate more static levels, in which we could specify the
branching such that the level would not generate vastly different maps, which
would detract from our experiment and to prevent the generation from going
out of bounds of the map’s allowed measurements.
The mission generator still uses the same principle as described in the
paper, a graph representation is used with directional and bidirectional relations, and the nodes in combination with the relation determine what the
node represents in our generation process. Since we wanted to avoid stringing together multiple of the same tasks after one another (alternate between
fight and puzzle) such that the player is not bored with too much repetitiveness and that the main path always contains the same amount of fights and
puzzles such that we will get about as many results for fights and puzzles;
during a level with a specific difficulty in the case of static difficulty this is
even more important as it keeps our data comparable between levels.
Once the mission generator parses the graph representation we have
our mission representation, which contains the data for rooms and their
connection to other rooms, barrier placement, for chest rooms their chest
contents, start and exit rooms as well as the overall level information about
difficulty, theme and type of map (tutorial, normal or boss). This data is
sent to the space generator to generate our level layout.
5.3.3
Space generator
Our space generator uses the Maze generator to produce the layout of our
level, the map. The Maze generator takes in three parameters: Wall width
(x pixels by y pixels), corridor width (x pixels by y pixels) and total maze
measurements (x corridors by y corridors). The maze generator supplies us
with a grid where each node contains information on which wall is open,
closed or contains a small pathway to the other node.
Starting with the entrance area at the most western room in the middle
of the ”maze” we assign the room its number that is associated with a room
in our mission representation and we assign the connected rooms to the
20
Figure 6: Visual representation of the maze generation parameters
adjacent rooms that has the most space left (a random pick between the
room with the most adjacent unassigned rooms in 2 degrees, see fig. 7).
Connections to other rooms are designated as a small pathway, if a room is
larger than one node the walls between the nodes are designated as open.
Figure 7: Placing a room within a grid. Grey is an already present room.
Green, Red and blue indicate possible locations, the colored rectangles signify the unassigned rooms in 2 degrees.
21
Fight rooms are square and take up the size of one room, puzzles need
the extra space to be viable as puzzles and take up the size of two rooms.
Boss rooms are also square but take up four rooms.
We now have a grid as our spatial representation with which to further
improve our layout. The next step involves carving paths inside the nonpuzzle rooms (puzzle rooms need all the space available). For this we use the
Maze generator again. We first create a grid of the room with the corridor
size of 16 pixels by 16 pixels (the size required for our character to be able to
walk through), we then subdivide our room over the length and width into
sections, and use the center cells (with a random offset) of each section to use
our maze generation algorithm on. Using the smaller section representation
we use path-finding to open up cells from each center cell to another cell
where the section representation indicated that there is an opening. The
opened cells are widened resulting in a pathway that connects the entrance
with all exits in the room, the still unopened cells in the grid are used for
prop placement.
5.3.4
Placement of assets
Our assets that we use are made up of sprites and tiles. Sprites are objects
in the game that have an animation based on their internal state that can be
interacted with by input of the player. Tiles are static and are placed on a
layer, tiles cannot be interacted with but can affect the player (water, holes,
spiked floor all cause damage); they can be arranged in a pattern to form a
prop (for example a tree is made of tiles that are placed on multiple layers
in a pattern). We created many of these props to be used together with each
theme. The produced spatial representation of the previous section is used
to fill the level with sprites, tiles and props.
In case of the Forest and Swamp themes the walls and non-walkable
sections are filled with trees that are placed in a repeating pattern that is
common in the The Legend of Zelda: A Link to the Past. These trees are
only placed if there is space to place an entire tree. The remaining nonwalkable space in the rooms is filled randomly with either water (the player
jumps a certain distance when touching the edge of the water), throw-able
sprites (bushes or rocks), smaller props or designated as empty space.
In case of the Caves the wall props are placed and remaining space in
the walls section is filled with a specific tile, which is also common for The
Legend of Zelda: A Link to the Past. The non-walkable space in rooms is
filled randomly with either spiked flooring (which damages and stops player
movement), pits (where the player directly falls into, receives damage and is
22
placed back on the last spot where the player was safely standing), throwable sprites (mostly rocks), or smaller props.
The change in asset placement is used to linearly increase difficulty besides the already present adaptable difficulty, as well as to provide a noticeable change in the way you can move around in the environment. A question
on the overall feel of the level was therefore added to the in-game questionnaire to acquire insight into the effect of the themed props and placement.
Figure 8: Themes: Top-left, Caves. Top-right, Swamp. Bottom, Forest.
5.4
Combat Generation
Combat generation is done via the fight generator. There are two settings
due to the experimental setup. A static linearly scaling difficulty and an
adaptable difficulty.
Both difficulty settings use a specific set of enemies. These enemies are
displayed in the figure below.
23
Figure 9: The enemies: From left to right, Hardhat, Snapdragon, Minillosaur, Green knight.
The enemies each have their own behavior pattern as well as some shared
behavior:
• The enemy types Hardhat, Minillosaur and Green knight moves towards the player once it is within 100 pixels of the player.
• The player loses half a heart when the player’s character touches an
enemy.
• The enemy types Hardhat, Snapdragon and Green knight move about
randomly before spotting the player.
• The Hardhat repels the player backward when it is hit with a sword
strike.
• The Minillosaur starts out as an immobile egg, and cracks open once
the player is within range.
• The Snapdragon always moves about randomly in a diagonal pattern
and ignores the player.
• The Green Knight’s sword repels the player’s sword attacks, preventing
damage but repels the Green Knight backwards.
The static difficulty is handpicked and linearly increasing between levels,
hardhats only spawn in the caves on this difficulty setting and the amount
of enemies spawned are based on play-tests of ”average” players.
24
Level number
Tutorial
Forest
Swamp
Caves
Enemy spawn combinations
[M×4] OR [G×2] OR [S×3]
[M×5] OR [G×3] OR [S×5]
[M×7] OR [G×4] OR [S×6]
[M×8 + H×2] OR [G×4 + H×2] OR
[S×6 + H×2] OR [H×5]
Table 2: The settings for static combat, H: Hardhat, S: Snapdragon, M:
Minillosaur, G: Green knight.
The adaptable difficulty is made with Flow experience [8, 14] in mind,
balancing the enemy composition in a room in such a way that it tries to
find the optimal weight per enemy type based on the player’s performance
and actions using linear regression. The combat’s difficulty is then moved
up and down in a sinus wave pattern such that the highest difficulty is the
optimal difficulty given the player (A difficulty that the player would just
about handle).
We suspect that (hardcore) players that enjoy challenges would prefer
the adaptable difficulty over the static difficulty because it presents a more
suited challenge a lot earlier than the static difficulty and that casual players
enjoy short combat with low difficulty and therefore would prefer the static
difficulty in the earlier levels.
Sadly our combat generator cannot differentiate between the two types
of players because of the lack of data, as such players might quit because of
the adaptable difficulty being too difficult or players might quit during the
static difficulty in later levels when the difficulty becomes too much.
To decrease the amount of noise in the difficulty of the combat we implemented resource drops in a deterministic way. The amount of hearts
available to the player will determine the amount of mistakes the player can
make and consequently the difficulty of the fight. If some fights would grant
a lot of hearts and others none then it would be detrimental to our data.
As such the loot table (see table 3 for details) is determined beforehand, an
item is dropped from the loot table randomly with no returns whenever an
enemy is killed or a throwable prop is picked up. When the loot table is
empty it is refilled.
Type of drop
Amount
Hearts
1
Arrows
2
Bombs
1
Magic
1
Single Rupee
4
Table 3: The loot table used for enemies and picking up throw-able props.
25
The boss always has a static difficulty (requires 6 hits with bomb explosions) and is relatively easy as long as the player has the convention of
looking for a monster’s weakness, pays attention to auditory (any hit with
a sword bounces off) and visual cues (small minions drop bombs) and reads
the bomb bag acquisition description that the bombs are throwable or knows
the Zelda convention that the boss is always beaten with the use of the last
equipment piece found in the dungeon.
The qualities that would help improve the player’s performance with
combat are reaction time and planning. Timing sword swings decreases the
difficulty of the Green knight enemies the most and is useful against all
enemy types. Hardhats are more easily dealt with using items or throwable
props. Using any type of item from the shop will decrease the difficulty of
the fights. Timing the sword swings would indicate a larger precision and
less sword swings overall.
5.5
Puzzle Generation
A puzzle is generated when the player enters a room that has been designated
as a puzzle room by the mission generator. Which puzzle type is generated
is based on the amount of puzzles of each type has been generated, the type
is then chosen randomly between the types that have been generated the
least amount of times such that the generated amount remains the same but
with random ordering.
Like the combat generation we implement two variants of the difficulty,
a static linearly scaling difficulty and an adaptable version. The adaptable
version of the puzzle generation carries a simpler version than the combat
generation. The difficulty increases or decreases based on the time it took to
complete the puzzle, the life lost during the puzzle and whether the player
has died during the puzzle or in the case of the sokoban puzzle type the
player pressed the optional ”Quit puzzle” button. The cut-off points have
been determined with play-tests of persons with average to high level of skill.
The static difficulty was also determined with play-tests by taking the
maximum attainable difficulty level of the play-tester with the lowest skill
level as the ceiling of our linearly scaling difficulty.
Details on the difficulty setting of a particular puzzle type can be found
in the subsequent puzzle type specific sections.
26
5.5.1
Maze
We create our maze in the designated room using Prim’s algorithm, this
results in a maze without loops with a fixed start- and endpoint and random
paths in between with a good amount of branching. The difficulty of a maze
of the above type (no loops, completely random and with fixed start- and
endpoints) is usually defined by the size of the maze which comes down
to the combined length of paths that are most likely to go towards the
endpoint in the maze. A maze is unlikely not to be solved when the player
can see the entire maze and without loops in the maze a player can reach the
end in a reasonable amount of time with just trial and error. In our game
world we cannot make very large mazes without impacting the game’s flow
and performance due to the nature of this research and the game’s engine
respectively.
Figure 10: The hazards present in the maze: left to right, fireball statue,
bouncing hazard, reactive spike block, pit.
As such the following features were added to increase the difficulty of the
maze:
• Darkness: the entire maze is covered in darkness and the player can
only see a small area ahead of him. The player can use the Magic
mirror item at the cost of magic points to illuminate the maze for a
short time.
• Pits: Some of the dead ends have black pits which cause the player to
be transported back to beginning of the maze as well as lose life points.
• Bouncing hazards (official enemy name is Bubble): These invulnerable
enemies bounce their way through the maze and touching them causes
the player to lose life and magic points.
27
• Fireball spitting statues: Another invulnerable enemy which shoots a
projectile at the player roughly every four seconds, the player can use
his shield to block the projectile or evade it through lateral movement.
This projectile moves through walls in the maze.
• Reactive spike blocks: Yet another invulnerable enemy which automatically moves at a fast speed towards the player when the player is
directly horizontally or vertically aligned with the enemy and within a
certain range, the enemy cannot move past the walls of the mazes.
The details on the settings for static and adaptable difficulty can be
found in table 5 and 6. The settings for generating a maze puzzle given a
difficulty level can be found in table 4.
Difficulty level
Fireball statues
Bouncing hazards
Pitfalls
Reactive spike blocks
Darkness
1
0
0
No
No
Yes
2
1
1
Yes
No
Yes
3
2
2
Yes
No
Yes
4
3
3
Yes
Yes
Yes
5
4
4
Yes
Yes
Yes
Table 4: The generation parameter settings of the maze per difficulty.
Static
Level name Difficulty
Tutorial
2
Forest
3
Swamp
4
Caves
5
Table 5: The settings of the static difficulty for the moving floor puzzle.
28
Adaptable
Hearts lost AND/OR Time spent (seconds)
COP = 20 × (( difficulty +1)/2)
≤ 1 HL AND TS ≤ COP
≤ 1 HL AND COP < TS ≤ COP ×1.5
≤ 2 HL AND COP < TS ≤ COP
> 3 HL OR TS > COP ×1.5
> 4 HL OR death
Difficulty
change
+1
+0.5
+0.5
-0.5
-1
Table 6: The settings of the adaptable difficulty for the moving floor puzzle.
The amount of hearts lost (HL) or the time spent (TS) versus the cut-off
point (COP) is used to determine the change in difficulty and the difficulty
level rounded down is used.
The likely qualities that would help in solving this puzzle type is reaction time to avoid spike blocks, pits, projectiles and bouncing hazards and
planning to block the projectiles, which would indicate a relatively small
amount of life lost and a larger amount of time taken. Reckless behaviour
would most likely result in falling in pits, relatively large losses of life, but
small amount of time spent on completing the maze.
5.5.2
Moving Floor
The moving floor puzzle is a maze that has large corridors in which the floor
moves in a particular pattern and speed and where touching the wall causes
damage to the player, the player can see the movement of the floor and has
full visibility of the environment.
Figure 11: Examples of the moving floor puzzle generation within a level.
Left has difficulty 3, right has difficulty 5.
The likely qualities that would help in solving this puzzle is planning
29
and timing: The floor moves in a repeating pattern for a set amount of
time at a constant speed with visual acceleration and deceleration, if the
player positions the hero properly the player can maneuver through the
puzzle without harm. However, reckless action by the player will result in
constant harm from correcting mistakes while the floor continues to execute
its pattern.
The details on the settings for static and adaptable difficulty can be
found in table 8 and 9. The settings for generating a moving floor puzzle
given a difficulty level can be found in table 7.
Difficulty level
Corridor width
(× character width)
Speed
Pattern
1
4
2
3.5
3
3
4
2.5
5
2
48 pixels per second
Random between
up-down / left-right / up-right-down-left
Table 7: The generation parameter settings of the moving floor puzzle per
difficulty.
Static
Level name Difficulty
Tutorial
2
Forest
3
Swamp
4
Caves
5
Table 8: The settings of the static difficulty for the moving floor puzzle.
30
Adaptable
Hearts lost AND/OR Time spent (seconds)
COP = 20 × (( difficulty +1)/2)
≤ 1 HL AND TS ≤ COP
≤ 1 HL AND COP < TS ≤ COP ×1.5
≤ 2 HL AND COP < TS ≤ COP
> 3 HL OR TS > COP ×1.5
> 4 HL OR death
Difficulty
change
+1
+0.5
+0.5
-0.5
-1
Table 9: The settings of the adaptable difficulty for the moving floor puzzle.
The amount of hearts lost (HL) or the time spent (TS) versus the cut-off
point (COP) is used to determine the change in difficulty and the difficulty
level rounded down is used.
5.5.3
Sokoban
Sokoban is an already existing game which entails the following: The object
of the game is to move blocks on top of switches by pushing them with a
character, there exist at least as many movable blocks as there are switches,
the player wins when every switch has a block on top of it. The game
depends heavily on the layout of the walkable space and the placement of
the movable blocks and switches.
Generating a Sokoban puzzle can be done via random generated layouts
which are then tested for difficulty and feasibility by a bot [21]. Our game
however relies on on-the-fly generation of puzzles and as such a generation
pipeline would no longer be in real-time and goes beyond the purpose of our
research. There exist many fun handcrafted puzzles which already have a
generalized text format and therefore we use a downloaded set of handcrafted
layouts [10] that are small enough to fit in our rooms and are generated in
game by parsing the adjusted downloaded set and placing them in the room
at which point our algorithm makes the layout playable.
The adjustments have been made to the downloaded set:
• An entrance and exit have been added, the exit is blocked by a wall,
which opens up when all switches have blocks on them. All movable blocks and switches also disappear when the player completes the
puzzle.
