Building Players: Revisiting Bartle’s Taxonomy of Players
SUBMITTED IN PARTIAL FULFILMENT FOR THE DEGREE OF MASTER OF SCIENCE
Thomas van Dam
10002918
MASTER
INFORMATION
STUDIES
GAME STUDIES
FACULTY OF SCIENCE
UNIVERSITY OF AMSTERDAM
August 24, 2015
!
1
stSupervisor
2
ndSupervisor
Dr. ing. S.C.J. (Sander) Bakkes
Dr. Frank Nack
Building Players: Revisiting Bartle’s Taxonomy of
Players
Thomas van Dam
University of Amsterdam
Graduate School of Informatics
Science Park 904, Amsterdam
thomasvdam@me.com
ABSTRACT
In order to better classify single-player games by the behaviours they facilitate, we propose a player model based on Bartle’s Taxonomy of Players. Through use of a questionnaire we evaluated the performance of this model compared to Bartle’s model, but were unable to find strong evidence that made the new model stand out. We also argued that creative player behaviour should be considered as a unique and separate kind of behaviour. This notion was supported by the data, as the creative behaviour served as a significant identifier for sandbox style games.
Keywords
Player modelling, Bartle’s Taxonomy of Players, sandbox, games
1. INTRODUCTION
With the ever increasing pace at which new games are released it becomes more and more difficult to find games that we like to play. While the internet can help in finding new games that we might like, either through professional reviews or comments from other individuals, the information provided is mostly a subjective view of how the game was experienced. While genre classifications offer some guidance, there can still be a wide variety of games within a genre that might or might not appeal to one’s personal taste. By looking at the kinds of behaviour that games facilitate and players can enjoy, it is possible to create a model that provides an objective base for reviewers and consumers to evaluate the games they play or maybe want to play. This is known as player modelling, and its uses go beyond simply finding fun new games to play, both for academic purposes as business purposes. The academic value lies in its usefulness as a common ground for researchers to base their ideas and theories on. This ensures that the parties involved in the academic discussion are on the same page regarding the observations made on player behaviour. Additionally, this means that researchers can focus more on their subject matter, rather than having to explicitly explain how they are interpreting displayed behaviour. For a business player modelling can be useful since they could use the model of a certain player to recommend new games to the player that facilitate behaviour in line with the gathered model.
A common player model that has been referenced numerous times and keeps generating interest is the Bartle Taxonomy of Players [19]. Perhaps the reason for this is that it was one of the earliest, or that it is one of the more simple models, but nevertheless it remains one of the most well known models out there [19]. Yet there has been a fair amount of criticism regarding Bartle’s model, including by Bartle himself, who claims that his taxonomy might be incomplete for other types of games besides MUDs [7]. Games have evolved quite
substantially since 1996, logically there are new ways to behave in games which the Bartle Taxonomy of Players is not tuned to. However, this does not mean that Bartle's model is without use. It has been shown that numerous different player models are closely related to Bartle’s model and that the approach he takes with this model has its merits [19]. As we will discuss in more detail in the background chapter these models come with their own shortcomings. Most notably a lack for expressing creative behaviour in games, which we believe is important enough to consider separately. So for this study we take Bartle’s Taxonomy of Players and modify it so that it caters to a wider variety of games, including games that focus on creative behaviour. In addition, we will look whether creative behaviour in games is worth giving its own category. So briefly put the main research questions for this study are:
•
Can a new model differentiate between single-player games better than Bartle’s model?•
Is it worthwhile to consider creative behaviour in games separately?2. BACKGROUND
In order to better understand what we set out to do it is important to consider the context of this research. We will first look at player modelling in a broader sense, before diving into the specifics of Bartle’s Taxonomy of Players. We will then take a look at other existing player models and their relation to Bartle’s, before finally considering the position of creative gameplay in player modelling.
2.1. Player Modelling
Player modelling is a research area that focuses on analysing how players go about in playing the games that they play, and then using this information for various ends [4,15,20]. In addition, there are multiple ways to go about player modelling [4]. In this thesis we will exclusively deal with constructing a model based on the behaviour of the player displayed within a game. However, instead of working from the game’s point of view to gain insight into the player, we use the model to classify games based on the kinds of behaviour that they facilitate. So rather than gaining insight into the individual player as to modify the game for optimal enjoyment, we focus on a more general view of the game, looking at how multiple players play the game to gain insight into the game. This dual purpose of player models makes them very versatile, which is why it is used in game development [15,20] as well as game analysis [6,19].
2.2. Bartle’s Taxonomy of Players
2.2.1 The model and player types
Barte’s Taxonomy of Players grew out of a long discussion about the reasons why players play MUDs. While summarising the contents of this discussion Bartle saw a pattern emerging; most reasons for playing could be grouped up in four distinct
categories [6]. This formed the base for his taxonomy, which can be observed in Figure 1.
Bartle constructed two axes to map his four categories to, based on the sources of interest that each category has in the game. On the x-axis there is a focus on players on the left, versus a focus on the game world on the right. The y-axis goes from a focus on acting at the top, to a focus on interacting on the bottom. The player types are situated in the quadrants associated with their interests. A closer look at each of the player types follows
1) Achievers
Achievers focus on acting on the game world, which boils down to doing things in the game. They care little about the other players in the game, or about the intricacies of the game if it does not result in them gaining more points.
2) Explorers
Explorers are interesting in interacting with the game world, always looking for new things in the game. They thrive on being surprised by the game, but not so much by other players.
