A Model of Player Motivations
Asking MMORPG players why they play reveals a dazzling array of varied motivations. Indeed, this wide variation illustrates why MMORPGs are so appealing - because they are able to attract people with very different motivations for playing.
I play MMORPGs with my husband as a source of entertainment. Overall it can be a cheaper form of entertainment where you can spend quite a bit of time with a significant other. To play well you end up developing more ways of communicating. [DAoC, F, 31]
I like the whole progression, advancement thing ... gradually getting better and better as a player, being able to handle situations that previously I wouldn’t have been able to. [EQ, M, 48]
No one complains about jobs or other meaningless things. It's a great stress reducer. I like that I can be someone else for a couple hours. [SWG, M, 28]
Currently, I am trying to establish a working corporation within the economic boundaries of the virtual world. Primarily, to learn more about how real world social theories play out in a virtual economy. [EVE Online, M, 30]
Being able to articulate and build an empirical model of these underlying motivations provides an important foundation to several other avenues of research. First, it gives us a meaningful way to differentiate players from one another as well as allowing us to explore, for example, how older gamers are different from younger gamers. Second, a model of player motivations provides a tool to explore in-game preferences and behaviors. For example, which players are most likely to become guild leaders or which players are most likely to exhibit problematic usage?
====Bartle’s Player Types are a well-known model of player motivations. In that paper, Bartle provides important insight into how players may differ from one another and he suggests a categorization of 4 Types (Socializer, Achievers, Killers and Explorers) based on two underlying axes. Recently, Bartle further developed this model into a model of 8 Player Types (see Designing Virtual Worlds by Bartle, 2004).
Bartle’s theoretical model, while providing important insight, suffers from several limitations.
1) Proposed components of each Type may not be related. For example, Bartle proposes that role-playing and socialization both fall under the same Type, but they may not be highly-correlated.
2) Proposed Types may overlap with each other. For example, aren’t members of raid-oriented guilds both Achievers and Socializers? But in Bartle’s Types, they are on opposite corners of the model.
3) The purely theoretical model provides no means to assess players as to what Type they are. But more importantly, without resolving the problem in (1), any attempted assessment of players based on this model might be creating player types rather than measuring them.
In essence, it would be hard to use Bartle’s model on a practical basis unless it was validated with and grounded in empirical data. For example, Bartle suggested that different Player Types influenced each other in certain ways. But unless we have a way of assessing and identifying players of different Types, theories built on top of Bartle’s model are inherently unfalsifiable. While a “Bartle Test” (not made by Bartle) does exist, the dichotomous, forced-choice nature of that assessment tool merely perpetuates the assumptions of Bartle’s Types rather than validating them. In this article, I present a methodology used to validate Bartle’s model and how the results are similar and different from Bartle’s proposed model.
I used an iterative process to validate, expand and refine a player motivation model empirically over the past few years. First, a list of possible motivations for playing an MMORPG was generated from existing literature (such as Bartle’s Types) or open-ended responses from earlier surveys (see here and here).
These motivations were then converted into survey questions, such as:
How important is it you to level up as fast as possible?
- Not Important At All
- Slightly Important
- Moderately Important
- Very Important
- Tremendously Important
The full list of questions used and information on administering the assessment tool is provided here.
Respondents then rated each statement on an online survey. In the current data set, 3200 respondents completed an inventory of 39 items. A factor analysis was then performed on this data to separate the statements into clusters where items within each cluster were as highly correlated as possible while clusters themselves were as uncorrelated as possible. This methodology achieved three goals:
1) Ensured that components of each motivation are indeed related.
2) Ensured that different motivations are indeed different.
3) Provided a way to assess these motivations.
I’d like to stress the iterative nature of this endeavor. The open-ended responses and brainstorming hint at the boundaries of the territory, tested by the factor analysis, at which point I return to open-ended responses to better explore the areas the factor analysis identified as coherent constructs. Respondent responses then inevitably shed light on nuances of motivations that I generate further statements to explore.
The current data set revealed 10 factors that then neatly factored into 3 overarching factors. We can think of these as subcomponents and main components respectively. Detailed information on the factor analyses is provided on the last page of this article. The 3 main components are presented here with their subcomponents.
The descriptions below emphasize what it means to score high on the subcomponents. Scoring low on these subcomponents is just as revealing. For example, a player who scores low in the Socializing subcomponent would prefer game mechanics that don’t force them to interact with others (i.e., character dependencies in EQ - binds, teleports, rezzes). For the sake of brevity, the “flip side” of every subcomponent is not explicitly stated.
The Achievement Component:
Advancement: Gamers who score high on this subcomponent derive satisfaction from reaching goals, leveling quickly and accumulating in-game resources such as gold. They enjoy making constant progress and gaining power in the forms offered by the game - combat prowess, social recognition, or financial/industrial superiority. Gamers who score high on this subcomponent are typically drawn to serious, hard-core guilds that can facilitate their advancement.
Mechanics: Gamers who score high on Mechanics derive satisfaction from analyzing and understanding the underlying numerical mechanics of the system. For example, they may be interested in calculating the precise damage difference between dual-wielding one-handed swords vs. using a two-handed sword, or figuring out the resolution order of dodges, misses, and evasions. Their goal in understanding the underlying system is typically to facilitate templating or optimizing a character that excels in a particular domain.