• A switch was added at the entrance to reset the puzzle and teleportation tiles (which teleports the player on top of the reset switch) have
been added at multiple spots instead of a wall, such that the player
31
can always reset the puzzle even if the player blocks off the path to the
reset switch with a movable block.
• Difficulty annotation has been added to each layout based on playtests.
• The layouts that are similar and use the same trick to solve the puzzle
have been grouped together and ordered in difficulty.
An example of the layout and details on the handcrafted puzzles can be
found in figure 12.
Figure 12: Example of the Sokoban generation: Layout group 11, 1st in the
group, difficulty 3. (#): Wall, (.): Switch, (@): Entrance, (E): Exit, ($):
Block, ( ): Empty, (R): Teleport.
The player always completes each group in order of difficulty and is not
suddenly presented with a higher difficulty version while the player hasn’t
completed the lower difficulty version. The layout used is randomly picked
from the list of layouts of a certain difficulty where the layout is the first in
the group or the player has completed the layouts that came before in the
group’s ordering. When no layouts fit the criteria we look for layouts one
level of difficulty lower and fitting the same criteria. If no fitting layouts
are found, we pick a difficulty level higher than the current. The adaptable
difficulty takes the difference between current difficulty and selected difficulty
into account when deciding whether to change the current difficulty setting
when the puzzle is completed.
Sokoban is a type of puzzle that requires the player to spot the trick
used to complete the puzzle, this can take a long while depending on the
player. Being able to plan ahead is vital in completing these puzzles. We
have therefore added a ”Quit” option for this puzzle where after a certain
time a message is displayed on screen saying that they can press a button
32
to instantly complete the puzzle. This is to ensure that the research is
not hindered by the player’s inability to solve the puzzle. The adaptable
difficulty and logging both take the quit option into account.
The details on the settings for static and adaptable difficulty can be
found in table 10.
Static
Level name Difficulty
Tutorial
2
Forest
3
Swamp
4
Caves
5
Adaptable
Time spent (seconds) Difficulty change
≤ 90
+1
> 90 and ≤ 135
+0.5
> 135
-0.5
> 135 & pressed quit
-1
Table 10: The Sokoban settings for the adaptable and static difficulty. In
case of the adaptable difficulty level is rounded down when used.
5.6
Exploration Rewards and Shop
Most games apply some form of extrinsic rewards to complement the intrinsic
rewards of playing, completing a challenge or completing a non-mandatory
task. The shop was added to give players more choice in their combat play
style and to present extrinsic rewards for exploration.
The player has a possible income of at least 300 rupees in between shops,
with a bonus on top for killing enemies. All items cost 300 rupees to buy
with the exception of Apples (15 for 3) and Fairies (50 for one).
On top of that one of the chests in the second segment in between shops
contains a full heart container which restores the player’s life points and
increases the maximum life points a player can have.
The following items were added to the shop:
• Bottle: An extra bottle to carry an extra fairy, the amount of life the
player has available is an indicator for how reckless the player can be
before having to restock on life points. The player can only buy one
extra bottle.
• Fairy: Buying a fairy while the player has an empty bottle results in
a bottled fairy which is released upon the players death, restoring the
player back to life with full life. If the player has no empty bottles the
fairy is instead consumed immediately, restoring the player to full life.
• Three apples: The player can own a maximum of 10 apples, and the
33
player buys 3 at a time. An apple can be consumed by the player to
restore one heart.
• Magic mirror : The player can use this item at the cost of magic points
to blind all enemies on the screen, effectively immobilizing them and
during this time, contact with the enemies no longer causes damage.
The player can also utilize this item to illuminate a dark maze for a
short time.
• Bow : The player can shoot arrows in a straight line either horizontally
or vertically which kills normal enemies if it hits. The player can find
arrows by lifting bushes, lifting rocks or killing enemies. The player
can hold up to 10 arrows.
Figure 13: Items in action: Top, bow and arrow. Right, fairy release after
death. Left, magic mirror.
34
Figure 14: The shop.
The shop also contains an NPC that asks the player to fill in an in-game
questionnaire about the last themed segment they completed. The details of
these questions can be found in the Appendix. The NPC, the shop and the
in-game questionnaire also acts as a an interruption of the flow of the game
such that the data gathered in the following segment is much less based on
the feeling of the segments that came before. Although the time spent on
the game would likely still have a large effect, given the average player’s time
they can stay focused.
6
User Profiling
Before we can find out if our user profiling can be linked to game-play
preference we should first define it. The following sections will describe
the elements that will be included in the user profile. It should be noted
that the user profiling is meant to capture a broad area of measurements
such that extremes on the measurements could clearly show us differences
in preference, behaviour and performance.
The user profiling will determine what kind of data we gather before the
game starts and will be used in the analysis of the game metrics and the
preferences.
6.1
Overview of the User Profiling
In the following sections we will discuss the following:
35
• Personality test, which personality test should be used in our user
profiling and what are the details of the personality test.
• The assumptions and hypotheses regarding the following topics:
– Game literacy
– Personality influence
– Preference of difficulty
6.2
Personality Test
The following properties were taken into account while choosing the personality test to use:
• Ability to compare with other related work to ensure that this research
is relevant.
• The scales available to be measured and specificity of scales. This is to
ensure that overarching behaviour patterns that the scales represent
can be related to behaviour logging.
• Length of the test. A lengthy and time consuming test means less
willing participants.
• Established test, such that it has enough empirical backing to ensure
the validity of our data.
Most of the related work uses the Five Factor Model (FFM) or a test
based on the FFM to measure personality dimensions [29, 1, 28]. Other
options were present in the related work such as the BrainHex [22] which is
specifically aimed at games, the downside is that it does not have specific
behaviour related scales and presents a type based result which definitions
are too broad, the empirical backing is also still being acquired.
Another option was the Meyer-Briggs Type Indicator (MBTI). This personality test has a lot of empirical backing but presents a type model. In
game studies the comparative failure of trait models is less than those of
type models [5]. Because type models define a range rather than a point on
a scale it could introduce noise into our analysis.
As such the FFM stands out, the NEO-PI-R version [7] in particular
since it has enough specificity with six facets for each of the five dimensions
of the FFM and an abundance of empirical backing. The NEO-PI-R is a
commercial product however and has a very lengthy test. Thus we opted for
36
a publicly available version that measures the same constructs as the NEOPI-R but with a bit less precision and a third of the amount of questions.
The IPIP-NEO-120 [16] is a personality test that consist of 120 questions
which all give a score relating to one of the five traits and one of the trait’s
facets.
6.2.1
FFM Traits
The FFM models personality using five traits and explained using the
OCEAN mnemonic as explained in [29]:
• Openness: This factor refers to openness to experiences. People with
high scores tend to be imaginative, prefer variety, have aesthetic appreciation, show intellectual curiosity (not necessarily intelligent), keep
close attention to their emotions, and embrace new ideas. On the other
side of the spectrum, people with low Openness scores are more conventional, prefer known situations, the expression of emotions are minced,
and their array of interests is narrower.
• Conscientiousness: This trait is related to the capability of self-control.
When a person has a high Conscientiousness score, she is well organized
and reliable, can identify clear objectives, plan accordingly, and has
strong will. A low score indicates that people have more difficulties in
achieving goals, involving weak-will, and lower organization skills.
• Extraversion: This factor addresses an array of sociable styles. High
scores are associated to a preference for larger groups, being talkative
and active, engaging with people, talking assertively, adopting optimistic and cheerful postures. An introvert is not unfriendly but reserved,
this person would prefer to be independent rather than following the
flock of people, and would keep an even pace as opposed to being slow.
• Agreeableness: This factor gauges the altruistic tendency of people.
A high score is related to empathy, to being helpful to others and
expecting that others will be helpful as well, and to easily development
of trust, whereas a low score is associated to competition rather than
helpfulness, to being suspicious, and to the tendency of prioritising
one’s intentions.
• Neuroticism: The Neuroticism factor covers emotional stability. A
person with a high score in this factor is prone to experience negative
states such as stress, fear, guilt, anger, shame, aversion, and depression.
37
Neuroticism
Extraversion
Openness To Experience
Anxiety
Friendliness
Imagination
Anger
Gregariousness
Artistic Interests
Depression
Assertiveness
Emotionality
Self-Consciousness Activity Level
Adventurousness
Immoderation
Excitement-Seeking Intellect
Vulnerability
Cheerfulness
Liberalism
Agreeableness Conscientiousness
Trust
Self-Efficacy
Morality
Orderliness
Altruism
Dutifulness
Cooperation
Achievement-Striving
Modesty
Self-Discipline
Sympathy
Cautiousness
Table 11: IPIP personality facets for the big five traits.
Whereas low scores can be interpreted as a relaxed person who has
a stable temper and can maintain this attitude when coping tense
situations.
6.2.2
FFM Facets
Each trait is divided into six facets, these facets reflect parts of the traits
that they represent. In our research we try to narrow down any effect that
we find on a trait. To see if the behaviour that is tied to the effect can be
explained using one of the facets.
The facets are all defined using a single word which and very straightforward in their interpretation. For every trait we display the IPIP NEO-PI-R
facets in table 11.
6.3
Assumptions and Hypotheses
In this section we will sum up the most relevant assumptions that have helped
shape this research, after stating the assumptions we give our hypothesis for
the data analysis based on the assumptions.
38
6.3.1
Game Literacy
Game literacy is the knowledge and experience a player has about games.
A high game literacy means that a player has learned many tricks, knows
conventions of game developers, video game tropes in both game mechanics
and flow of a game. Having knowledge about a challenge before it is presented will certainly help in completing that challenge. We therefore assume
that players that have many hours played per week and those that explore
many different genres of games will have a higher game literacy than those
who don’t. Our hypotheses are therefore that the players who play many
hours a week will play longer than those that play a few hours a week. Those
that have played many games of the action RPG genre will have a better
performance than those who have not.
6.3.2
Personality Influence
Personality traits and facets can give us insight into the motivations of behaviour, if the found behaviour matches the behaviour that can be extrapolated from a personality trait or facet, then those personality traits or
facets could be used to predict certain behaviour in game and possibly the
resulting performance or preferences.
Openness to experience has already been linked to game literacy in other
research, most hardcore players are both introvert and have a high openness
to experience1 . In an experiment using a Neverwinter Nights scenario as an
alternate way of profiling the player within the FFM model [28], the players
which completed the game quickly also showed high scores in Openness to
experience. We expect the players with high openness to experience that are
interested in games will perform better than those that have low openness
to experience but still interested in games.
Conscientiousness is the innate drive to micromanage and strive for efficiency, this could indicate that achievements, badges and on screen objectives
are likely to have a positive effect with people that score average to high on
the conscientious scale. As well as having an inventory or statistics to manage. Recklessness associated with a low Conscientious score could require a
more lenient but higher paced gameplay with less elements to manage. As
such we hypothesise that a player with a high conscientiousness score will
use more items and takes a more careful approach to puzzles and fights.
1
The link between openness to experience and game literacy is likely due to the time
that someone has invested in seeking new experiences, if a person has an interest in video
games and has a high openness rating then it is likely that the person has played many
different games over the years.
39
Extraversion has been linked to more energetic game-play, as well as
more involvement in conversations, as seen in [28]. High extraversion also
relates to the preference to large social groups which could be simulated by
adding bots with which the player should work together with and scaling
the difficulty. We hypothesise that more energetic game-play indicates more
movement, more exploration, more interaction with NPCs.
Agreeableness has also been correlated to the amount of involvement in
conversations and courteous behavior in games, as well as the amount of
instructions that is gathered during play. We predict that a player with a
high agreeableness score will have less trouble finding their way around town
and as a result spends less time in the village and finds more items than a
player with a low score.
Neuroticism as measure of resistance to frustration is a good tool to have
when deciding on starting difficulty, difficulty curve and the general amount
of challenges to present to the player. The hypothesis is that a player with
a low neuroticism score will be less likely to quit after failing a challenge or
when a challenge is tedious or arduous.
6.4
Difficulty Preference
The requirement for getting into a flow is a balance of skill and challenge
[8]. If the challenge is too low or too high then that will cause apathy or
frustration respectively and reduce the flow experience. If the player gets
into a flow, the sense of time is reduced and the player is more likely to
spend more time on a game.
Hardcore gamers and casual gamers each have their own level of skill.
Balancing the difficulty would therefore be required for players of different
skill levels to reach a state of flow. Hardcore gamers would have had more
practice and more knowledge on games and as such could handle a higher
difficulty level than the casual gamers.
Our hypothesis is therefore that a player that performs well will like
higher difficulty levels and players with a lower performance will like lower
difficulty levels.
7
Method
In this section we will describe the methodology used in this paper. This includes introducing all the elements of the experiment, discussing the design,
explaining the procedure of the experiments, the data that we collect and
how we will process that data.
40
7.1
Method Overview
The following flowchart will give an overview of the different parts of this
experiment and how they relate to each other.
Figure 15: The method overview. Our participants which we have acquired
through acquisition methods went to the website and followed the research
protocol by following the directions on the website. The data that has been
gathered will be analysed based on our design and data categories.
In the sections to come we will go over each topic displayed int the
method overview (figure 15). The following topics will be discussed:
• Acquisition, how we have acquired participants for our research and
the amount of participants in each group.
• The measurement tools, the tools that we use to gather our data. This
includes the demographic questionnaire, the personality test, the game
that is played and the post-game questionnaire.
• Design, the groupings that are relevant to our experiments.
• Procedure, which contains the following topics:
– Research protocol, each step of the experiment that the participant goes through.
– Measurements, groups of dependent variables that we have measured.
41
– Analysis, our strategy to analyse the data that we gathered. Describing which data groups we will compare with each other with
certain tools.
7.2
Participants and acquisition
The participants were recruited for this research via personal connections,
flyers and personal handouts of flyers. The flyers particularly asked for
gamers with some gaming experience. The flyers were handed out and
spread around at the University of Amsterdam. In total 25 participants
had completed all the instructions. 21 participants were male and 4 were
female, the average age was 25.52. All participants have either experience
playing games on a desktop, laptop or console. Of these 25 participants, 16
have been presented the adaptable difficulty version and the remaining 9 the
static difficulty version, assignment of the versions was done randomly.
Figure 16: Demographic data bar plots.
42
7.3
The Demographic questionnaire
This questionnaire was used to determine the demographic of the participant,
which included the participant’s age, gender, time spent gaming, game genre
preference and if they had played The Legend of Zelda: A Link to the Past.
This questionnaire is also used to determine whether a participant belongs
to the casual or average to hardcore gamer group. A full description of the
demographic questionnaire is available in section A of the Appendix.
7.4
The personality test
The test is 120 questions large and was designed to capture the NEO-PI-R
facets with a high correlation between constructs [16]. The test includes
a detailed explanation of how the participant is supposed to interpret the
questions and what the scale for each question represents. The questions
were interleaved such that there is no question that is followed by a question
of the same category.
A few examples of the questions that are present in the personality test:
• Questions that affect the score of the conscientiousness facet
achievement-striving:
– Do more than what’s expected of me.
– Work hard.
– Put little time and effort into my work.
– Do just enough work to get by.
• Questions that affect the score of the neuroticism facet immoderation:
– Go on binges.
– Rarely overindulge.
– Easily resist temptations.
– Am able to control my cravings.
• Questions that affect the score of the extraversion facet activity level:
– Am always busy.
– Am always on the go.
– Do a lot in my spare time.
– Like to take it easy.
43
Questions are represented in a 5-point Likert scale ranging from very
inaccurate (1) to very accurate (5), to get the score for a particular facet
or dimension we add together all the scores of the related facet/dimension.