3) Socialisers
As the name suggests, socialisers focus on interacting with other players. They want to get to know new players and engage in social activity with them. For them, the game world is mostly a backdrop to their social engagements.
4) Killers
Killers are looking to impose themselves on others, acting on players rather than the game world. They thrive on demonstrating how superior they are to other individuals, merely beating computerised opponents is not enough for them.
2.2.2 Strengths of Bartle’s Taxonomy
Perhaps one of the biggest strengths of Bartle’s model is its relative simplicity. With just four player types, divided over two distinct axes it is easy to comprehend and intuitive to use. Additionally, the use of a scale allows for player models to have varying degrees of interest in the aspects of the game. A player is usually not limited to one style of play, and can dabble in other styles from time to time. Bartle can account for this by assigning values to each of the axes for a player, creating a multi-dimensional model rather than just a single player type.
The fact that classifications similar to that of Bartle are widespread also adds merit to quality of this type of classification. As Stewart notes, a great deal of player models are very similar to Bartle, and thus to one another [19]. Further on in this thesis we will take a closer look at these other models. In addition to scientific player models, there are also industry examples of companies that use a classification which shares similarities with Bartle’s model. Most notably is the model employed by Wizards of the Coast in their design of
new cards for Magic: The Gathering [16]. They use a cast of three player types: Timmy, Johnny, and Spike, which roughly correspond to Bartle’s Socialisers, Explorers, and Achievers. In addition, they also allow for players to associate with multiple playing styles in varying degrees of intensity. A possible reason for not having a Killer equivalent in the model Wizards of the Coast employ might be that the multiplayer aspect of the game is in most cases mutual. Players agree to play a game with each other, whereas in MUDs the players are placed in a game with random other players.
2.2.3 Shortcomings of Bartle’s Taxonomy
In a world where new games and game genres pop up on a regular basis, the fact that Bartle’s model is twenty years old does not do it any favours. Since then games have moved at a breakneck speed, introducing various new ways to enjoy games along the way. Bartle’s Taxonomy is simply not equipped to classify behaviour in most modern games, although this does not mean the model itself is flawed. It is just that when an industry develops so rapidly a twenty year old model can hardly be expected to have the same relevance. Which brings us to another shortcoming of Bartle’s Taxonomy of Players; that it was initially designed for MUDs only, a quite specific kind of game which was popular at that time. This has made it difficult to use the model in different games, even Massive Multiplayer Online Role-Playing Games, which share many similarities with MUDs [7]. This greatly reduces the effectiveness of the model, especially when considering the fact that MUDs (and MMO’s in general) are steadily declining in popularity [8].
Pigeonholing Bartle’s model even further is the fact that it was developed based on an online multiplayer game. This means that all games which focus more on delivering a single player experience are hard to classify using Bartle’s model.
2.3. Other Player Models
Up to now we have almost exclusively dealt with Bartle’s Taxonomy of Players, but there are numerous other models out there that aim to categorise players by their playing style. A particularly interesting model is the Four Keirsy Temperaments [13], which uses a categorisation very similar to Bartle’s. These were not derived from people playing games, but rather a pattern Keirsey observed from the sixteen types of the Myers-Briggs personality model. These four categories are high level constructs of personality traits, which can be seen as a superset of Bartle’s player types [19]. Even though Keirsey’s Temperaments are not specifically tailored to games, they do allow for categorisation based on the type of behaviour a person exhibits in the world, or in a game world [19].
Another four type model is the model constructed by Bateman, the Demographic Game Design model (DGD1) [9]. Through observation of video games Bateman came to four player types that are all slightly different from the four Bartle types. However, as Stewart notes, it is possible to construe the types of the DGD1 model as hybrids of the Bartle types [19]. By elaborating on the Hardcore and Casual modes described by Bateman, Stewart created six types that function as all possible hybrid combinations of the Bartle types.
This brings us to the Unified Model, which is the brainchild of Stewart [19]. In this model he incorporates many different player models, as we already touched upon in the previous paragraphs. He shows that a number of the most well-known player models as well as game design models share so many conceptual elements that it is possible to combine them all in a single model [19].
2.4. Sandbox Games
In all the different aforementioned models we observed that most did not explicitly deal with the creative aspect that some players enjoy in video games. The popularity of sandbox games such as Minecraft shows that there is a desire for games
with no explicit purpose other than to build or create whatever the player desires. Most models regarded building as a component of simulation, where the player wants to copy something from the real world. While the unified model does consider creative building, it is shoehorned into Bartle’s explorer category [19]. Research has shown that sandbox players are motivated by a unique set of motivators that are not reflected in any existing player model [10,21]. We believe that the sandbox aspect of games should be seen as its own category.
3. A.C.E2, A NEW MODEL
To address the shortcomings in Bartle’s model we took his model and made some adjustments to it in order to make it applicable for a wider variety of games rather than just the MUDs from Bartle’s model. We will first look at how the axes changed in the model compared to Bartle’s model, and then proceed to look at the new player types in more detail.
3.1. Model Overview
Figure 2 shows the axes and player types in the ACE2 model. Straight away we can see that it is very reminiscent of Bartle’s Taxonomy of Players. It also uses two axes and four player types. Below we discuss the motivation for both of the axes.