Some questions are negatively keyed, which means that the point scale is
reversed (5 - 1).
A full description of the personality test can be found in section B of the
Appendix.
7.5
Dynamic Zelda
The game that was played by the participants, we gather data during play
and direct mapping of the designated game metrics to log files. We gather
behaviour data mainly from the village, fights and level exploration:
• Village behaviour data involves how much the participant engaged in
conversation and how much preparations the participant had made in
the village before heading out to the levels.
• Exploration behaviour data involves which areas the player had visited
in the village and how much the player went off the main path in the
levels.
• Fight behaviour data involves movement, usage of items or abilities as
well as sword usage.
We also gathered performance data from the puzzles and fights, the
metrics are mostly focused on how much health the player had lost during
the fight or puzzle combined with the time it took to complete it.
And lastly we collected level ratings for the experienced difficulty by
using a short questionnaire with 5 questions.
Figure 17: Screenshot of the in-game questionnaire.
44
A full description of the game is available in section 5 and a full description of the game metrics is available in section C of the Appendix.
7.6
The Post-game questionnaire
This questionnaire was used to support the game metrics that were gathered
and to explicitly state the preference and the perceived experience of the participant regarding all parts of the game. This includes questions about the
overall difficulty of the fights and puzzles and whether the participant also
thought that this difficulty fit this type of game. We also asked introspective
questions regarding their own in-game behavior within the village, fights and
exploration.
Almost all the questions that were posed in the post-game questionnaire
were of the same nature as the in-game metrics and are only used as support
if the in-game metrics are not clear enough.
The most important question in the post-game questionnaire is to give
the ranking of the preference for each game element (fights, block-pushing,
moving floor, maze, exploration, talking to NPCs), ranking the most preferred with 1 and the least preferred with 6.
A full description of the Post-game questionnaire is available in section
D of the appendix.
7.7
Design
The experiment used a 2×1 between-subjects design. The independent variable was the static vs adjustable difficulty. Static vs adjustable gives us
an indication if the current execution of the adjustable difficulty adds more
value to the game’s experience than a linearly scaling difficulty curve. Casual
gamers and average to hardcore gamers are not split because of the existing
distinction is unclear and relative and only the participants which clearly
fall into one or the other category will be compared later during analysis.
The participants are randomly assigned to either the static or the adjustable difficulty group to avoid assignment bias. We sadly cannot avoid
any individual variability or environmental bias because of the nature of the
experiment (at home when the participant has enough time and want to
play). The random assignment might cause an unbalanced group that could
open the door for generalization issues.
The between subjects design is only to test the static vs adjustable
difficulty, when we compare our user profile data with our game metrics
we consider both groups equivalent and process them as one group. We
45
assume that the difficulty will have the most effect on performance metrics
and we also need data where the difficulty is below or above the player’s
level of skill, where a properly working adjustable difficulty setting would
deny us that data.
7.8
Procedure
In this section we will discuss the procedure of the experiment, the protocol
that was followed for each participant during the experiment, the measurements that were made and the analysis strategy that will state what we will
test for and what kind of data we will be using during these tests.
7.8.1
Research Protocol
During acquisition potential participants were asked if they play a video
game once in a while, an explanation of the research followed where the
intent behind the research was stated, the steps required were mentioned
and that the participants would be able to participate in their own time.
Participants were asked to visit a website that explained the details of
the experiment. These details included the reasoning behind the experiment,
an explanation of the used game engine, the current version of the game,
the game’s control schema and the steps required for participation with
time estimations for each step. Participants were also informed that they
could spread out the participation in the research if necessary, but that
participation in one sitting was preferred. A full description of the website
can be found in section E of the Appendix.
The steps required for participation were visiting the website and completing the following steps:
• Fill in the online Demographic questionnaire.
• Fill in the online Personality test.
• Downloading, installing and playing the game until they didn’t feel
like playing or finished the game.
• Mailing the logs to a designated e-mail address.
• Fill in the online Post-game questionnaire.
When a participant had sent all the required logs to the e-mail address
and had filled in all questionnaires, the participant then received his personality test results via e-mail.
46
7.8.2
Measurements
The dependent variables that were measured for every player are:
• Preference regarding parts of the game (Fights / Dark maze /
Block-pushing puzzle / Moving-floor puzzle / Talking to characters /
Walking around and exploring).
• Performance of the player regarding the puzzles (Dark maze /
Block-pushing puzzle / Moving-floor puzzle).
• Performance and behavior of the player during the fights.
• Behavior of the player in the village.
• Exploration data for each segment.
• The participant’s personality score.
7.8.3
Analysis strategy
For our analysis we look for strongly correlated variables using the Spearman
Rank-Order Correlation Coefficient. We are not looking for direct cause but
primarily for monotonic relationships by which the participants’ preferences,
behaviour or performance can be predicted or explained using the personality
trait and facet definitions. We will also be comparing distributions using the
Kolmogorov–Smirnov test. The following combinations will be used with the
above tests:
• Personality dimensions and facets correlations with performance, preference and behaviour variables.
• Performance correlations with player preference and perceived difficulty.
• Comparing casual player preference distribution with non-casual player
preference distribution.
• Comparing the ratings distribution of the static difficulty and the adaptable difficulty setting.
To link personality with preference we will investigate the correlations
and try to draw meaningful conclusions from our data given our definitions
and assumptions.
47
By comparing ratings distributions of our adaptable difficulty setting
and the static linearly scaling difficulty setting we can give an indication of
whether or not our difficulty adjustment has made an improvement to our
game’s experience.
If any interesting relationships between personality and preference arise
we can investigate whether the personality trait can be predicted with any of
our game metrics and as such provide a meaningful way of profiling a player’s
personality through game metrics and consequently predict the preferences.
8
Experiments
In this section we will present the results from our experiments that are
relevant to any research question or hypothesis that was presented earlier in
the paper in the research questions section (1.1) and the assumptions and
hypotheses section (6.3). The results are then followed up by a short analysis
on the topic of the relevant research question or hypothesis.
Experiment 1: The difficulty level of the fights and puzzles is predetermined per level.
Experiment 2: The difficulty level for a game element (fight or puzzle)
is based on the participant’s performance for that particular game
element.
For our experiments we asked participants to follow the instructions on a
website, these instructions included filling in questionnaires, playing a game
and sending the log files after quitting or completing the game. The allotted
time requirements were somewhere between an hour and two hours and our
instructions were to be followed in a specific order, preferably when the
participant had enough time to complete the instructions in one sitting.
We could see in the acquired data and participants’ responses that there
were still some issues with the game after presenting it to our participants,
these issues were usually fixed straight away and the participant could resume playing. This will most likely introduce some noise. Some participants
had quit when a crash occurred, some when the game became too difficult,
some out of boredom. The break from the instructions meant that some
forgot about the instructions after the game and forgot to send any log files
or complete the post-game questionnaire, the data from these participants
has been left out of the dataset.
48
8.1
Exploration Method
Our data was first subjected to the Shapiro-Wilk test on each measured
variable. From this data we could see that the majority of the data is
skewed in either direction and that we could reject the hypothesis that the
data came from a normal distribution. This means that we cannot use
any statistical tools that have an assumption of normality. Thus the use
of Pearson product-moment correlation coefficient and Principle Component
Analysis is not advised.
We will instead keep to investigating possible monotonic relationships
by using Spearman Rank-Order Correlation Coefficient on the variables
that were mentioned in the hypotheses section. We will explore our research questions using Fast Independent Component Analysis (python library: sklearn.decomposition.FastICA with standard settings), where the
number of components are approximated by using the number of components
that explained 90% of the variance in the output of Principle Component
Analysis (python library: matplotlib.mlab.PCA).
The only part that is regrettable is that ICA in general is not a deterministic algorithm and that we might not catch all the components or we could
get some small artifacts in our results.
The FastICA exploration will be done by correlating the resulting components of the FastICA algorithm with the actual data, we can then see
which of the components correlate strongly with each other. We take the
strongest correlating variables of a component and try to qualitatively give
an explanation to the correlation between these selected variables. In most
cases found below we will only discuss the interesting components that have
strong correlating variables outside their variable group (e.g. strong correlations between personality dimensions or facets are not of interest, only those
that correlate with a game metric).
But even when the data correlates strongly with the components, the
found component parts (e.g. personality and game metric) should also
strongly correlate with each other, because this is the part that we are interested in.
8.2
Gamer type results
We have hypothesized that the hardcore gamer preference would diverge
from the casual gamer preference. To test this we have compared the distributions of preference between participants using the Kolmogorov–Smirnov
test, the results can be seen in figure 18.
49
Within our data we define hardcore gamers as followed: the participant
spends a considerable chunk of time on games (10+ hours per week) and has
a preference for somewhat challenging content (hard difficulty or higher).
Casual gamers are defined as having an interest in games, but not wanting
to put in time and effort to get really good at a game, with often a preference
for simple games. Within our data that is defined as less than 10 hours per
week and Easy difficulty or lower.
We also have a gray area where we have potential hardcore gamers that
enjoy difficult games but don’t have or invest a lot of time to play, or the
possibly casual gamer that puts in considerable amounts of time but doesn’t
want a challenge.
Figure 18: Boxplots of the preference ranking distribution of participants
that classify as hardcore, casual or as unknown. Rank 1 is most liked, rank
6 is least liked. Only the maximum and minimum outliers are shown as
circles. The green triangles represent the mean.
As can be seen in figure 18, the hardcore gamer preference for the moving
50
floor puzzle is ranked better than the moving floor puzzle ranks for the casual
gamers, but most likely due to the outlier in the hardcore gamer pool we
cannot say for certain that they are strictly different distributions (p = 0.086
two-sided). Variables other than the moving floor puzzle preference have a
too high of a p-value to consider the possibility that their distributions are
different for hardcore gamers and casual gamers.
Though we cannot reject the hypothesis that they are of the same distribution, the reason why the moving floor puzzle ranks worse for the casual
gamers is the nature of the puzzle, it requires more finesse with controlling
the character than the other game elements, and this is where the hardcore
gamers have the advantage.
8.3
Difficulty setting results
To investigate if the adaptable setting was an improvement on our static
setting we will again use the Kolmogorov–Smirnov test to determine if the
adaptable setting distribution is significantly different from the static setting.
The boxplots of the two distributions can be seen in figure 19
Figure 19: Boxplots of the average Likert score of the participants on whether
they liked the amount of challenge presented for the fights and puzzles, the
overall experience of the levels and the rating of the amount of exploration
available. The Likert scale ranges from strongly disagree to strongly agree.
Only the maximum and minimum outliers are shown as circles. The green
triangles represent the mean.
51
Variable
Overall experience
Fights
Puzzles
D
0.224
0.125
0.150
p
0.921
0.999
0.999
Table 12: Two sample KS tests between the average Likert score of liking
the following: the game elements difficulty for fights and puzzles, the overall
experience of the levels and the rating of the amount of exploration available.
Likert scores are from the ingame questionnaire.
Figure 20: Boxplots of the average Likert score of the participants on whether
they found the difficulty too hard for the fights and puzzles and whether the
puzzle difficulty fits this type of game. The Likert scale ranges from strongly
disagree to strongly agree. Only the maximum and minimum outliers are
shown as circles. The green triangles represent the mean.
52
Variable
Fights too hard
Maze puzzles too hard
Moving floor puzzles too hard
Block-pushing puzzles too hard
Puzzle difficulty is fitting
Puzzles too easy
D
0.326
0.562
0.138
0.215
0.437
0.152
p
0.484
0.031
0.999
0.920
0.160
0.997
Table 13: Two sample KS tests between the average Likert score of find
the following elements too hard: The difficulty. Likert scores are from the
postgame questionnaire
8.4
Personality results
In figure 21 and 22 we can see the range of the measured personality dimensions of our participants. We can see that our population has a skewedness
towards the higher ends for all of the dimensions except neuroticism. The
skewedness could be an indication of the kind of persons that would be more
likely to participate in lengthy research. This could have had an influence
on our dataset, given that extremes in personality are also more extreme
in their actions, which means that we do not have participants that are exceptionally competitive or conservative in their experiences, nor do we have
that many reckless persons. But none the less we will try to find meaningful
relations within our given dataset.
Figure 21: Boxplot of the measured personality dimension scores. Green
triangles represent the means.
53
Figure 22: Boxplots of the measured personality facet scores. Green triangles
represent the means.
54
8.5
FastICA and Spearman’s rho exploration results
To find support or arguments against our hypotheses we will look at the
FastICA results and see if any of our hypotheses match a found correlation.
In our FastICA exploration we will use the most generic of our game metrics,
this will cause our correlations to be less diluted and we can report on metrics
that could also be found in other games of this genre.
Some of the components that were found have a very small fraction attributed to one of the groups (Personality scores or game metric/element
rank), this most likely indicates an artifact which only indicates correlation
within a group and should be disregarded. We display all the correlated components parts that have a p-value less than 0.15, considering our data and
research objective we should not discard the somewhat weaker correlations
without thought.
Do not pay a lot of mind to the low modifier values on some of the
variables, by using Spearman Rank-Order Correlation Coefficient our correlation is unaffected by low or disproportionate numbers, we should only
pay attention to the proportions of each variable within a group because this
has influence on the ranking of the data.
8.5.1
Gamer type and personality
We first explore how our demographic metrics correlate with our participants’ personality scores, the strongest correlations can be found in table
14. All of the components found are very interesting for our research.
We can see that conscientiousness negatively correlates very strongly
with all of the hardcore gamer type metrics, which indicates that within
out group of participants the conscientious participants were much less into
games.
We can also see that the strongest correlation within the conscientiousness facets is the negative correlation between achievement-striving, selfdiscipline and many of the hardcore gamer related variables, it is expected
that these scores are like this, as hardcore gaming takes up a lot of time the
achievement striving facets will be scores low with statements in the personality test such as ”Do just enough work to get by”. The same goes for
self-discipline with statements such as ”Waste my time”. As played hours
increase so does practice, as such the difficulty preference is also higher.
The negative relation between achievement striving and the many different genres played and playing games within favorite genre tells us that
low achievers among gamers play more game genres and have developed a
55
favorite genre which is also in line with the hardcore gamer.
Personality facets/dimensions
(-0.067 · C4:Achievement-striving)
+(-0.029 · C5:Self-discipline)
(-0.009 · N1:Anxiety)
+(-0.032 · N3:Depression)
+(-0.021 · N4:Self-consciousness)
+(-0.13 · N5:Immoderation)
+(-0.018 · N6:Vulnerability)
(0.068 · O2:Artistic Interest)
+(0.104 · O3:Emotionality)
+(-0.075 · O6:Liberalism)
Gamer type variables
(0.097 · Hours Played)
+(0.038 · Difficulty Preference)
+(-0.019 · Many different genres)
+(0.019 · Within favorite
genre)
+(0.02 · Action Adventure)
(-0.107 · Difficulty Preference)
+(-0.055 · Many different genres)
+(-0.002 · played Zelda:ALTTP)
rho
0.768
p
p<0.0005
0.402
0.047
(-0.076 · Within Fav genre)
0.347
0.089
Table 14: FastICA components that weakly correlate the gamer type metrics with the personality facets and dimensions. Gamer type metrics: Hours
Played, Difficulty Preference, Many different genres, Within Fav genre, Action Adventure, played Zelda:ALTTP.
The second strongest correlation is also interesting, it points towards
neuroticism as an indicator of difficulty preference as well as the lack of
interest in playing different genres.
The individual correlations are displayed in table 15. In the table we
can see why immoderation is a large part of the component, it correlates
strongly with hardcore gamer metrics and playing the action adventure genre
in particular. The other neuroticism facets did not correlate significantly
with difficulty preference, as such immoderation is likely an expression of
being a hardcore gamer where playing for long hours will categorize as binges
which brings practice which in turn might increase the difficulty preference
due to the amount of practice.
The third correlation is interesting but not as as clear as the other correlations in the table. Instead we took a look at the individual correlations
between variables of openness to experience and our gamer type variables.
Intellect is strongly correlated with difficulty preference, intellect is measured with statements in the personality test such as: ”Have difficulty understanding abstract ideas”. This indicates that participants that have a higher
56
intellect score will most likely have less trouble grasping the conventions used
in games and as such will require more challenging material for it to be as
rewarding.
57
Facets
Neuroticism
Extraversion
Openness to Experience
Agreeableness
Conscientiousness
O1:Imagination
O2:Artistic Interest
O3:Emotionality
O4:Adventurousness
O5:Intellect
O6:Liberalism
C1:Self-efficacy
C2:Orderliness
C3:Dutifulness
C4:Achievement-striving
C5:Self-discipline
C6:Cautiousness
E1:Friendliness
E2:Gregariousness
E3:Assertiveness
E4:Activity Level
E5:Excitement-seeking
E6:Cheerfulness
A1:Trust
A2:Morality
A3:Altruism
A4:Cooperation
A5:Modesty
A6:Sympathy
N1:Anxiety
N2:Anger
N3:Depression
N4:Self-consciousness
N5:Immoderation
N6:Vulnerability
Hours
played
Difficulty
preference
0.193
-0.378**
-0.018
0.169
-0.66****
0.076
-0.306*
-0.064
-0.05
0.15
0.02
-0.429***
-0.404***
-0.098
-0.768****
-0.539****
-0.079
-0.199
-0.318*
-0.404***
-0.503***
-0.2
-0.134
0.234
0.075
0.126
-0.145
0.149
0.071
0.07
0.132
0.069
0.3*
0.386**
-0.033
0.213
-0.019
0.155
0.057
-0.442***
0.192
-0.099
-0.054
-0.055
0.44***
0.135
-0.213
-0.226
-0.254
-0.45***
-0.509****
-0.202
-0.065
-0.075
-0.046
-0.265
0.118
0.218
0.007
0.008
0.08
0.066
-0.005
-0.021
0.085
0.1
-0.039
0.271
0.474***
0.148
Many
different
genres
0.221
-0.291
-0.025
-0.018
-0.448***
0.178
-0.301*
-0.158
-0.08
0.165
0.041
-0.389**
-0.338**
-0.385**
-0.43***
-0.264
-0.279
-0.374**
-0.195
-0.17
-0.278
-0.214
0.075
-0.086
-0.097
0.038
-0.028
0.079
0.032
0.108
0.115
0.163
0.233
0.197
0.148
Within
favorite
genre
0.017
-0.287
0.055
0.144
-0.416***
0.12
-0.371**
-0.188
0.24
0.094
0.182
-0.294
-0.365**
-0.163
-0.637****
-0.154
-0.119
-0.387**
-0.158
-0.245
-0.182
-0.235
0.04
0.035
0.106
0.079
0.113
0.239
-0.054
-0.078
-0.05
-0.002
0.185
0.112
-0.103
Action
Adventure
genre
0.316*
-0.115
0.149
-0.128
-0.259
0.457***
0.116
0.113
-0.224
0.34**
-0.087
-0.413***
0.142
-0.461***
-0.246
-0.497***
-0.064
-0.032
-0.075
-0.133
-0.363**
0.066
0.088
-0.116
-0.158
0.09
-0.094
-0.081
0.256
0.24
0.292
0.21
0.163
0.572****
0.123
Table 15: Moderate to strong correlations between personality facets and
the gamer type metrics. *: p<0.15, **: p<0.1, ***: p<0.05, ****: p<0.01
58
8.5.2
Game literacy and performance
In the tables below we have correlated the difficulty reached for puzzles
and the predicted fight difficulty for the adaptable difficulty setting. In the
adaptable difficulty setting we could properly measure how skilled a participant is by measuring the average difficulty level on which the participant
was playing.
Both table 16 and 17 state that having played The legend of Zelda: A link
to the past together with the difficulty preference are the strongest indicators
for performance, after which those who play Action Adventure genre games
will have some familiarity with the Moving-floor type puzzles.
The fight element’s predicted difficulty is our weakest measurement, because the difficulty level is kept relative to ensure the oscillating flow mechanic works properly. Since each monster has a separate weight and relative
difficulty it is also not correct to take the weights as measurements of total
difficulty. The predicted difficulty was often very high when the player was
playing well and as a result a lot of enemies were spawned. As such it is
partly indicative of higher performance but also somewhat noisy since a bad
player can get a high predicted fight difficulty as well, but not that easily
and not as high as when a lot of enemies spawn.
Gamer type metrics
(-0.06 · Difficulty Preference)
+(-0.065 · Action Adventure)
+(-0.207 · played Zelda:ALTTP)
(-0.014 · Difficulty Preference)
+(0.022 · Many different genres)
+(0.112 · Action Adventure)
+(0.104 · played Zelda:ALTTP)
(0.141 · Many different genres)
+(0.102 · Within Fav genre)
+(0.068 · Action Adventure)
+(0.107 · played Zelda:ALTTP)
(0.091 · Action Adventure)
Difficulty reached
(0.048 · Block-pushing difficulty)
(0.084 · Moving floor difficulty)
(-0.01 · Predicted fight difficulty)
(0.242 · Maze difficulty)
rho
-0.706
p
0.002
0.723
0.002
-0.576
0.02
0.321
0.225
Table 16: FastICA components that strongly correlate the gamer type metrics with the difficulty reached for the adaptable difficulty setting. Gamer
type metrics: Hours Played, Difficulty Preference, Many different genres,
Within Fav genre, Action Adventure, played Zelda:ALTTP.
59
Game element
Hours
played
Difficulty
preference
Block-pushing
Maze
Moving-floor
Fights
0.286
0.391*
0.043
0.463**
0.55***
0.509***
0.455**
0.198
Many
different
genres
0.077
0.369
0.136
0.365
Within
favorite
genre
0.224
0.422*
0.024
0.4*
Action
Adventure
genre
0.413*
0.321
0.544***
0.164
Played
Zelda:ALTTP
0.511***
0.442**
0.741****
0.467**
Table 17: Moderate to strong correlations between the gamer type metrics
with the difficulty reached for the adaptable difficulty setting. *: p<0.15,
**: p<0.1, ***: p<0.05, ****: p<0.01
8.5.3
Game element preference
In this section we use the FastICA algorithm to explore the possible link
between the game element preference ranking and the performance metrics
and personality dimensions and facets. If necessary we investigate further
by using the individual correlations.
Game element rank
(-0.016 · Maze)
(0.051 · Fight)
Performance metrics
(0.096 · time spent)
+(-0.041 · got hurt)
+(-0.115 · falls)
+(-0.019 · Reached difficulty)
(-0.2 · hurt)
+(0.066 · lift)
+(0.014 · useItem)
rho
0.25
p
0.239
0.043
0.837
Table 18: FastICA components that correlate the performance game metrics
with the game element ranking. Sokoban metrics: time spent, completed,
retries, vfm time (time before the first move), reached difficulty. Maze metrics: got hurt, falls (fell into a pit), time spent, reached difficulty. Moving
floor metrics: got hurt, time spent, reached difficulty. Fight metrics: hurt
(times player got hurt), lift (lifting a rock or bush to throw), useItem (using
an item from the player’s inventory), spin (special attack with the sword),
predicted difficulty.
The found components for performance can be seen in table 18, all of the
p-values are high and as such we can conclude that (relative) performance
does not have a link with preference. To further give arguments that performance has no link to preference we also calculated the correlations between
60
the achieved difficulty level and preference ranking in the adaptable setting
which can be seen in table 19.
Game element
Block-pushing
Maze
Moving floor
Fights
rho
0.062
0.039
0.168
0.002
p
0.82
0.886
0.534
0.995
Table 19: Correlations between the game element’s preference rank and the
game element’s achieved difficulty level for the adaptable difficulty setting.
In table 20 we can see the the correlation between the strongest component parts for personality dimensions/facets and game element ranking.
We can see that there are some interesting correlations which might indicate
that there is a link.
The most interesting ones are the ones that could be used to get insight into the player or be used as a predictor for preference. For instance
the decreased preference for talking to NPCs with decreasing achievementstriving and increased preference for the moving-floor puzzle. This could be
the result that secrets and bonuses in other games are often gained through
conversation with NPCs, and that the convention now results in an increased
preference. Since these facets also correlate strongly with hardcore gamers
another possibility is that the hardcore gamer participants feel comfortable
with that kind of information gathering and casual gamers less so. A low
score for achievement-striving also indicates a more hardcore gamer type and
since the required skill with the controls is higher than the other puzzles.
The higher requirement could indicate that participants that did not have
this required skill liked it a lot less than participants that could properly
complete it without getting their character hurt too much.
Another example from the table is the increase in extraversion facets
that relates to the decreasing preference for fights, we would expect that a
person that is more assertive, active or excitements-seeking would be more
attracted to fights, not less.
We also see that many strong correlations are present that include agreeableness or its facets, comparing the found component parts to the definition
of agreeableness has not given us much insight into the preference other than
that if the agreeable participants try to follow up on instructions quickly then
they’ll spent less time on talking to NPCs even if they would have liked it if
they did, which might cause a change in preference ranking.
61
We further investigated personality and game element ranking with individual correlations, see table 21. In this table we can find many strong
correlations between personality dimensions/facets and game element ranking, we will discuss all the correlations that have a p-value of less than 0.05.
Just like in table 20 we see that the moving floor and talking to NPCs
preference correlates with the agreeabless dimension as well as achievment
striving correlate with moving floor and talking to NPCs.
Imagination correlates strongly with the preference for exploration, which
would fit the description of a daydreamer or someone who likes to get lost
in thought.
Assertive participants like the dark maze more, or maybe it is more
accurate to say that less assertive participants found the dark maze to be
disheartening.
Trust correlates strongly with the moving floor puzzle, one explanation
could be that if the trust facet would have captured the participant’s inclination towards finding patterns and believing in them, since the moving
floor puzzles variations all have a distinct pattern, making it easier and more
properly solvable by someone that has the pattern down, otherwise such a
strong correlation with such a social facet makes little sense.
Modesty correlates strongly with the block-pushing puzzle preference.
Again a primarily social facet correlates with a puzzle preference, one explanation could be that a person who thinks highly of themselves finds extra
enjoyment from the eureka sensation that these kinds of puzzles bring compared to other participants that might not feel the need to strengthen their
own belief that they are better than others.
Sympathy correlates strongly with the dark maze preference. Another
social facet that correlates with a puzzle type. Sympathy is a personality
trait that speaks of emotionality when another person or being is in trouble
or in distress. Whether the emotionality is a rewarding sensation or not
is unclear, but if it is then that emotionality might also be triggered by
dark maze setting. Other than that the description of the facet nor the
test statements on which the facet is based bring insight into this strong
correlation.
Anger correlates strongly with the block-pushing puzzle. An explanation
could be that the more angry a person becomes when not being able to solve
a block-pushing puzzle the more contrast there is in their emotions when
they do get that eureka sensation, which they experience more strongly then.
Another explanation is that the participants that are easily frustrated quit
sooner into the game and never get to a hard enough block-pushing puzzle
to cause them to get irritated, causing it to rise in preference by the decrease
62
in preference of other elements.
63
Personality facets/dimensions
(-0.002 · C4:Achievement-striving)
(-0.152 · A2:Morality)
+(-0.118 · A4:Cooperation)
(-0.102 · A3:Altruism)
+(-0.094 · A6:Sympathy)
(0.032 · A1:Trust)
+(-0.042 · A4:Cooperation)
+(-0.068 · A5:Modesty)
+(0.009 · A6:Sympathy)
(-0.026 · Agreeableness)
(-0.058 · Agreeableness)
(-0.086 · E3:Assertiveness)
+(-0.061 · E5:Excitement-seeking)
+(0.08 · E6:Cheerfulness)
(-0.01 · N1:Anxiety)
+(-0.016 · N2:Anger)
+(-0.01 · N3:Depression)
+(-0.022 · N6:Vulnerability)
(-0.005 · Openness to Experience)
+(-0.004 · Agreeableness)
(0.021 · Neuroticism)
+(-0.025 · Extraversion)
+(0.008 · Conscientiousness)
(-0.039 · C1:Self-efficacy)
+(-0.123 · C4:Achievement-striving)
+(-0.054 · C5:Self-discipline)
(0.139 · A1:Trust)
(-0.064 · E1:Friendliness)
+(-0.041 · E2:Gregariousness)
+(0.023 · E4:Activity Level)
+(-0.046 · E5:Excitement-seeking)
+(-0.04 · E6:Cheerfulness)
(0.018 · O1:Imagination)
+(0.086 · O2:Artistic Interest)
+(0.092 · O3:Emotionality)
Game element rank
(0.031 · Moving floor)
+(-0.106 · Talking to NPCs)
(0.111 · Talking to NPCs)
rho
-0.485
p
0.014
-0.478
0.016
(0.031 · Dark maze)
0.466
0.019
(-0.087 · Fights)
+(0.038 · Block-pushing)
0.442
0.027
(0.12 · Moving floor)
(0.072 · Fights)
+(-0.089 · Talking to NPCs)
(-0.093 · Fights)
0.43
0.424
0.032
0.035
0.393
0.052
(0.093 · Block-pushing)
0.379
0.061
(-0.099 · Exploring)
0.365
0.073
(-0.094 · Block-pushing)
0.351
0.085
(0.018 · Talking to NPCs)
0.332
0.105
(0.134 · Fights)
(-0.063 · Block-pushing)
-0.328
0.311
0.109
0.13
(0.101 · Dark maze)
-0.307
0.135
Table 20: FastICA components that strongly correlate preference ranking
with the personality scores. A lower rank means more preferred.