3.1.1 Horizontal axis
As observed earlier, part of the weakness of Bartle’s model lies in the fact that it is geared towards a very specific kind of game: MUDs. Since we wanted to create a model that was applicable to a wider variety of games we took a more abstract approach to games. However, we quickly discovered that the multiplayer aspect of games adds so many intricacies to the kinds of behaviour that players display that we decided to limit this model to single-player games. While this seems detrimental, it allowed for a greater degree of nuance than what would have been possible had we included all kinds of games. Since Bartle’s x-axis dealt with the distinction between the virtual world and its player inhabitants, we were no longer able to use this axis. Instead we came up an axis that deals with different ways of enjoying games. There are numerous reasons why players enjoy playing games [2,14], but that these can be divided into two main categories which we labeled Aesthetics and Mechanics. Whilst the term Aesthetics might be somewhat confusing due to its use in the MDA model [12], we felt that it best conveyed what we meant by it. Namely, the elements of the game that do not belong to the gameplay. This includes the narrative of a game, its visual style (or lack thereof) [18], the soundtrack, but also the emotional responses that can be triggered by the game, which is why we decided to keep the reference to the MDA model. On the other side of the scale we have the Mechanics, which are the elements of the game that comprise the gameplay of a game.
3.1.2 Vertical axis
The vertical axis is exactly the same as it is in Bartle’s model, since we felt that the distinction Bartle [6] makes between acting on the game world and interacting with the game world is also present in single player games.
3.2. ACE2 Types
We will now describe all four player types, which the model derives its name from (Achievers, Creators, Explorers, Engagers), in detail and highlight the kinds of gameplay that they enjoy.
1) Achievers
The achievers in this model are very similar to the achievers in Bartle’s model, since they focus on acting on the game mechanics, which is almost the same as Bartle’s acting on the game world. Achievers enjoy winning and gaining points, but also enjoy obtaining mastery over the mechanics of the game.
2) Explorers
While the explorers in this model share some traits with the explorers in Bartle’s model, they are more distinct than the achievers. They also seek to learn about the game’s intricacies and quirks, but more focused on the gameplay side. Exploring terrain is not as interesting to them as it is to Bartle’s explorers. They will often look for interesting interactions in games, such as unique combo’s in deck building games such as Hearthstone or novel use of game mechanics. An example of the latter is ‘snaking’ in Mario Kart DS, a technique that uses the drifting mechanic, which was intended for taking corners, to increase the speed of the vehicle on straight sections of the track as well.
3) Engagers
Engagers are the first completely new type, and focus on interacting with the aesthetics of the game. They are more interested in the story or views a game provides, and not so much the gameplay. They will often look for games that trigger an emotional response, or that allows them to form an emotional bond with the characters in the game. Interactive novels are an example of games that lean heavily towards this category, as they oftentimes provide minimal gameplay but instead deliver a rich aesthetic experience.
4) Creators
Creators are the final player type in this model, and are also the type that sets it apart from most other models. While this kind of behaviour is often a minor part of a different category, or even completely disregarded, here it has its own player type. Creators, like engagers, are drawn towards the aesthetics of a game, but seek to act on them rather than interact with them. This manifests as creating structures or visuals within the game, effectively using the game as a creative outlet. Creators can also use the game to create their own aesthetic experience as to trigger an emotional response in others who experience their work. This includes creating levels that may elicit certain feelings from the player.
4. METHOD
In order to examine how ACE2 held up compared to Bartle’s taxonomy of players, we constructed a questionnaire in which participants were asked to rate how strong the focus on a particular kind of behaviour was in selected games. By looking at how the focuses are divided for both models we were able to compare the performance for the models on the selected games. We shall first discuss what games were selected for this questionnaire, and subsequently discuss how the items of the questionnaire were constructed. Finally, we shall discuss briefly how the questionnaire was presented to the participants.
4.1. Selected Games
To ensure a wide variety of games we created a list of at least reasonably well-known games from many different genres.
Game genres are still a quite active topic of debate in the scientific community, despite the fact that the notion has been around for many years now [1]. There have been many categorisations by many researchers [5,11,17], but we believe that the genres put forward by Bakkes and the inclusion of one more encompass most games [3]. Below we will briefly discuss the genres as listed by Bakkes and finally the games selected for each genre.
4.1.1 Game genres
Action: Action games mostly challenge the reaction speed of
their players. It is one of the most basic genres, with gameplay often emphasising combat.
Adventure: Adventure games focus on the narrative provided
by the game, and often require the player to explore the game world and interact with the game characters.
Role-playing: Role-playing games (RPGs) usually also focus
on a narrative in the game, but they differentiate themselves by letting the player take the role of a character that evolves over the course of the game.
Simulation: Simulation games aim to mimic the real world, or
sometimes a fictional world. They value relative realism highly.
Strategy: Strategy games challenge the decision making
capabilities of the player. Where action games test the physical capabilities of the player, strategy games test the mental capabilities. Most strategy games seek to keep the role of chance to a minimum.
Sandbox: This is the genre we added to the list put forward by
Bakkes [3]. As we observed in our evaluation of player models, the sandbox game is a unique type that comes with its own motivators and play styles. Without this genre certain games become difficult to classify. While it is true that many sandbox games add elements from other genres, mostly to appeal to players that pertain to different player categories, the differences are substantial enough to warrant a separate genre. The key feature of a sandbox game is that it does not provide the player with a goal to accomplish, which is almost always the case in other genres. A sandbox game is about the goals the players set for themselves, which is why a pure sandbox game attracts a specific kind of player [21].
4.1.2 Selected games
For each genre we selected three games in an attempt to cover as many of the sub-genres as possible. Some of the selected games were part of a series in which multiple games were nearly identical in terms of the gameplay they provided. In such cases all these games were grouped under the series.