64
Personality
dimensions/facets
Neuroticism
Extraversion
Openness to Experience
Agreeableness
Conscientiousness
O1:Imagination
O2:Artistic Interest
O3:Emotionality
O4:Adventurousness
O5:Intellect
O6:Liberalism
C1:Self-efficacy
C2:Orderliness
C3:Dutifulness
C4:Achievement-striving
C5:Self-discipline
C6:Cautiousness
E1:Friendliness
E2:Gregariousness
E3:Assertiveness
E4:Activity Level
E5:Excitement-seeking
E6:Cheerfulness
A1:Trust
A2:Morality
A3:Altruism
A4:Cooperation
A5:Modesty
A6:Sympathy
N1:Anxiety
N2:Anger
N3:Depression
N4:Self-consciousness
N5:Immoderation
N6:Vulnerability
Fights
Block-pushing
Dark maze
Moving floor
0.015
0.195
-0.048
-0.19
-0.071
-0.074
-0.008
0.014
0.096
-0.069
0.003
-0.067
-0.036
-0.31*
0.123
0.077
-0.107
-0.032
0.158
0.143
0.323*
0.287
-0.099
-0.328*
-0.153
-0.117
0.117
-0.005
-0.141
0.117
-0.024
0.047
-0.006
-0.103
0.074
-0.297*
0.331*
-0.116
0.024
0.305*
-0.163
0.223
-0.107
-0.107
0.045
-0.305*
0.322*
0.127
0.193
0.229
0.131
0.108
0.33*
0.383**
0.303*
0.155
0.229
0.148
0.348**
-0.125
0.054
-0.231
-0.439***
0.322*
-0.21
-0.477***
-0.318*
-0.003
-0.093
-0.312*
0.138
-0.146
-0.281
-0.24
-0.164
-0.244
-0.242
-0.249
-0.283
-0.097
0.275
-0.343**
-0.005
-0.024
0.034
-0.053
-0.132
0.023
-0.237
-0.449***
0.043
0.054
-0.002
-0.075
-0.143
-0.372**
-0.031
0.225
-0.502***
0.048
0.315*
-0.03
0.006
0.18
0.271
0.268
-0.156
-0.11
-0.43***
0.008
0.015
-0.003
0.03
-0.235
-0.254
0.198
0.03
-0.164
-0.174
0.403***
0.139
-0.148
-0.309*
-0.274
-0.046
0.153
-0.263
-0.066
-0.448***
-0.369**
-0.229
-0.044
-0.129
-0.252
0.332*
0.357**
0.167
0.051
-0.222
0.393**
Talking
to NPCs
-0.001
-0.203
0.183
0.401***
-0.213
0.068
-0.015
0.108
0.234
0.04
-0.042
-0.068
-0.104
-0.084
-0.459***
-0.064
-0.107
-0.13
-0.151
0.13
-0.359**
-0.284
-0.182
0.018
0.39**
0.277
0.312*
0.203
0.143
-0.072
-0.051
0.127
-0.093
-0.027
-0.024
Table 21: Correlations between the game element’s preference rank and
personality scores for dimensions and facets. A lower rank means more
preferred. *: p<0.15, **: p<0.1, ***: p<0.05, ****: p<0.01
65
Exploring
-0.034
-0.068
0.371**
0.277
0.103
0.403***
0.031
0.293
0.33*
0.284
-0.172
0.138
0.168
0.138
-0.249
-0.203
0.343**
0.003
0.068
0.061
-0.29
-0.113
0.036
0.198
0.311*
0.331*
-0.05
0.085
0.291
-0.096
0.016
0.143
-0.048
0.181
-0.282
Game element
Fights
Block-pushing
Dark maze
Moving floor
Talking to NPCs
Exploring
Hours
played
Difficulty
preference
-0.176
-0.112
-0.014
-0.425***
0.329*
0.267
-0.023
-0.024
0.127
-0.262
0.175
-0.084
Many
different
genres
-0.015
-0.291
0.051
-0.221
0.458***
0.028
Within
favorite
genre
0.017
-0.38**
-0.094
-0.239
0.501***
0.207
Action
Adventure
genre
0.042
0.084
-0.124
-0.252
-0
0.205
Played
Zelda:ALTTP
Table 22: Correlations between the the gamer type metrics and the game
element’s preference rank. A lower rank means more preferred. Gamer type
metrics: Hours Played, Difficulty Preference, Many different genres, Within
Fav genre, Action Adventure, played Zelda:ALTTP. *: p<0.15, **: p<0.1,
***: p<0.05, ****: p<0.01
In table 22 we can see that there is a correlation between gamer type
metrics and preference, but that this is unrelated to the actual performance.
The block-pushing puzzles were a frequent occurring type of puzzle in
The Legend of Zelda: A link to the past. It makes it likely that nostalgia
then influences the preference of the block-pushing puzzle.
Participants that play games for many hours a week tend to have an
increased preference for the moving floor puzzle. The moving floor puzzle
likely appeals to participants that have a finer control over the character.
Participants that play games often have a decreased preference for talking
to NPCs, this does not mean that they talk to NPCs less, but it could also
mean that they tend to skip a lot of dialog. Participants that play a lot
of games often know that most information attained from the NPCs is not
important for the game, only the act of talking to them because it is often
the trigger for other events.
66
0.115
-0.339**
0.17
0.174
0.138
-0.103
8.5.4
NPC and conversation interaction
Personality facets/dimensions
(0.026 · E1:Friendliness)
+(0.006 · E2:Gregariousness)
+(0.049 · E3:Assertiveness)
+(0.001 · E4:Activity Level)
(0.068 · A4:Cooperation)
+(0.085 · A5:Modesty)
(0.028 · A2:Morality)
+(0.006 · A3:Altruism)
+(0.028 · A5:Modesty)
+(0.006 · A6:Sympathy)
(-0.014 · N1:Anxiety)
+(-0.002 · N2:Anger)
+(-0.028 · N4:Self-consciousness)
+(-0.024 · N6:Vulnerability)
(0.057 · E4:Activity Level)
+(-0.075 · E5:Excitement-seeking)
Game metric
(0.006 · AreasVisited)
rho
0.443
p
0.03
(-0.046 · npcs talked to)
+(-0.106 · AreasVisited)
(0.028 · npcs talked to)
+(0.008 · Extra options taken)
+(0.006 · AreasVisited)
0.509
0.011
-0.419
0.041
(0.01 · npcs talked to)
+(0.011 · AreasVisited)
0.391
0.059
(-0.025 · Extra options taken)
0.335
0.11
Table 23: FastICA components that strongly correlate the village game
metrics for conversations with the personality scores. Conversation metrics:
NPCs talked to, Extra options taken (conversation options taken minus
NPCs talked to). AreasVisited is a combined metric, the sum of all boolean
values of visited village areas.
In table 23 we can see a similar effect regarding Agreeableness and the
preference for talking to NPCs, the cooperation facet is a common factor
in both. One possible explanation is that participants that score high in the
cooperation facet want to start the quest as soon as possible, given that the
entire premise and directions were given at the very start. With regards to
extraversion we have predicted that the play-style would be more active, have
more interactions with NPCs as well as more exploration, the correlation
found in table 23 seems to support the exploration part.
We can also see that a lower score for the neuroticism facets selfconsciousness, vulnerability and anxiety also relates to the amount of NPCs
talked to and the amount of areas visited. An explanation could be that participants with high neuroticism scores project their social anxieties onto the
characters in the game, causing the participants to avoid talking to NPCs.
In table 24 we can see that participants that play a lot of Action Adventure
67
genre games talk to a lot of NPCs and explore the other conversation options
that are available. This is to be expected as it is a convention that bonuses
or secrets can be found in Action Adventure games through conversation.
Game metric
Hours
played
Difficulty
preference
npcs talked to
Extra options taken
AreasVisited
-0.069
0.303
-0.13
0.009
0.204
-0.016
Many
different
genres
0.197
0.024
0.165
Within
favorite
genre
0.141
0.119
-0.079
Action
Adventure
genre
0.515***
0.686****
0.296
Played
Zelda:ALTTP
-0.105
0.025
-0.225
Table 24: Correlations between the the gamer type metrics and the village
game metrics for conversations. Conversation metrics: NPCs talked to, Extra options taken (conversation options taken minus NPCs talked to). AreasVisited is a combined metric, the sum of all boolean values of visited village
areas. Gamer type metrics: Hours Played, Difficulty Preference, Many different genres, Within Fav genre, Action Adventure, played Zelda:ALTTP. *:
p<0.15, **: p<0.1, ***: p<0.05, ****: p<0.01
Personality dimension/facets
Extraversion
E1:Friendliness
E2:Gregariousness
E3:Assertiveness
E4:Activity Level
E5:Excitement-seeking
E6:Cheerfulness
NPCs talked to
0.342*
0.22
0.372**
0.252
0.082
0.133
0.491***
Extra options taken
0.2
0.201
0.134
0.038
-0.091
0.303*
0.217
AreasVisited
0.33*
0.325*
0.305*
0.385**
0.143
0.116
0.417***
Table 25: Correlations between the the extraversion dimension/facet scores
and the village game metrics for conversations. Conversation metrics: NPCs
talked to, Extra options taken (conversation options taken minus NPCs
talked to). AreasVisited is a combined metric, the sum of all boolean values
of visited village areas. *: p<0.15, **: p<0.1, ***: p<0.05, ****: p<0.01
8.5.5
Village preparation
In the game the player could prepare for the journey by buying specific items
from the shops, the apples which each heal a heart on use and the bottle
which can be found in the bush patch left of the player’s house and which
68
can be filled at the witch’s shop which buys you a diluted potion which also
heals a single heart.
Personality facets/dimensions
(0.023 · N1:Anxiety)
+(0.034 · N2:Anger)
+(0.012 · N3:Depression)
+(0.029 · N6:Vulnerability)
Game metric
(-0.087 · apples)
+(-0.028 · BottleStats)
rho
0.39
p
0.059
Table 26: FastICA components that strongly correlate the village game
metrics about preparation before heading out with the personality scores.
Preperation metrics: apples, BottleStats (0: no bottle, 1: got empty bottle,
2: filled bottle).
In table 26 there is a moderately strong correlation between neuroticism
facets and preparation metrics. An explanation could be that participants
that score lower on neuroticism facets are more relaxed while playing and
think more about planning ahead than if they were to be somewhat more
emotional. But a more likely case is that the personality plays a much smaller
part than the conventions within action adventure games that preparation
is often useful or even vital, this explanation is supported by the correlation
of the component parts in table 27 and that our game metrics pointed out
that 11 out of 25 did not even find the empty bottle and 10 out of 25 did not
buy any apples of which 7 participants did not even visit the plaza where
the apples were sold, which undermines the personality aspect if not every
player has the convention of preparation.
Gamer type metrics
(0.007 · Many different genres)
+(0.036 · Within Fav genre)
+(0.04 · Action Adventure)
Game metrics
(0.109 · apples)
+(0.079 · BottleStats)
rho
0.586
p
0.003
Table 27: FastICA component that strongly correlates the gamer type metrics with the village preparation metrics. Gamer type metrics: Hours Played,
Difficulty Preference, Many different genres, Within Fav genre, Action Adventure, played Zelda:ALTTP. Preparation metrics: apples, BottleStats (0:
no bottle, 1: got empty bottle, 2: filled bottle).
69
Personality dimension/facets
Agreeableness
A1:Trust
A2:Morality
A3:Altruism
A4:Cooperation
A5:Modesty
A6:Sympathy
apples
-0.12
0.061
-0.145
-0.028
-0.007
-0.164
0.013
BottleStats
0.139
0.374**
0.087
0.241
-0.129
-0.031
0.132
Time spent
-0.187
0.121
-0.216
0.065
-0.325*
-0.145
0.069
Table 28: Correlations between the the agreeableness dimensions/facets
scores and the village preparation game metrics. Preparation metrics:
apples, BottleStats (0: no bottle, 1: got empty bottle, 2: filled bottle).
*: p<0.15, **: p<0.1, ***: p<0.05, ****: p<0.01
8.5.6
Fights
In this section we investigate the fight variables that have a form of choice or
strategy involved. Usage of items or lifting an item means keeping yourself
alive by finding resources or by using apples, it can also mean the use of
items and lift that the player has uses bombs, throwable props or other
items to soften up the enemies. The use of the spin attack is useful where
many enemies are afoot, but when improperly timed the enemy will hurt the
player before the attack is fully charged.
Our hypothesis was that a high conscientiousness score would lead to
a more careful approach, but a lower extraversion score might be a better
indicator of using a more ”steady” approach. An argument for the low
extraversion can be made if high extraversion would be related to being
heavily influenced by the environment, as this would reduce any effect that
conscientiousness would have on the planning capability during a fight. It
should be noted that our conscientiousness data is skewed and that we do not
know what kind of impact lower scores could have had on the components.
70
Personality facets/dimensions
(0.038 · E1:Friendliness)
+(-0.001 · E2:Gregariousness)
+(0.032 · E3:Assertiveness)
+(0.033 · E4:Activity Level)
(0.05 · E1:Friendliness)
+(-0.047 · E5:Excitement-seeking)
(0.075 · Neuroticism)
+(-0.05 · Extraversion)
+(-0.105 · Conscientiousness)
(0.038 · E2:Gregariousness)
+(0.066 · E3:Assertiveness)
+(-0.108 · E5:Excitement-seeking)
+(0.124 · E6:Cheerfulness)
(-0.031 · C1:Self-efficacy)
+(-0.015 · C2:Orderliness)
+(-0.033 · C4:Achievement-striving)
+(-0.013 · C5:Self-discipline)
(0.06 · O2:Artistic Interest)
+(-0.005 · O3:Emotionality)
+(0.031 · O5:Intellect)
Game metric
(-0.023 · itemUse)
rho
0.463
p
0.02
(-0.174 · spin)
0.433
0.03
(-0.048 · spin)
-0.466
0.019
(-0.08 · lift)
0.362
0.075
(0.009 · spin)
+(0.014 · lift)
+(0.011 · useItem)
0.338
0.098
(-0.004 · spin)
+(-0.008 · lift)
+(-0.015 · useItem)
0.329
0.108
Table 29: FastICA components that strongly correlate the fight game metrics
with the personality scores. Fight metrics: lift (lifting a rock or bush to
throw), useItem (using an item from the player’s inventory), spin (special
attack with the sword).
Although with a weaker correlation, the participants which played games
more often and have a higher difficulty preference were more likely to use
the tools at their disposal. This is in line with our definition of a hardcore
gamer.
71
Gamer type metrics
(-0.065 · Hours Played)
+(-0.05 · Difficulty Preference)
+(-0.016 · Within Fav genre)
Game metric
(0.109 · spin)
+(0.048 · lift)
+(0.039 · useItem)
rho
-0.345
p
0.091
Table 30: FastICA components that moderately correlate the gamer type
metrics with the fight game metrics. Gamer type metrics: Hours Played,
Difficulty Preference, Many different genres, Within Fav genre, Action Adventure, played Zelda:ALTTP. Fight metrics: lift (lifting a rock or bush to
throw), useItem (using an item from the player’s inventory), spin (special
attack with the sword).
8.5.7
Block pushing puzzle
Looking at table 31 we can see a large effect of the openness to experience
on the performance metrics of the block pushing puzzle. Again we expected
to see conscientiousness to appear here as planning is key in solving these
kinds of puzzles, although the puzzle could also be reset an infinite amount
of times, which undermines the planning aspect. Openness to experience
has been linked to hardcore gamers [5], but the FastICA exploration did
not come up with any strong correlations regarding gamer type metrics
and the block-pushing metrics. A reason for these results could be that
the metrics that were recorded do not have a clear strategy or behavioural
pattern behind them, as such we might not have the right measurements to
find any meaningful relations between these metrics and personality scores
and the internal workings that they represent.
72
Personality facets/dimensions
(-0.036 · Openness to Experience)
+(-0.05 · Agreeableness)
(0.056 · A2:Morality)
+(0.083 · A3:Altruism)
+(0.046 · A4:Cooperation)
+(0.038 · A5:Modesty)
+(0.044 · A6:Sympathy)
(0.003 · N1:Anxiety)
+(0.012 · N3:Depression)
+(0.095 · N4:Self-consciousness)
+(0.079 · N5:Immoderation)
Game metric
(-0.09 · time spent)
+(-0.036 · retries)
+(-0.062 · average vfm time)
(0.044 · time spent)
+(0.059 · retries)
rho
0.401
p
0.052
0.365
0.079
(-0.079 · time spent)
+(0.007 · retries)
+(-0.062 · average vfm time)
0.356
0.088
Table 31: FastICA components that strongly correlate the block-pushing
game metrics with the personality scores.
Block-pushing metrics:
time spent, completed, retries, vfm time.
8.5.8
Moving floor puzzle
In table 32 we see again that a facet that is associated with the hardcore
gamer type is associated with performance. This was to be expected, an
experienced player has better control over the character which means better
performance on the moving floor puzzle.
Cooperation also correlates strongly with the performance metrics, but
this is not one associated with the hardcore gamer metrics. Our less cooperative participants performed better at the moving floor puzzle. This could
just be an artifact or it has to do with an underlying element of cooperation.
Sadly the definition and the test statements have not given any insight.