4.1.2.1 Action games
Possibly one of the most iconic characters in the game industry, Mario has spawned many different kinds of games. A popular series of action platformer games is the Super Mario
Bros. series. Since the core gameplay is very similar between
all instalments, all are considered valid: Super Mario Bros.,
Super Mario Bros. 2, Super Mario Bros. 3, New Super Mario Bros, New Super Mario Bros. 2, New Super Mario Bros. Wii, New Super Mario Bros. U.
Street Fighter IV is the most recent instalment of the Street Fighter series. As the name suggests, it is a fighting game,
requiring quick reflexes and precision timing from the player. Since IV introduces some new features, this is the only instalment that is valid.
Halo is a franchise of mostly first-person shooter games, also a
staple of the action genre. While the multiplayer component of this game is often considered more important, it features an elaborate single-player component. Due to some new elements introduced in the game, only Halo III and Halo IV are considered valid.
4.1.2.2 Adventure games
Sam & Max is an early adventure game that has recently seen a
reboot. The focus on narrative and exploring the game world make it easily identifiable as an adventure game. Any of the recent episodes is valid: Sam & Max Save the World, Sam &
Max Beyond Time and Space, Sam & Max: The Devil's Playhouse.
Similar to Sam & Max, Tales of Monkey Island is an early adventure game that was rebooted not too long ago. Any of the following episodes is considered valid: Launch of the
Screaming Narwhal, The Siege of Spinner Cay, Lair of the Leviathan, The Trial and Execution of Guybrush Threepwood, Rise of the Pirate God.
The Walking Dead: The Game is an original adventure game
based on the television series of The Walking Dead, in turn based on the comic book series of the same name. It is an interactive story game, with branching dialogue options and multiple endings depending on the decisions of the player. Any of the episodes in season 1 and 2 are valid. Season 1: A New
Day, Starved for Help, Long Road Ahead, Around Every Corner, No Time Left, 400 Days. Season 2: All That Remains, A House Divided, In Harm's Way, Amid the Ruins, No Going Back.
4.1.2.3 Role-playing games
Baldur’s Gate is a classic role-playing game that is based on
the tabletop RPG Dungeons and Dragons. Very little changed between iterations and the remakes, so all the following titles are valid: Baldur's Gate, Baldur's Gate II, or their Enhanced editions.
Pokémon, aside from being a worldwide phenomenon, is also a
longstanding series or RPG games. Players take their characters through and adventure, and along the way the characters grow and (literally) evolve. Over the years the core gameplay has changed very little, sometimes to the dismay of critics, so all main games are considered valid for this study:
Red, Blue, Yellow, Gold, Silver, Crystal, Ruby, Sapphire, Emerald, FireRed, LeafGreen, Diamond, Pearl, Platinum, HeartGold, Soulsilver, Black, White, Black 2, White 2, X, Y, Omega Ruby, Omega Sapphire.
Final Fantasy is another longstanding series from Japan, with
most games emphasising narrative. However, the gameplay differs substantially between certain titles in the series, so for this study only the following Final Fantasy games are considered valid: VII, VIII, IX, X, X-2, XII, XIII, XIII-2,
Lightning Returns: Final Fantasy XIII.
4.1.2.4 Simulation games
Sim City is a popular simulation game, that has also seen some
use in educational context. It aims to simulate the planning and development of a city, with the player acting as a supreme entity pulling the strings. Sim City 2000 and Sim City 3000 are similar enough for the purpose of this study.
Euro Truck Simulator places the player in the role of a truck
driver in Europe. As the name suggests, it simulates driving cargo from one place to another in a truck. Both Euro Truck
Simulator and Euro Truck Simulator 2 are valid.
Nintendogs is a more social simulation game. It simulates the
keeping of a dog or cat, with which the player can interact through various means. While there are many games in this series, they are all so similar that they are all considered valid.
Nintendogs: Dachshund & Friends, Lab & Friends, Chihuahua & Friends. Nintendogs: Best Friends, Dalmatian & Friends. Nintendogs + Cats: French Bulldog & New Friends, Golden Retriever & New Friends, Toy Poodle & New Friends.
4.1.2.5 Strategy games
Civilization is a long running series of turn-based strategy
games. The turn-based nature allows for players to think through their strategies carefully, whilst the game provides
many different routes to victory. Due to the similar nature, both
Civilization IV or Civilization V are valid.
StarCraft is also a long running series, but these are real-time
strategy games. This puts some more emphasis on the physical reaction speed of the player compared to the Civilization series. Little has changed gameplay wise between the games, so StarCraft, with or without the expansion Brood War,
StarCraft II; Wings of Liberty, and StarCraft II: Heart of the Swarm are all valid.
Last in the strategy genre is Portal, a puzzle game with some focus on narrative as well. Whilst it may not be the purest puzzle game, it is one that became quite popular, which is unusual for puzzle games. Since the two are so similar, both
Portal and Portal 2 are valid.
4.1.2.6 Sandbox games
Minecraft is possibly one of the most well-known sandbox
games, as it exploded in popularity since its release in 2011. It incorporates elements from other genres, but at its heart it is a three-dimensional creative block builder where players can create whatever they want.
Garry’s Mod is a three-dimensional physics sandbox based on
the Half-Life 2 engine, but later made into a standalone game. It features no goals, but allows players to manipulate items and props in the virtual world.
Terraria is often described as a two-dimensional Minecraft, as
it shares many properties with the game. The game leans a bit more heavily on the elements it borrows from other genres, but ultimately there is no goal in the game towards which the player is steered, other than the goals players set themselves.
4.2. Questionnaire Items
All items in the questionnaire took the form of a question about how strong a focus was in the game in question, which the participant could answer on a five point Likert scale ranging from “Very Strong” to “Barely There”. In addition, participants could also answer “Not Applicable” should they feel the item was not relevant to the game in question, or “Can’t Remember” should they be unable to remember whether said element was present in the game or not.
Included in Appendix A are all the questionnaire items as they were presented to the participants. The first line of each item contains the main statement, the proceeding lines contain a more elaborate explanation.
For each Bartle type there are two items, and for each ACE2 type there are three items. We tried to balance this, but due to the preciseness of Bartle’s types it proved impossible to come up with additional items that were sufficiently unique. Likewise, we tried lowering the number of items for the ACE2 model, but since we aimed to create more abstract types, having only two items was too limiting in covering the scope of each type. Below we shall discuss the items based on their associated player type, starting with the Bartle types.
4.2.1 Bartle types
All the items for the Bartle types were constructed using Bartle’s original paper [6], by looking at which kinds of behaviour Bartle associated with each type.
4.2.1.1 Bartle's Achievers
For Bartle’s achievers we constructed items A1 and A2 (Appendix A, item 1 and 2). According to Bartle, players that seek to act on the world aim to beat various game goals and set out to achieve them, which is reflected in A1. Oftentimes they look to increase some measure of their prowess, be it experience points or hoarding large amounts of currency. This is reflected in A2.
4.2.1.2 Bartle's Explorers
For Bartle’s explorers we constructed items A4 and A5. A4 is reflective of the explorers desire to explore the breadth of the
game, whereas A5 reflects the desire to explore the depth of the game. These are the two main ways in which players find out as much as they can about the game according to Bartle [6].
4.2.1.3 Bartle's Socialisers
A8 and A9 were constructed for Bartle’s socialisers. In his taxonomy, socialisers are predominantly concerned with their social status (A9) and engaging in social activity (A8).
4.2.1.4 Bartle's Killers
Lastly, for Bartle’s killers we constructed A10 and A11. This category was the hardest to construct items for, since their main interest in the game is to cause distress in other players (A10). However, Bartle briefly mentions that sometimes the imposition upon others does not have to be detrimental for the other player, and in rare cases killers might forcefully help another player [6], which is why A11 was included.
4.2.2 ACE2 types
For the ACE2 types we spent a large quantity of time observing various games, including but not limited to the list of games selected for the questionnaire. Most of these observations were in the form of YouTube “Let’s Plays” or footage of players streaming their gameplay. By keeping track of the various kinds of behaviours that these players exhibited in all these games we were able to categorise all this in three items per ACE2 type. Since the Bartle achievers and explorers share traits with the ACE2 achievers and explorers, they also reuse some of their items. This had the added benefit of keeping down the number of items for the participants.
4.2.2.1 ACE2’s Achievers
This category shares the most overlap with its Bartle equivalent, using both A1 and A2 in addition to A3. We felt that another element of achievement in some games can come from the mastery of the game, most notably in action puzzle games such as Tetris, but also games such as Super Meat Boy, which provide incredibly tough gameplay where part of the enjoyment can come from mastering the mechanics of the game.
4.2.2.2 ACE2’s Explorers
Since exploring territories was a little too specific, we dropped A5 for the ACE2 explorers and only reused A4, in addition to the new A6 and A7. A6 and A7 are very similar, since they both deal with finding solutions to levels or problems. The difference lies in how the player goes about in doing it. A7 is the most common form of play, where the player seeks to find the best solution, whether it be using as few moves as possible to complete a game or placing units for optimal gains in
Civilization. A6 is more about finding unusual ways to
accomplish a goal, for example using only a certain kind of pokémon in a Pokémon game or creating a incredibly convoluted solution in SpaceChem.
4.2.2.3 ACE2’s Engagers
For the ACE2 engagers we created items A12, A13, and A14. Some players play games purely to enjoy the narrative that is provided, sometimes caring very little for the gameplay itself. An example of this can be observed in the Warcraft games, which some players will play just to experience the story. Another way to engage with a game is in the visuals it provides, even though this sometimes comes at the cost of reduced gameplay. An example of a game that facilitates this behaviour is Monument Valley, where the main emphasis of the game is on the Escher-like vistas. This is reflected in A13. Lastly, some players seek to build ‘relations’ with the non-player characters in a game. Dating simulation games are a prime example of games in which this behaviour is displayed, but it can also be observed in games such as Mass Effect and
4.2.2.4 ACE2’s Creators
Finally, for the ACE2 creators we constructed items A15, A16, and A17. A fair amount of modern games feature level creators as a way to increase the lifespan of the game. Some players really enjoy this, and spend a great deal of their time with the game on creating new levels, which is reflected in A15. Level creation is featured in games like Worms, Super Meat Boy, and
Minecraft in the form of adventure maps. Minecraft can also
double as the virtual canvas for players to display their architectural creations or pixel art, A16 reflects this kind of behaviour. Whilst less common, some games also allow for players to create their own narrative, either through creating a custom level or campaign (Minecraft and Crashlanders) or through the creation of movie clips (machinima). Examples of the latter can be observed in the Rockstar Editor in GTA V, or the Halo scene creator. A17 reflects this kind of player behaviour.