The correlation with neuroticism, extraversion and conscientiousness is
most likely the same one as in table 15 that correlates the same dimensions
with ”Played Zelda:ALTTP”. Participants that have not played the game
before are at a disadvantage and do not have a feel for the controls yet. For
those who have played it and got far in the game, this type of puzzle should
be familiar.
The correlation between activity level and the performance metrics looks
to be an expression of the gamer type metrics ”Hours played” and the
resulting lack of feel for the controls.
73
Personality facets/dimensions
(-0.012 · C5:Self-discipline)
(0.002 · A4:Cooperation)
(0.076 · O1:Imagination)
+(0.045 · O3:Emotionality)
+(-0.072 · O4:Adventurousness)
+(0.051 · O5:Intellect)
+(-0.099 · O6:Liberalism)
(0.109 · Neuroticism)
+(-0.076 · Extraversion)
+(-0.038 · Conscientiousness)
(0.03 · E4:Activity Level)
Game metric
(-0.11 · time spent)
+(-0.1 · got hurt)
(-0.1 · time spent)
+(-0.104 · got hurt)
(0.018 · time spent)
(0.002 · got hurt)
(0.1 · time spent)
+(0.091 · got hurt)
rho
0.638
p
0.001
-0.515
0.01
-0.448
0.028
-0.407
0.048
0.382
0.065
Table 32: FastICA components that strongly correlate the moving floor
game metrics with the personality scores. Moving floor metrics: got hurt,
time spent.
Game metric
time spent
got hurt
Hours
played
Difficulty
preference
-0.29
-0.417***
-0.293
-0.272
Many
different
genres
0.093
-0.185
Within
favorite
genre
0.129
-0.2
Action
Adventure
genre
-0.414***
-0.248
Played
Zelda:ALTTP
Table 33: Correlations between the moving floor game metrics and the
gamer type metrics. Moving floor metrics: got hurt, time spent. Gamer
type metrics: Hours Played, Difficulty Preference, Many different genres,
Within Fav genre, Action Adventure, played Zelda:ALTTP. *: p<0.15, **:
p<0.1, ***: p<0.05, ****: p<0.01
8.5.9
Maze puzzle
Comparing tables 34 and 35 we can see that gamer type metrics correlate
much stronger with the performance metrics than the personality scores, the
personality facets found in the components are also related to gamer type
metrics.
We can see that the hurt and falls increase with the hardcore gamer
metrics. This can be easily explained as the difficulty in the adaptable
74
0.03
-0.297
setting only goes up when the player finishes within a certain time frame,
if the player struggles somewhat with the controls then time is wasted and
the difficulty doesn’t increase or doesn’t increase as much. We can see that
the more hardcore gamer types perform better in terms of time. As such
the players that got further were mostly the hardcore gamer types and had
to deal with more hazards including the pitfalls, which in turn increases the
amount the character got hurt and the amount of falls.
Personality facets/dimensions
(-0.06 · E4:Activity Level)
(0.002 · C1:Self-efficacy)
+(0.081 · C4:Achievement-striving)
+(0.025 · C5:Self-discipline)
(-0.086 · A1:Trust)
+(-0.051 · A2:Morality)
+(-0.051 · A3:Altruism)
+(-0.054 · A6:Sympathy)
Game metric
(-0.092 · time spent)
+(0.106 · falls)
(0.04 · time spent)
+(-0.101 · got hurt)
+(-0.042 · falls)
(-0.043 · got hurt)
+(0.047 · falls)
rho
0.366
p
0.079
0.324
0.122
0.32
0.128
Table 34: FastICA components that strongly correlate the maze game metrics with the personality scores. Maze metrics: got hurt, falls, time spent.
Game metric
time spent
got hurt
falls
Hours
played
Difficulty
preference
-0.335*
0.611****
0.449***
-0.714****
0.355**
0.342*
Many
different
genres
-0.534****
0.328*
0.405***
Within
favorite
genre
-0.424***
0.382**
0.214
Action
Adventure
genre
0.1
0.427***
0.45***
Played
Zelda:ALTTP
Table 35: Correlations between the maze game metrics and the gamer
type metrics. Maze metrics: got hurt, falls, time spent. Gamer type metrics: Hours Played, Difficulty Preference, Many different genres, Within Fav
genre, Action Adventure, played Zelda:ALTTP. *: p<0.15, **: p<0.1, ***:
p<0.05, ****: p<0.01
8.5.10
In-game exploration
In table 36 we can see an increase in exploration and total time spent on the
game with the increase of the openness facets.
75
-0.344**
0.087
0.358**
Imagination and artistic interest were both found to correlate with the
”Action Adventure genre” and ”played Zelda:ALTTP” gamer metrics. This
could indicate that the experience or interest toward these type of games is
heightened by both of those facets.
We have also investigated the individual correlations of the personality
scores and again no strong correlation was found. Investigating the individual correlations with gamer type metrics did give us results as can be
seen in table 37.
Looking at the individual correlations we can see that playing action
adventure genres strongly correlates with the total time spent on the levels,
the amount of exploration and the resulting rewards retrieved. The many
different genres as well as the ”played Zelda:ALTTP” metrics had a role in
the amount of time spent on the game. This means that the amount of time
the participants spend on the game, and the amount of optional content the
participants want to partake in is mostly influenced by their game history,
learned conventions and perhaps nostalgic feelings towards those games.
Another interesting correlation is the hardcore gamer type metrics with
the average percentage of time spent on the optional path. From our data
we can see that this is likely due to the players that get far, will eventually
lose interest in exploration and deviation from the main path, resulting in a
lower percentage of time spent on the optional path.
Personality dimensions/facets
(-0.032 · O1:Imagination)
+(-0.031 · O2:Artistic Interest)
+(0.013 · O3:Emotionality)
+(0.006 · O5:Intellect)
Exploration metrics
(-0.036 · Total time Spent)
+(-0.036 · Total unique rooms visited)
+(-0.019 · Average percentage of unique rooms visited)
+(-0.041 · Average rewards retrieved percentage)
Table 36: FastICA component that strongly correlates the personality scores
with the exploration metrics. Exploration metrics: Time Spent, Unique
rooms, time per unique room, average reward retrieved percentage, average
time spent optional percentage
76
rho
0.455
p
0.022
Game metric
Hours
played
Difficulty
preference
Within
favorite
genre
0.311*
0.232
0.184
-0.055
Action
Adventure
genre
0.615****
0.514****
0.476***
0.174
Played
Zelda:ALTTP
0.281
0.15
0.045
-0.138
Many
different
genres
0.43***
0.278
0.186
0.058
Total time Spent
Total unique rooms visited
time/unique room
Average percentage of
unique rooms visited
Rewards retrieved
percentage
Average percentage of time
spent on optional path
0.197
0.036
-0.05
-0.127
0.063
0.3*
0.127
0.073
0.526****
0.284
-0.363**
-0.477***
-0.278
-0.307*
-0.158
-0.436***
Table 37: Correlations between the exploration game metrics and the gamer
type metrics. Exploration metrics: Time Spent, Unique rooms, time per
unique room, rewards retrieved percentage, average time spent optional percentage. Gamer type metrics: Hours Played, Difficulty Preference, Many
different genres, Within Fav genre, Action Adventure, played Zelda:ALTTP.
*: p<0.15, **: p<0.1, ***: p<0.05, ****: p<0.01
8.6
Summary of Findings
• Distribution comparisons of hardcore and casual participants show
that we cannot reject that they are drawn from the same underlying
distribution.
• The skewedness in the personality dimension/facets distributions could
be an indication of the kinds of persons that are likely to participate
in lengthy research (1+ hours).
• Conscientious participants were much less into games.
• Low achievers (achievment-striving, conscientiousness facet) among
gamers play more game genres and have developed a favorite genre.
• Neuroticism could be used as an indicator of difficulty preference and
lack of interest in playing different genres.
• A high immoderation (neuroticism facet) score is an indicator of being
a hardcore gamer that binges by playing games for long hours.
• A high intellect (Openness to experience facet) score indicates less
trouble grasping conventions in games.
77
0.492***
0.381**
0.28
0.19
• Having played ”The Legend of Zelda: A link to the past” before and
having a high difficulty preference are the strongest indicators for performance.
• Game element preference correlations:
– Performance metrics do not correlate with game element preference ranking.
– Achievement-striving correlates positively with the preference for
talking to NPCs and the moving floor puzzle, likely is that hardcore gamers like conversations less than challenges that involve
controls.
– Increase in extraversion facets decreases the preference for fights.
– Many strong correlations are present that include agreeableness
or its facets, using the definitions has not given us insight into the
relation to preference. Indication of one or more latent variables.
– Openness to experience facet imagination correlates strongly with
the preference for exploration.
– Extraversion facet assertiveness correlates with the preference for
the dark maze.
– Agreeableness facet trust correlates strongly with preference for
the moving floor puzzle.
– Agreeableness facet modesty correlates strongly with the blockpushing puzzle preference.
– Agreeableness facet sympathy correlates strongly with the dark
maze preference.
– Agreeableness facet anger correlates strongly with the blockpushing puzzle.
– ”Played Zelda:ALTTP” correlates with the block-pushing puzzle
preference, which might be a nostalgia effect.
– The moving floor puzzle preference correlates with hours played.
– ”Many different genres played” and ”Play a lot within favorite
genre” metrics correlate negatively with the preference for talking
to NPCs, likely due to the convention that conversations with
NPCs are often not that important, only the act of talking to
them which can trigger events.
78
• Agreeableness facet cooperation might be an indication of the tendency
to follow instructions and as such leave the village as soon as possible,
resulting in less NPCs talked to and less preparation.
• Participants with high neuroticism scores might project their own anxieties onto the characters in the game, resulting in less NPCs talked to.
• Participants that play a lot of action adventure genre games are likely
more familiar with the convention of bonuses and secrets that are
unlocked through conversation, as such talk more to NPCs.
• Participants which played games more often and have a higher difficulty preference were more likely to use the tools at their disposal.
• Total time spent on the game and the amount of exploration correlates
with the openness facets.
• The ”play action adventure genre a lot” and ”played Zelda:ALTTP”
metrics correlated with the openness facets imagination and artistic
interest. This could indicate that the experience or interest toward
these type of games is heightened by both of those facets.
• Players that get far will eventually lose interest in exploration and
deviation from the main path, resulting in a lower percentage of time
spent on the optional path.
8.7
Discussion of the Results
In this section we address each hypothesis and research question with our
found results.
8.7.1
Discussing research questions
In this section we will discuss the relevant data that we have found per
research question and to present an answer to the research question based
on that data.
RQ 1: Can we link performance to preference?
In the game element preference section in tables 16 and 15 we can see that the
relative performance and reached difficulty does not correlate significantly
79
with the preference ranks, as such we can rule out performance as a link to
preference.
RQ 2: Can we link personality to preference?
In tables 20 and 21 we see many different preference correlations that are
significant, but in some of these cases it is indirectly measuring latent variables that we already know of, one such latent variable, the hardcore gamer
type, can be seen correlating significantly with achievement-striving.
The agreeableness facets are all measured with social statements in the
test and they seem to capture something else than the gamer type or the
social aspect because the agreeableness facets do not correlate with gamer
type metrics, but do correlate with performance metrics (all puzzles) and
village metrics such as the NPCs talked to, the amount of exploration in the
village.
The neuroticism facets correlate with preference, but just like the agreeableness facets they do not correlate with gamer type metrics but they do
correlate with village metrics (NPCs talked to, village exploration and preparation) and performance metrics (block-pushing and moving floor puzzle).
The extraversion facets also correlate with preference but less significantly, it does correlate with village metrics and some performance metrics
but is not useful for determining preference.
The link between personality and preference is present within our data,
but the correlations are most likely caused by latent variables that are expressed through the personality facets.
RQ 3: Can we link difficulty preference and game knowledge to
preference?
In the game element preference section in table 22 we can see that difficulty
preference does not correlate significantly with any of the preferences. As
such we will rule out difficulty preference as a link to game element preference.
Hours played correlated significantly with the moving floor puzzle, and we
expect this to be the case because of the finer control required for the puzzle
and the amount of practice someone has from playing games for so many
hours, considering that the preference did not correlate with performance
means that not having the finer control did not influence the preference in
a bad way.
80
The demographic metrics ”Many different genres played” and ”Play a
lot within my favorite genre” have a significant negative correlation with the
Talking to NPCs preference rank, it is speculated that this is because of the
convention in games that conversation is rarely important and most of it can
be skipped. To further support this, in table 22 the amount of NPCs talked
to did not correlate significantly with either ”Many different genres played”
or ”Play a lot within my favorite genre” metrics as such the preference does
not express itself in actually engaging with the NPCs, but more the overall
liking of the engagement with the NPCs.
The knowledge about Zelda:ALTTP seemed to have influenced the preference towards block-pushing but this result was not significant. We suspect
nostalgic feelings to have an influence.
RQ 4: Should a game generation algorithm also focus on Dynamic
difficulty adjustment?
In the difficulty setting results section we had performed the KolmogorovSmirnov test to determine if the distributions were significantly different
from each other. The test showed that we cannot reject the hypothesis that
the data is drawn from the same distribution and the proportion of hardcore
and casual gamers per group is about the same (4 hardcore and 2 casual in
static, 6 hardcore and 4 casual in adaptable). Therefore we conclude that
our version of adaptable difficulty was not experienced significantly different
from the static linearly increasing difficulty.
The significant distribution difference for Mazes is caused by a the difficulty adjustment settings, which caused the maze to stay too easy for the
adaptable difficulty setting.
Our stance is that Dynamic difficulty adjustment should be in a game if
it is appropriate for the game itself, having a bad parametrized DDA system
is likely no different and possibly worse than creating static scenarios from
which to randomly select.
RQ 5: Can we create a game that enables us to gather data to test
the possibility of these links and the dynamic difficulty adjustment?
We have created a game that gathered performance and behavioural metrics as well as ratings for each level via an in-game questionnaire, the game
contains many elements which were relevant within our limited scope. The
ratings have been used to compare the distributions between static and ad81
aptable difficulty settings. The performance metrics and behavioural metrics
have enabled us to explore most of the chosen user profile metrics.
The game also makes a clear distinction between elements which creates
less noise in our data, the game was received well among our participants.
No participant actually disliked the game, but some had points of critique
towards the balance of content in the game (exploration / conversations /
fights / puzzles) or the results of the content generation algorithms (looked
too computer generated).
RQ 6: If any links can be found is there a viable player profiling
solution among them?
Gamer type can be measured through difficulty reached and performance,
the gamer type can then be used to estimate preference by assuming the
effects of certain gamer type aspects (mastery over controls, conventions
such as dialog skipping and bonuses/secrets via conversation).
The effect of personality on preference is hard to assume since the underlying metric which we are measuring is unknown, therefore it is harder
to apply this to new cases where we have no data. It is more likely that
personality scores are not the correct metrics to use to profile players for
their preference or their predicted performance in certain game elements.
8.7.2
Discussing personality effect hypotheses
Hypothesis: Higher openness to experience results in better performance
Only the intellect facet correlates significantly with difficulty preference,
and difficulty preference correlates significantly with the performance in the
block-pushing puzzle and maze puzzle. But anything requiring more reaction based actions is more related to the hours played and whether they have
dealt with such an obstacle before.
Hypothesis: Low conscientiousness likes more high paced gameplay, high conscientiousness uses more items and careful planning
The achievement striving facet negative correlation with moving floor puzzle
and the positive correlation with talking to NPCs is significant. But both
of these effects can also be attributed to gamer types since this personality
facet also (negatively) correlates significantly with being a hardcore gamer.