4.2.3 Objectivity of the items
Since we constructed our own items for this questionnaire we were very mindful of the fact that we could influence the results favourably for ACE2 just by how we chose the items. To prevent this we took special care to focus on the actual behaviours we observed in gameplay footage, rather than on what would best differentiate the new model from Bartle’s model.
4.3. Questionnaire Procedure
Upon loading up the questionnaire the participant was greeted with an introduction screen where the goal of the questionnaire was briefly explained, as well as explaining what was expected of the participant in their answering of the questions. In Appendix B we included a screen capture of the initial screen. Special care was taken to ensure that the participants knew what was expected of them without guiding them too explicitly. While the text does steer the participants towards a certain answer, trial runs showed that without this explanation participants were confused by some of the items. Some of this confusion remained, but we will discuss this later.
Upon starting the questionnaire the participant was presented with a screen in which they could select the games with which they felt comfortable enough to answer questions about. This screen is included in Appendix C.
For every selected game the participant was asked to fill in the form shown in Appendix A. In addition to the questions, the screen also showed the games in question, and a small reminder on how to judge certain questions. Again, whilst this seems to steer towards a certain result, trial runs showed that people were confused with some items without the reminder, and several testers recommended to add this paragraph. All questions in the screen had additional info in their tooltips,
which could be accessed by hovering over them with the mouse cursor. All items and their tooltips are included in Appendix A.
After finishing the questionnaire the participant was shown one of two screens, depending on whether they selected at least a single game or none at all. We included these in Appendix E and Appendix F respectively. Participants could opt to leave their email address, stored separately from the questionnaire results, to be informed of the results upon completion of the thesis.
4.4. Questionnaire Analysis
When analysing the results we transformed the answers given by the participants into scores for each item. We subsequently combined these with items in the same category and took the average. In the case a participant answered “Can’t Remember” we did not take this answer into account in the calculations. This gave us a score for every category for both models, which we mapped on the plots shown below in Figures 3 and 4. The scores range from 0 to 5, where 0 means that this player type is not represented in the game at all according to the participants, and 5 that this is one of the main foci of the game.
The reasoning behind this was that now games are identifiable through their shape on the plot, as well as making for an easier visual comparison of differences between the models in the results. The visualisation is interactive and allows for a more hands on approach in analysing the data. In addition, it makes the gathered data more easily digestible by those who are interested in the results.
5. RESULTS
First we will discuss how the models compare over all games., looking at the overall picture of the data. Second, we will take a closer look at each of the genres and how the model dealt with them.
5.1. All Games
By calculating the average for all player types among all games for both models we were able to create the plot that can be observed in Figure 5 on the following page. The socialisers and engagers, as well the killers and creators have been put on the same ends of the axes in order to make comparison easier. While the two shapes are similar, the ACE2 model has three directions in which it expands, whereas Bartle’s model only expands in two directions substantially. This indicates that participants were able to categorise with a higher degree of nuance in ACE2, since more relevant options were available to them. When looking at the data for all games separately rather than averaged, this becomes clearer still, although the picture is a little messy. Appendix G1 plots the results for all games individually using Bartle’s model. We observe that for Bartle’s
Figure 3: Bartle Axes
model that the killer and socialiser axes are sparsely populated with medium to low scores. Appendix G2 also plots the data for the individual games, but using ACE2 instead. We see that the achiever, explorer, and engager axes are densely populated with high scores for the ACE2 model. While the creator axis is also sparsely populated, the values on there reach higher scores, which suggests that for the games in which it was relevant, it was highly so.
5.2. Genres
Appendices H to M include the graphs for each of the six genres. We will go over them one by one, noting interesting differences between the two models or peculiar features of the models for that particular genre.
5.2.1 Action games
Appendix H1 plots the averages of the data for action games for both models. We observe a slight difference between the two models. Overall, the Bartle killers are more relevant for action games than the ACE2 creators, but not significantly so p < 0.07.
However, things become more interesting when taking a look at how the individual action games score in H2 for Bartle and H3 for ACE2. As one can see in H3, the graphs for the three games are all fairly similar to the average in H1. This indicates that the average is a fair representation of the three action games. Yet when looking at Bartle things are a little different. All three games have distinct shapes, and none of them are close to the average in H1. It seems that the more low level nature of Bartle’s model is unable to generate a good abstract view of the action genre.
5.2.2 Adventure games
Appendix I1 plots the averages for adventure games for both models. The shapes for adventure games are very distinct when compared to other genres, for both Bartle’s model and the ACE2 model. All individual games were similar in shape to their average, so both models seem to be able to identify adventure games fairly smoothly. However, Bartle’s model only utilises two of the four axes, whereas ACE2 uses three. This allows for a higher degree of nuance in the categorisation of adventure games.
5.2.3 Role-playing games
Appendix J1 plots the averages for Role-playing games for both models. Like adventure games, role-playing games all have similar shapes and are thus close to their average for both models. Bartle mainly utilises two of the axes, but it does not completely ignore the other two axes. The ACE2 model is again capable of showing more nuance by using three axes, but the creator axis is almost completely ignored.
5.2.4 Simulation games
Appendix K1 plots the data for simulation games for both models. Simulation games feature quite distinct shapes in both models, although all scores across both models are on the lower side. It seems that simulation games do not fit it quite as well as the other genres.
In appendix K2 and K3 we plot the individual games for Bartle and ACE2 respectively. Here we observe a similar situation to the action genre, except this time it is the ACE2 model that has a clear outlier game. We will come back to this in the findings chapter.