As we do not find any other correlations within the conscientiousness dimension it is more likely that it is attributed to the personality trait but rather
the gamer type. The same goes for the item use and abilities which require
82
more careful planning, in fights the hardcore gamer personality facets of conscientiousness negatively correlate with these variables, which again points
more towards an increase in use of items and abilities when the ability as a
player goes up.
Hypothesis: High extraversion means more energetic game-play, more exploration, more interaction with NPCs
In table 23 in the NPC and conversation interaction section we can see that
primarily friendliness and assertiveness facets correlates significantly with
the amount of areas visited in the village. In table 25 we can see that the
cheerfulness facet correlates significantly with the amount of NPCs talked
to and the areas visited in the village. In both the maze and moving floor
puzzle we can see that the activity level facet causes the player to make more
mistakes.
From this follows that certain facets of the extraversion dimension imply more energetic game-play, more exploration and more interaction with
NPCs.
Hypothesis: High agreeableness spends less time in village, follows instructions better, finds more items
From individual correlations in table 28 we can see that the time spent correlates with the cooperation facet, the correlation is not very strong, most
likely because other behaviour will influence the time spent in the village.
The bottle stats correlates with trust, this might be because the participants
with higher trust scores trusted the witch more easily.
The correlations are weak regarding the agreeableness facets, the preparation is correlated much more strongly with the hardcore gamer type
metrics. As such we cannot with certainty say anything about the relation
between agreeableness and the time spent, the instruction following and
finding more items.
Hypothesis: Lower neuroticism scores result in longer playtime and lower
difficulty preference
In the table 15 we can see that the immoderation facet correlates strongly
with difficulty preference, a gamer with a high immoderation score will likely
be someone that plays a lot of games and is more likely to prefer higher
difficulty settings. In the table 36 we see that no significant neuroticism
component parts were found for total time spent. Instead we found that
total time spent correlates strongly with imagination and artistic interest,
which both correlate strongly with the ”Play Action Adventure genre often”
83
metric and that correlates strongly with the with the total time spent.
As such we conclude that the immoderation facet which is linked to the
hardcore gamer type is related to the difficulty preference.
8.7.3
Game literacy and difficulty preference
Hypothesis: A player that performs well will like higher difficulty levels
We can see in table 17 that the difficulty preference correlates significantly
with the reached difficulty level in case of the puzzles. The reason that we
are unable to find the same result for fights is most likely due to our metrics
which only captures the relative difficulty and not the difficulty level as we
do with the puzzles.
With these correlations we can safely assume that within our data the
participants who like higher difficulty levels also reach a higher difficulty (i.e.
perform better).
9
Conclusion
For this master thesis a game was created to mimic the feel of an old classic
game ”the Legend of Zelda: A link to the past”, added to it were elements
that were necessary for this research, the procedural content generation
related to difficulty. The story created from scratch to match the game’s
length and to point the player in the right direction. Generating levels via
algorithms which contain enough comparable elements while keeping the
game relatively interesting. We also implemented a system for adaptable
difficulty to try and find support for DDA systems in games.
The main purpose of this thesis was to explore the possibility of user profiling using different kinds of metrics as possible bridges between behaviour
and preference.
Research question 1: Can we link performance to preference?
During our exploration of the performance link to preference we found that
there were no significant correlations between them as such the measurement
of difficulty level reached does not have a link to its corresponding game-play
element type. This means that we can rule out most performance related
game metrics for our preference categorization.
Research question 2: Can we link personality to preference?
84
We also tried to find a link between personality scores and preference. By
correlating the FFM traits and facets with the reported preference rankings
we found that one of the significant correlations was also correlated with
gamer type metrics. In particular the facet Conscientiousness:Achievementstriving. We also found that the agreeableness dimensions and facets correlated strongly with different types of preference, but given the social nature
of what the agreeableness dimensions tries to measure we suspect one or
more latent variables which are influencing the preference.
Many of the constructs in the personality test were not meant to measure
behaviour in a game setting, so we now have significant correlations between
preference and one or more unknown underlying motivations or behaviours.
This means we cannot link personality scores to preference.
Research question 3: Can we link difficulty preference and game
knowledge to preference?
Difficulty preference, just like the performance metrics does not correlate
significantly with any of the preference. As such we will rule out difficulty
preference as a link to preference.
Game knowledge metrics are of themselves quite broad metrics and as
such it was expected that they would measure latent variables. Combining the metrics with performance and behavioural data we found that the
preference for talking to NPCs decreases as the participant has played more
games. However, given that the participants amount of NPCs engaged with
isn’t correlated with these metrics we suspect it to be a preference towards
the actual engagement. We speculate that it is the participants that have
played a lot of games that can navigate a game successfully without requiring the detailed information that the NPCs have and as such skip most of
the dialog.
We also found a possibly nostalgic effect of the block-pushing game element type, this element occurs frequently within the The Legend of Zelda:
A link to the past. Participants that had played it before had a moderately
significant correlation with the preference for block-pushing.
Research question 4: Should a game generation algorithm also focus
on dynamic difficulty adjustment?
By comparing the distributions of the ratings of each game element we found
that we cannot reject the hypothesis that the data is drawn from the same
85
distribution. Therefore we conclude that our version of adaptable difficulty
was not experienced significantly different from the static linearly increasing
difficulty.
In the case of a game generation algorithm we suspect that there should
be at least a small amount of difficulty correction based on what kind of rule
is added. Even though our stance is that personalisation is key to a practical
form of game mechanics generation. If our data is correct it could be that
difficulty might not be that important as long as it is playable and winnable
and that the expected difficulty follows at least a logical pattern (such as a
linear one).
Adding dynamic difficulty adjustment to game mechanics generation
might make the game difficulty more unstable, it could have a worse result
than adding default mechanics which reduce or manage difficulty inherently.
In regards to procedural generated content a dynamic difficulty adjustment implementation should be in a game if it is appropriate for the game
itself, having a bad parametrized DDA system is likely no different and possibly worse than creating static scenarios from which to randomly select.
Research question 5: Can we create a game that enables us to gather
data to test the possibility of these links and the dynamic difficulty
adjustment?
We have created a game which we have used as a data gathering platform
for our research. The implementation contains many elements which are
relevant for our limited scope. We have successfully used these game metric
in the analysis of our user profile metrics and the user ratings regarding the
game and difficulty were mostly positive. The game can also be expanded
or changed for later research because of the nature of the modifications.
We conclude that we have succeeded in creating such a game.
Research question 6: If any links can be found is there a viable player
profiling solution among them?
While exploring we found that many of the personality facets have underlying
motivations and variables that are not captured by the personality test alone.
Among these underlying variables were the gamer type metrics that told us
the common personality facets among hardcore gamers.
We were not able to link personality and preference as well as hoped. In
our results and analysis sections we talked about possible explanations for
86
the correlations that we found, we were not successful in finding a possible
explanation for all of them, but we did succeed in finding one for most of
them. The personality link that is present should be researched in terms of
underlying motivations and strategies that stem from them.
We were not able to find within our data that performance is linked to
preference of a certain game element. We did find some correlations which
pointed towards gaming conventions and the amount of time a participant
already spends on games. Difficulty preference was not related to any preference within our data.
In the end we have not found a viable profiling solution for all of the
game’s elements, either due to missing the right metrics or not having the
right tools to measure the metrics.
Problem statement: Can we categorize game-play preference with
user profiling?
Given that our data has found some very strong correlations between our
chosen metrics and the preference, even though they likely point to latent
variables we can see that these latent variables at least exist. Our guess is
that they point to gaming conventions (game-play culture) which the player
is familiar with combined with motivations for certain decisions or actions.
This is based on the assumption that anything unfamiliar could be harder to
understand and subsequently appreciate, additionally that the personality
scores did seem to correlate mostly with the behavioural metrics which points
to underlying constructs which were partially measured by the personality
test.
Our conclusion is that we cannot categorize game-play preference with
our chosen user profiling metrics, but that it is likely that metrics exist that
could lead to successful user profiling regarding preference.
10
Game Developer Insights
From our list of findings there are a few that can be useful to the design
process of a game developer. The following pieces of advice can be gathered
from our data:
• When creating a dynamic difficulty adjustment system the following
should be taken into account:
87
– If performance measures are unclear or when too many variables
cannot be properly measured then it is likely that the DDA will
perform sub-par to designed and tested scenarios.
– Players that play longer are also more likely to be hardcore
gamers, which in turn will increase their difficulty preference.
– Using different types of tools available and at the right moment is
also a good indicator of intelligence and adaptability of the player,
which in turn increases their difficulty preference.
– Those that display neurotic behaviour are more likely to have a
lower difficulty preference.
– Those that are familiar with the conventions of your game are
more likely to perform better. This could be indicated by if a
player skips the tutorial sections of your game.
• For when designing the environment, the players that get further into
a game and that eventually get bored by the type of content will either
require a change in game-play or a decrease in optional content as the
game progresses.
• Regarding NPCs and conversations. The players that skip conversations will expect that all the information to complete the game is available to them even when not listening to or reading what the NPCs have
to say. This is something to keep in mind when the mandatory action
is to perform instructions based on what is said in conversations.
• Those that have more imagination are more interested in exploration,
as such if your game speaks to the imagination it is a good idea to
implement some form of exploration.
11
Proposed Future Work
The game was designed to be used in later research as a platform for data
gathering or implementation of other algorithms that require game elements
that are more diverse than for example Infinite Mario.
The implementation still need some work before it is completely usable
as a general purpose platform like Infinite Mario.
• The game engine is written in C++ which allows communications with
many machine learning libraries, an API should be made available to
the Lua script with which the content of the game is written.
88
• Data gathering should be automated, this would likely also be a modification to the game engine to allow the engine to contact a server
through the Lua script.
• Alternatively the C++ source code might be able to be hosted online
which would make the game more accessible to participants.
We have found many faults with the current research’ design, the scope
was too large, there were too many variables to keep track of. The gathering
of participants was slow, the data gathering method slowed that down even
more and the project did not take advantage of the current trends.
If this research would be continued then the game would have to be
simplified as well as give the player more choice in game-play, the more
often chosen or longest played type of game could then be considered more
preferred.
The research could also go more towards game generation, and provide
a way for the player to interact with the generation process, the generation
process would have to have the option to implement certain conventions or
elements, this would limit the scope to game-play and exclude most of the
added things such as NPCs and story based puzzles or preparation. The
preference could then be traced back to the type of elements that have been
selected or interacted with the most. This would make it more appropriate
for researching game-generation aspects as well.
Any future research should also consider other trait measurement tools
as opposed to FFM. Brainhex might be a good starting point from which to
further work out a way of categorizing game-play preference.
A few possible improvements with the use of AI algorithms would be:
• Implementing a more advanced form of enemy AI by using dynamic
scripting [24] or best-response learning [2] to improve the dynamic
difficulty adjustment.
• Implementing a better parametrized generation algorithm for level generation that could make more organic structures within the levels, most
likely work out the more detailed version of the already implemented
version that is described in [9].
• A more detailed version of the dynamic difficulty adjustment could be
implemented for the puzzles which would make use of unsupervised
learning based on the game input of the player and more detailed
behavioural information.
89
• Some form of game mechanic generation could be implemented which
could for example change the available items or the main weapon of
the player.
12
Acknowledgement
Appreciation and acknowledgement goes out to S.C.J Bakkes for guiding
this master thesis in the right direction and staying on as guidance even
when finding work at another university. To Arjen Swellengrebel, who did
most of the work for the combat generation process and helped me find more
motivation for this project while he was still among us and sadly the loss of
motivation when he passed away.
References
[1] Magy Seif-El Nasr Stefanie Tignor C. Randall Colvin Alessandro Canossa, Jeremy B. Badler. In your face(t) impact of personality
and context on gameplay behavior. 2015.
[2] Sander Bakkes, Pieter Spronck, and EO Postma. Best-response learning
of team behaviour in quake iii. 2005.
[3] Sander Bakkes, Chek Tien Tan, Yusuf Pisan, et al. Personalised gaming.
Journal: Creative Technologies, 3, 2012.
[4] Sander Bakkes, Shimon Whiteson, Guangliang Li, George Viorel
Vişniuc, Efstathios Charitos, Norbert Heijne, and Arjen Swellengrebel.
Challenge balancing for personalised game spaces. In Games Media
Entertainment (GEM), 2014 IEEE, pages 1–8. IEEE, 2014.
[5] Chris Bateman, Rebecca Lowenhaupt, and Lennart E Nacke. Player
typology in theory and practice. In Proceedings of DiGRA, 2011.
[6] Cameron Browne and Frederic Maire. Evolutionary game design. Computational Intelligence and AI in Games, IEEE Transactions on, 2(1):1–
16, 2010.
[7] Paul T Costa and Robert R MacCrae. Revised NEO personality inventory (NEO PI-R) and NEO five-factor inventory (NEO FFI): Professional manual. Psychological Assessment Resources, 1992.
90
[8] Ben Cowley, Darryl Charles, Michaela Black, and Ray Hickey. Toward
an understanding of flow in video games. Computers in Entertainment
(CIE), 6(2):20, 2008.
[9] Joris Dormans and Sander Bakkes. Generating missions and spaces
for adaptable play experiences. Computational Intelligence and AI in
Games, IEEE Transactions on, 3(3):216–228, 2011.
[10] Aymeric du Peloux. Sokoban handcrafted puzzles: Mini cosmos. http:
//sneezingtiger.com/sokoban/levels/microcosmosText.html.
[11] Robert J Harvey, William D Murry, and Steven E Markham. A “big
five” scoring system for the myers-briggs type indicator. In annual
conference of the Society for Industrial and Organizational Psychology,
Orlando, 1995.
[12] Erin J Hastings, Ratan K Guha, and Kenneth O Stanley. Automatic
content generation in the galactic arms race video game.
[13] Mark Hendrikx, Sebastiaan Meijer, Joeri Van Der Velden, and Alexandru Iosup. Procedural content generation for games: A survey. ACM
Transactions on Multimedia Computing, Communications, and Applications (TOMM), 9(1):1, 2013.
[14] Robin Hunicke and Vernell Chapman. Ai for dynamic difficulty adjustment in games.
[15] Alexandru Iosup. Poggi: generating puzzle instances for online games
on grid infrastructures. Concurrency and Computation: Practice and
Experience, 23(2):158–171, 2011.
[16] John A Johnson. Measuring thirty facets of the five factor model with
a 120-item public domain inventory: development of the ipip-neo-120.
Journal of Research in Personality, 51:78–89, 2014.
[17] Jesper Juul. A casual revolution. Reinventing Video Games and Their
Players, 2010.
[18] Chong-U Lim and D Fox Harrell. An approach to general videogame
evaluation and automatic generation using a description language. In
Computational Intelligence and Games (CIG), 2014 IEEE Conference
on, pages 1–8. IEEE, 2014.
91
[19] Sus Lundgren, Karl J Bergström, and Staffan Björk. Exploring aesthetic
ideals of gameplay. In Breaking New Ground: Innovation in Games,
Play, Practice and Theory: Proceedings of the 2009 Digital Games Research Association Conference, London. Brunel University.(Cited on
page 23.), 2009.
[20] Nicole McMahon, Peta Wyeth, and Daniel Johnson. Personality and
player types in fallout new vegas. In Proceedings of the 4th International
Conference on Fun and Games, pages 113–116. ACM, 2012.
[21] Yoshio Murase, Hitoshi Matsubara, and Yuzuru Hiraga. Automatic
making of sokoban problems. In Pacific Rim International Conference
on Artificial Intelligence, pages 592–600. Springer, 1996.