5.2.5 Strategy games
Appendix L1 plots the data for strategy games for both models. When taking a look at the Bartle model for RPGs in Appendix J and strategy games in Appendix L1, we can clearly see that the two shapes are very similar. While the ACE2 model also has similarities, it scores lower on the engager axis and higher on the creator axis.
Appendix L2 and L3 plot the data for all strategy games individually for Bartle and ACE2. Here we observe that the Bartle model has an outlier game that differs quite substantially from the average.
5.2.6 Sandbox games
Appendix M1 plots the averages of the data for the sandbox games for both models. Sandbox games generate distinct patterns in both models, making them easily identifiable. The Bartle model shows a little more variance in the individual games than ACE2, as can be observed in Figures 6 and 7, but
Figure 5: Average All Games
overall the individual games are similar to the average in both models.
When looking at the creators axis in the individual games (Appendices H3-M3), we can see that with a single exception all high scores are in the sandbox genre. The one exception is in simulation games, where the city builder Sim City also scores high on the creators axis. The difference in scores on the creators axis for sandbox game and any other genre is significant, with an unpaired t test giving a value of p < 0.04 for sandbox versus simulation, and p < 0.003 for sandbox versus other genres.
6. DISCUSSION
In this chapter we will discuss our findings and reflect on the limitation of this study.
6.1. Findings
When comparing the various axes with unpaired t tests across various genres we found very little significant differences, even though by observing the graphs there seems to be a substantial difference. An explanation for this is that it seems not all participants understood that the questionnaire was focused exclusively on single-player games and thus still used Bartle’s killers and socialisers, whilst one would assume that these play no role in single-player games. However, the creators type forms the exception to this, showing a overall highly significant (p < 0.04) difference between the sandbox genre and others. This supports our hypothesis that the creative player is a unique kind of player that should be considered separately from other player types.
Staying with the creative type, we want to briefly reflect on the ACE2 outlier in the simulation games, Sim City. Due to their nature, simulation games will often borrow elements from other game genres in order to create the best simulation. In the case of Sim City, which is a city builder type game, it is no surprise that the creator player type is strongly represented whereas it is not in the other simulation games. The answers for the items pertaining to the creator type for Sim City differ significantly from those for the other two simulation games,
Euro Truck Simulator and Nintendogs, with p < 0.0133. This
strengthens our hypothesis that creative gameplay is worth considering separately even further.
Lastly, while none of the results were significant, we did find that the ACE2 model made it easier to differentiate between genres by eye. When looking at Figure 8, the three shapes in Bartle’s model are nearly identical even though they belong to very different genres. This is more accurately reflected in
Figure 9, where the three games feature have distinct shapes, allowing for simple and intuitive identification when observing the data. This shows us that while the differences might not be significant in a statistical sense, the models do offer some use in creating intuitive comparisons that can help people in finding similarities and differences between games.
6.2. Study Limitations
Whilst the creator is quite solid as a player type, it does not fit perfectly in the model. It is located between Acting and Aesthetics, yet level creation is also considered as creator behaviour, even though this is quite possibly more related to mechanics than aesthetics. This shows that the model has clear limitations in the range of player behaviours it can model. The questionnaire featured three items per ACE2 category, whereas it only had two items for each Bartle category. Additionally, while creating the items a bias might have been introduced, which would lead the results to be favourable for the ACE2 model. To combat this we were careful with steering the answers too much when constructing the items of the questionnaire. However, it was difficult finding the balance between ensuring the participants understood what was expected of them whilst making sure not to influence their decisions too strongly. This partially explains the scores that the Killer and Socialiser category received. Considering the fact that the questionnaire dealt with single-player games, and these categories are very much multi-player oriented, we expected these to score lower than they did.
The number of respondents was limited, barely reaching forty filled in questionnaires. This meant that the less popular games received very few responses, with the lowest game only having a single respondent.
7. CONLUSION
At the onset of this thesis we set out to create a model that would perform better than Bartle’s model in differentiating between single-player video games, with a particular focus on sandbox style games. In the data we gathered through the questionnaire we managed find some evidence which indicates that the new ACE2 model allows for a nuanced labelling of single-player games, and that even though creative behaviour does not feature often, when it does it is a defining feature. However, the Bartle model performed better than we had expected, showing again that despite its shortcomings it is still a usable model. We hope that this research has shown that approaches to player modelling similar to Bartle’s are
Figure 8: Super Mario Bros., Sim
worthwhile pursuing, even though the model proposed in this thesis may not be the best way to do so.
8. FUTURE WORK
The research presented in this thesis was mostly exploratory, so future work should seek to solidify the observations made. There are multiple ways to go about this, most notably by conducting a more widespread survey with many more respondents. Additionally, using a broader selection of games can also be of help in further testing the validity of the model proposed in this thesis. To test how well the model preforms in classifying games based on the behaviours they facilitate, a study could be conducted that asks participants for a particular game they enjoy. Then, based on data gathered previously, the model can recommend a new game based on the prominent values in the game named by the participant. The participant can then spend some time with the game, and rate their appreciation of the game. This servers the dual purpose of testing out the model, as well as the assumption that players like a certain kind of behaviour, which they look for in games. Future research in player modelling in general should aim to encompass a wider variety of players and behaviour. It seems like using only four axes does not do justice to the variety of ways in which people enjoy games. Perhaps introducing a new axis will help in allowing more precise classification.
9. ACKNOWLEDGEMENTS
We would like to use this opportunity to thank Sander Bakkes for his invaluable feedback and support as supervisor to this thesis.. In addition, we would like to thank everyone who took the time to participate in the questionnaire and/or spread the word about it. Lastly, we would like to thank one particular artist, whose DJsets have provided the background music for the majority of the process: Nuno Dos Santos (https:// soundcloud.com/nunodossantos).
10. REFERENCES
1. Apperley, T. H. Genre and Game Studies: Toward a Critical Approach to Video Game Genres. Simulation
Gaming, 37-1 (2006), 6-23.
2. Avedon, E. M., Sutton-Smith, B. The Study Of Games. John Wiley, (1979).
3. Bakkes, S. C. J. Rapid Adaptation of Video Game AI. PhD disseration, Universiteit van Tilburg, (2010).
4. Bakkes, S. C. J., Spronck, P. H. M., Lankeveld, G. Player Behavioural Modelling for Video Games. Entertainment
Computing, 3 (2012), 71-79.
5. Bakkes, S. C. J., Spronck, P. H. M., Postma, E. O. Best-response Leaning of Team Behaviour in Quake III.
Proceedings of the IJCAI 2005 Workshop on Reasoning, Representation, and Learning in Computer Games, (2005),
13-18.
6. Bartle, R. Hearts, Clubs, Diamonds, Spades: Players Who Suit MUDs. (1996). http://mud.co.uk/richard/hcds.htm 7. Bartle, R. Player Type Theory: Uses and Abuses. Casual
Connect. (2012). https://www.youtube.com/watch?
v=ZIzLbE-93nc
8. Bartle, R. The Decline of MMOs. (2013). http://mud.co.uk/ richard/The%20Decline%20of%20MMOs.pdf
9. Bateman, C. 21st Century Game Design. Charles River Media, (2005).
10. Canossa, A. Give Me A Reason To Dig: Qualitative Associations Between Player Behaviour Found in
Minecraft and Life Motives. Proceedings of the
International Conference on the Foundations of Digital Games, (2012), 282-283.
11. Fairclough, C., Fagan, M., MacNamee, B., Cunningham, P. Research Directions of AI in Computer Games.
Proceedings of the 12th Irish Conference on Artificial Intelligence & Cognitive Science, (2003), 333-344.
12. Hunicke, R., LeBlanc, M., Zubek, R. MDA: A Formal Approach to Game Design and Game Research.
Proceedings of the AAAI Workshop on Challenges in Game AI, 4 (2004).
13. Keirsey, D. Please Understand Me II. Prometheus Nemesis, (1998).
14. Lazzaro, N. Why We Play Games: Four Keys to More Emotion Without Story. Game Developers Conference, March (2004).
15. Rohs, M. Preference-based Player Modelling for
Civilization IV. (2007).
16. Rosewater, M. Our Three Favorite Players: Timmy, Johnny, and Spike. (2002). http://archive.wizards.com/ Magic/magazine/article.aspx?x=mtgcom/daily/mr11b 17. Schaeffer, J. A Gamut of Games. AI Magazine, 22-3
(2001), 29-46.
18. Solarski, C. The Aesthetics of Game Art and Game Design.
Gamasutra, (2013).
http://www.gamasutra.com/view/feature/185676/ the_aesthetics_of_game_art_and_.php?print=1
19. Stewart, B. Personalities and Playstyles: A Unified Model.
Gamasutra, (2011).
h t t p : / / w w w. g a m a s u t r a . c o m / v i e w / f e a t u r e / 6 4 7 4 / personality_and_play_styles_a_.php?print=1
20. Thue, D., Bulitko, V., Spetch, M., Wasylishen, E. Interactive Storytelling: A Player Modelling Approach.
AIIDE, (2007), 43-48.
21. Webber, N. Controlling a Sandbox. Ctrl-Alt-Play: Essays
Appendix A
All statements were prefaced with “How strong is the focus on…”
— Achievers
1. Winning.
Beating levels or opponents in the game.
2. Gaining points.
Increasing a value, be it experience points, gold, achievement points, or anything similar.
3. Mastering the game.
Getting better and better at the game. The learning curve is a large part of the game.
— Explorers
4. Finding interaction between game elements.
Discovering how game elements interact with each other, finding the limits of the game engine.
5. Finding unexplored territories.
Discovering areas in the game that few other players have been to.
6. Finding alternate strategies.
Beating levels in different ways than what is most obvious; finding new ways to accomplish something.
7. Finding the optimal solution or setup.
Finding the optimal solution for a puzzle, or finding equipment/weapon combination that provide the
best stat boosts.
— Socialisers
8. Getting to know new players.
Meeting new players and communicating with them to get know them better.
9. Improving your social status in the community.
Getting more players to know you and see you in a positive light.
— Killers
10. Causing distress in other players.
Interacting with other players in the game world as to ruin their day. Often by killing their in game
character.
11. Imposing yourself on other players.
(Forcefully) interacting with other players in the game world.
— Engagers
12. Experiencing the narrative of the game.
The game features an extensive story.
13. Experiencing the visuals of the game.
The game provides stunning views, or features a particular art style.
14. Interacting with the Non-player Characters of the game.
Engaging in dialogue with computer controlled characters, or in other ways interacting with them.
— Creators
15. Creating new levels.
Constructing new levels that are playable by others.
16. Creating your own structures, landscapes, or visuals.
Using the game as a creative outlet. An example of visuals would be pixel art.
17. Creating your own narrative.
Appendix G
All games.
Appendix H
Action games.
1.
Appendix I
Adventure games.
Appendix J
Appendix K
Simulation games.
Appendix L
Strategy games.
1.
Appendix M
Sandbox games.