[22] Lennart E Nacke, Chris Bateman, and Regan L Mandryk. Brainhex:
Preliminary results from a neurobiological gamer typology survey. In
ICEC, pages 288–293. Springer, 2011.
[23] Nicole Peever, Daniel Johnson, and John Gardner. Personality & video
game genre preferences. In Proceedings of the 8th australasian conference on interactive entertainment: Playing the system, page 20. ACM,
2012.
[24] Pieter Spronck, Marc Ponsen, Ida Sprinkhuizen-Kuyper, and Eric
Postma. Adaptive game ai with dynamic scripting. Machine Learning, 63(3):217–248, 2006.
[25] Christophe Thiery. Solarus engine, an arpg game engine. http://www.
solarus-games.org/. Accessed: 2014-10-30.
[26] Christophe Thiery. Legend of zelda: Mystery of solarus dx, 2011.
solarus-games.org.
[27] Julian Togelius, Mark J Nelson, and Antonios Liapis. Characteristics
of generatable games. intelligence, 9:20, 2014.
[28] Giel van Lankveld, Pieter Spronck, Jaap van den Herik, and Arnoud
Arntz. Games as personality profiling tools. In Computational Intelligence and Games (CIG), 2011 IEEE Conference on, pages 197–202.
IEEE, 2011.
[29] Veronica Lorena Zammitto. Gamers’ personality and their gaming preferences. PhD thesis, Communication, Art & Technology: School of
Interactive Arts and Technology, 2010.
92
Appendix A
Demographic questionnaire
Question
What is your name or handle?
Please fill in your age.
What is your gender (biological
sex)?
How many hours a week do you play
video games?
What type of of system do play
video games on?
Answers or answer type
Text
Number
Male/Female/Other
0-4 / 4-10 / 10-20 / 20+
Desktop Computer / Laptop, Console, Handheld console, Phone, Tablet, Other[Text]
5-level Likert scale
I have played many different genres
of video games.
I play video games of the genre Action Adventure very often.
What kind of difficulty setting are
you drawn to?
Have you played ”The Legend of
Zelda: A link to the Past” before?
5-level Likert scale
Very easy / Easy / Normal / Hard
/ Very Hard / Impossible
No / Yes
Table 38: This table contains the questions and possible answers posed in
the demographic questionnaire.
93
Appendix B
Personality test
The following instruction was given at the beginning of the personality questionnaire:
How Accurately Can You Describe Yourself?
Describe yourself as you generally are now, not as you wish to be in the
future. Describe yourself as you honestly see yourself, in relation to other
people you know of the same sex as you are, and roughly your same age. So
that you can describe yourself in an honest manner, your responses will be
kept in absolute confidence. Indicate for each statement whether it is
1. Very Inaccurate, 2. Moderately Inaccurate, 3. Neither Accurate Nor
Inaccurate, 4. Moderately Accurate, or 5. Very Accurate as a description of
you.
Estimated time required to fill in this questionnaire is 20 to 40 minutes.
All the questions used in the personality questionnaire can be found at
http://ipip.ori.org/30FacetNEO-PI-RItems.htm.
The questionnaire was constructed by interleaving all questions such that
no question is adjacent to a question of its own category.
94
Appendix C
C.1
Game metrics
Village game metrics
Name
name
timeStamp
npcsTalkedTo
optionsTaken
totalOptions
percOptionsTaken
percNpcsTalkedTo
cureBrewer
cureWitch
apples
rupees
foundBottle
filledBottle
visitedBushArea
visitedWoodsExit
visitedPlaza
visitedBrewerArea
timeSpentVillage
villageFromSave
Village game metrics
Type
Description
string
character name
datetime
date and time of game start
integer
number of NPCs talked to
integer
number of conversation options explored
integer
total number of conversation options available
fraction
percentage of conversation options
explored
fraction
percentage of NPCs talked to
boolean
got the brewer’s cure
boolean
got the witch’s cure
integer
amount of apples gathered before
leaving the village
integer
amount of rupees gathered before
leaving the village
boolean
found the hidden bottle in the
bushes
boolean
filled the hidden bottle with the
witch’s red potion
boolean
visited the area around the bushes
boolean
visited the northeast area behind
the buildings
boolean
visited the plaza with the old lady
and the apple merchant
boolean
visited the area around the brewer
decimal
amount of seconds spent in the village
boolean
player had saved in the village and
continued later
Table 39: This table contains the full list of logged game metrics within the
village.
95
C.2
Fights game metrics
Fight game metrics: initial inventory, enemies and room layout data
Name
Type
Description
name
string
character name
mapID
string
ID of the current level
egg
integer
amount of manillosaur enemies
spawned
snapDragon
integer
amount of snapdragon enemies
spawned
hardhat
integer
amount of hardhat enemies spawned
knight
integer
amount of knight enemies spawned
boss
integer
has boss been spawned
startLife
integer
amount of life the player had at the
start of the fight
hasBow
boolean
does the player have a bow
hasMirror
boolean
does the player have the magic mirror
hasFairyBottle
boolean
does the player have a bottle filled
with a fairy
hasGlove
boolean
does the player have the power
gloves that enables white rock lifting
hasGlove2
boolean
does the player have the upgraded
power gloves that enables black rock
lifting
hasBomb
boolean
does the player have the bomb bag
pitfalls
integer
the length where pitfalls / water
meets walkable ground
spikes
integer
the length where spike floor meets
walkable ground
inside
boolean
is the current map inside the caves
surface
integer
amount of 8×8 pixel squares of
walkable space available
Table 40: This table contains the full list of logged game metrics of the
fights.
96
Name
finished
swordHits
bombUsage
bowUsage
mirrorUsage
applesUsage
explodeHits
thrownHits
fightTime
dirChange
lifeLost
clangs
uselessKeys
moving
standing
percStanding
avgAggro
killEgg
killSnapDragon
killHardhat
killKnight
Fight game metrics: end of fight data
Type
Description
boolean
finished off all remaining enemies
without leaving the area
integer
times the enemies were hit with the
sword
integer
times the bomb bag was used in the
fight
integer
times the bow was used in the fight
integer
times the magic mirror was used in
the fight
integer
times an apple was used in the fight
integer
times the enemies were hit by a
bomb explosion
integer
times the enemies were hit by a
trown bush or rock
decimal
seconds spent fighting
integer
times the player has changed the
direction of the character
integer
life the player lost during the fight
integer
times the sword of a knight was hit
during the fight
integer
input keys that have been pressed
that do nothing
decimal
seconds spent moving during the
fight
decimal
seconds spent standing still during
the fight
fraction
percentage of time standing still vs.
moving around
decimal
average enemies chasing the player
over time
integer
manillosaurs killed
integer
snapdragons killed
integer
hardhats killed
integer
knights killed
Table 41: This table contains the full list of logged game metrics of the
fights.
97
Name
free
freezed
grabbing
hurt
stairs
loading
spin
swing
tap
carry
lift
treasure
useItem
falling
backOnFeet
Fight game metrics: hero states recorded
Type
Description
integer
character returned to idle state
integer
character’s control has been frozen
integer
character is grabbing the environment
integer
character went into the hurt animation
integer
character went into the stairs animation
integer
character is loading a sword spin attack
integer
character is performing a sword spin
attack
integer
character swings his sword
integer
character taps the environment with
his sword while loading a sword spin
integer
character is carrying an object
above his head
integer
character lifts an object above his
head
integer
character opens a treasure chest
integer
character uses an item from his inventory
integer
character is falling in a pit or body
of water
integer
character is placed back on a walkable surface after falling
Table 42: This table contains the full list of logged game metrics of the
fights.
98
C.3
Puzzle game metrics
Name
name
mapID
puzzleType
difficulty
timeSpent
retries
gotHurt
falls
deaths
quit
completed
averageVFMtime
Individual puzzle game metrics
Type
Description
string
character name
string
ID of the current level
string
type of puzzle (maze / moving-floor
/ sokoban)
integer
difficulty rating (1-5)
decimal
seconds spent on the puzzle
integer
times the sokoban puzzle was restarted
integer
times the player got hurt during
maze or moving-floor puzzle
integer
times the player fell in a pit during
the maze puzzle
integer
times the player died during the
puzzle
boolean
player has pressed the optional quit
button in a sokoban puzzle that becomes available after 135 seconds
boolean
puzzle is completed by reaching the
farthest side
decimal
average seconds a player took before
moving his first block after restarting the sokoban puzzle
Table 43: This table contains the full list of logged game metrics of individual
puzzles.
99
Puzzles per segment game metrics
Name
Type
Description
name
string
character name
mapID
string
ID of the current level
totalTime
decimal
time spent puzzling
sokobanTotalTime
decimal
time spent on sokoban puzzles
sokobanRetries
integer
times the player restarted a sokoban
puzzle
sokobanQuits
integer
times the player quit a sokoban
puzzle
sokobanPuzzles
integer
sokoban puzzles encountered
sokobanVFM
decimal
average of averages of the the
seconds a player took before moving his first block after restarting the
sokoban puzzle
movingFloorTotalTime decimal
time spent on moving-floor puzzles
movingFloorGotHurt
integer
times player got hurt during
moving-floor puzzles
movingFloorDeaths
integer
times player died during movingfloor puzzles
movingFloorPuzzles
integer
moving-floor puzzles encountered
mazeTotalTime
decimal
total time spent on maze puzzles
mazeGotHurt
integer
times player got hurt during maze
puzzles
mazeFalls
integer
times player fell into a pit during
maze puzzles
mazeDeaths
integer
times player died during a maze
puzzle
mazePuzzles
integer
maze puzzles encountered
Table 44: This table contains the full list of logged game metrics of puzzles
per level segment.
100
C.4
Exploration game metrics
Exploration game metrics: initial settings, rewards and rooms
Name
Type
Description
name
string
character name
mapID
string
ID of the current level
staticDifficulty
boolean
static difficulty enabled
branchLength
integer
length of the optional path branches
fights
integer
number of fight-type rooms
puzzles
integer
number of puzzle-type rooms
outside
boolean
is outside or forest themed
missionType
string
tutorial / normal / boss type missions which determines layout
fightsPuzzlesRatio
fraction
the ratio of fights to puzzles
totalRooms
integer
total amount of rooms in the level
roomsVisited
integer
amount of rooms entered
uniqueRoomsVisited
integer
amount of unique rooms entered
percUniqueRoomsVisited fraction
percentage of rooms visited at least
once
rewardsAvailable
integer
optional treasure available
rewardsRetrieved
integer
optional treasure retrieved
heartAvailable
boolean
one optional treasure contains heart
container
heartRetrieved
boolean
heart container retrieved
percRewardsRetrieved
fraction
percentage of rewards retrieved
fightsEncountered
integer
fights encountered by the player
fightsFinished
integer
amount of fights completely cleared
of enemies by the player
puzzlesEncountered
integer
puzzles encountered by the player
puzzlesFinished
integer
puzzles finished by the player
typeEncounterRatio
fraction
encounter ratio of fights vs. puzzles
Table 45: This table contains the full list of logged room related game metrics
of the exploration for a single level.
101
Exploration game metrics: time
Name
Type
Description
timeSpent
decimal
time spent in the level
timeSpentOptional
decimal
time spent in the optional areas
timeSpentMain
decimal
time spent in the main route areas
timeSpentPuzzling
decimal
time spent on puzzles
timeSpentFighting
decimal
time spent fighting
timeSpentOther
decimal
remaining time spent elsewhere
fightsPuzzleTimeRatio
fraction
time ratio of fights vs. puzzles
percTimeSpentOptional fraction
percentage of time spent in optional
areas
percTimeSpentMain
fraction
percentage of time spent in the main
route areas
Table 46: This table contains the full list of logged time related game metrics
of the exploration.
C.5
In-game questionnaire
Name
exploration
puzzles
puzzlePreference
fights
overallExperience
Level segment questionnaire
Type
Description
likertscale
The player agrees that he liked the
(1-5)
amount of exploration available
likertscale
The player agrees that he liked the
(1-5)
challenge that the puzzles brought
string
Puzzle the player preferred: maze,
moving-floor
or
block-pushing
(sokoban)
likertscale
The player agrees that he liked the
(1-5)
challenge that the fights brought
likertscale
whether the player agrees that he
(1-5)
liked the overall feel and theme
Table 47: This table contains the logged game metrics of the in-game questionnaire about perceived challenge and the enjoyment of the level segments
in between shops.
102
Appendix D
Post-Game questionnaire
Question
Once again please fill in your name
or handle.
I never got lost during the game.
I liked to explore during the game.
I went exploring because...
I stopped exploring because...
I spent about this much time playing the game (estimate)
I spent more time trying to explore
than to find my way to the next
level.
I spent more time fighting than
puzzling.
I encountered more puzzles than
fights.
I tried to prepare as much as possible before heading out of the village.
I talked to as many characters as
possible.
I thoroughly explored all the conversation options.
I explored every nook and cranny of
the village.
I found the witch to be more trustworthy than the brewer.
Answers or answer type
Text
5-level Likert scale
5-level Likert scale
...of the rupees and items, ...of possible secrets, ...I want to explore,
Other[Text]
...of the difficulty, ...I got bored with
the game, ...I wanted to finish the
game fast, ...I explored everything
there was to explore, Other[Text]
10-25 min. / 26-40 min. / 41-55
min. / 56-70 min. / 70+ min.
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
Table 48: This table contains the first part of the questions and possible
answers posed in the post-game questionnaire.
103
Question
I found the puzzles to be too easy.
I found the puzzle difficulty to be
fitting for this type of game.
I think that I finished the blockpushing puzzles quickly.
I found myself having to reset the
block-pushing puzzle a lot.
I found the block-pushing puzzles to
be too hard.
I found myself thinking a long time
before making my first move with
the block-pushing puzzles.
I think that I finished the movingfloor puzzles quickly.
I often got my character hurt during
a moving-floor puzzle.
I found the moving-floor puzzles to
be too hard.
I think that I finished the maze
puzzles quickly.
I often got my character hurt during
a maze puzzle.
I found the maze puzzles to be too
hard.
Answers or answer type
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
Table 49: This table contains the second part of the questions and possible
answers posed in the post-game questionnaire.
104
Question
I often used my environment to deal
with enemies.
I often used items (bombs, bow,
mirror, apples, etc.) that I had at
my disposal.
During a fight I would move around
quite a lot.
During a fight I would swing my
sword as many times as possible.
During a fight I would time my
sword swings as best I could.
I often got my character hurt during
a fight.
I found the fights to be too hard.
I found the combat difficulty to be
fitting for this type of game.
I think that I finished the fights
quickly.
I found the boss fight to be too hard.
My favorite strategy during fights
regarding enemies is...
I found that the game was fun to
play.
I found that the story presented in
the game was interesting.
Rank all the elements in order of
that which you liked most.
[Optional feedback box]
Answers or answer type
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
5-level Likert scale
...luring them close / ...rushing towards them / ...avoiding
them / ...softening them up with
items/rocks/bushes
5-level Likert scale
5-level Likert scale
(most enjoyable)1-6(least enjoyable) ranking ofthe categories
Fights,
Block-pushing
puzzle,
Dark maze, Moving floor puzzle,
Talking to characters, Walking
around/exploring
Text
Table 50: This table contains the third part of the questions and possible
answers posed in the post-game questionnaire.
105
Appendix E
Website
Figure 23: The entry page of the website, it displays information about the
research and what a participant gets by participating in the research.
106
Figure 24: A summary of what is required for participation and information
about the engine.
107
Figure 25: The instructions given to the participant to complete the experiment.
108
Figure 26: The standard control scheme and contact information.
109