Competition: Gamers who score high on this subcomponent enjoy the rush and experience of competing with other gamers on the battlefield or economy. This includes both fair, constrained challenges - such as dueling or structured PvP/RvR, as well as unprovoked acts - such as scamming or griefing. Gamers who score high on this subcomponent enjoy the power of beating or dominating other players.
The Social Component:
Socializing: Gamers who score high on this subcomponent enjoy meeting and getting to know other gamers. They like to chit-chat and gossip with other players as well as helping out others in general - whether these be less-experienced players or existing friends. Gamers who score high on this subcomponent are typically drawn to casual, friendly guilds.
Relationship: Gamers who score high on this subcomponent are looking to form sustained, meaningful relationships with others. They do not mind having personal and meaningful conversations with others that touch on RL issues or problems. They typically seek out close online friends when they need support and give support when others are dealing with RL crises or problems.
Teamwork: Gamers who score high on Teamwork enjoy working and collaborating with others. They would rather group than solo, and derive more satisfaction from group achievements than from individual achievements. Gamers who score low on this subcomponent prefer to solo and find it extremely important to be self-sufficient and not have to rely on other gamers. They typically group only when it is absolutely necessary.
The Immersion Component:
Discovery: Players who score high on Discovery enjoy exploring the world and discovering locations, quests or artifacts that others may not know about. They enjoy traveling just to see different parts of the world as well as investigating physical locations (such as dungeons and caves). They enjoy collecting information, artifacts or trinkets that few others have.
Role-Playing: Players who score high on Role-Playing enjoy being immersed in a story through the eyes of a character that they designed. These players typically take time to read or understand the back-story of the world as well as taking time to create a history and story for their characters. Also, they enjoy role-playing their characters as a way of integrating their character into the larger ongoing story of the world.
Customization: Players who score high on this subcomponent enjoy customizing the appearance of their characters. It is very important to them that their character has a unique style or appearance. They like it when games offer a breadth of customization options and take time to make sure that their character has a coherent color scheme and style.
Escapism: Gamers who score high on Escapism use the environment as a place to relax or relieve their stress from the real world. These players may use the game as a way to avoid thinking about their RL problems or in general as a way to escape RL.
The subcomponents generated by the factor analysis are NOT player types. It is NOT the case that we have come up with 10 boxes that we can put players in, but rather, we have revealed 10 subcomponents that co-exist and together reveal the motivations of a player. Bartle assumed that your underlying motivations “suppressed” each other. In other words, the more of an Achiever you were, the less of a Socializer, Explorer and Killer you could be, but just because you like ice-cream doesn’t mean you will hate pasta. The assumption of polarized motivations is also not supported by the correlations of the current data set. The Achievement component is not negatively correlated with the Socializing component as Bartle’s model would predict. In fact, it is mildly positively correlated (r = .10, p < .001). A more detailed comparison between the Types vs. Components approach is presented in a separate article.
The factor analysis also revealed several important ways where the data differed from Bartle’s theoretical model:1) Socializing and Role-Playing: Bartle proposed that people who like to chat and make friends are also the people who like to role-play. These are in fact two independent motivations.
2) Achieving and Competing: While Bartle proposed that Achievers and Griefers were separate Types, they are in fact fairly correlated with each other. The Advancement and Competition subcomponents are correlated at r = .41, p < .001.
3) The Explorer Type: Bartle construed Explorer’s as people who enjoyed both exploring the world, gathering information as well as enjoying tinkering with the underlying system and mechanics. These are also in fact two different kinds of people. My earlier attempts to find the Bartle Explorer failed until I tried to look for those two constructs separately. In other words, there is a Discovery subcomponent that revolves around finding and accumulating knowledge that is separate from the Mechanics subcomponent that is interested at unraveling and tinkering with the game mechanics.
4) Immersion: There is also a separate set of motivations that didn’t exist in Bartle’s Types. The Immersion subcomponents revolve around story-line, role-playing, fantasy, customization and escapism and are independent of the Socializing motivations.
In his book, Designing Virtual Worlds (2004), Bartle critiqued an earlier model of player motivations also derived from factor analyses. Here I present and respond to some of those critiques.1) The motivations suggested by the survey are implicit in the questions. While that is true, the survey doesn’t implicitly presume a grouping of statements and that was the more important goal of the survey - to understand what statements did correlate to form a “subcomponent” rather than simply assuming such. For example, we found that socializing and role-playing are independent constructs.
2) Brainstorming motivations is as subjective as brainstorming player types. The important difference though is that the brainstormed motivations are then empirically tested to find validated constructs. The data showing discrepancies with Bartle’s original Types illustrate how player motivations can’t simply be brainstormed. They must be tested.
3) The labeling of the facets is not provided by the factor analysis. Nor are the labels of the Player Types inherent in any way. The Player Type labels suffer from the more serious problem of labeling a cluster of motivations that are not really correlated.
4) Some of the facets overlap, but some don’t. And the only way we can know this is by having a validated tool for assessment and then observing the underlying correlations. And in fact, the Player Types also overlap, but this was not apparent until we had a way of measuring those motivations. More importantly, the current model of 3 main components is largely uncorrelated with each other (all around r = .10).
====Details of Factor Analysis
A principal components analysis was performed to arrive at a parsimonious representation of the associations among the 39 items. 10 factors were extracted with eigenvalues greater than 1. Together, these factors accounted for 60% of the overall variance. The chart below shows the factor loadings of the survey items used.
The scores for all subcomponents were generated for each of the 3200 respondents using a regression method. Another principal components was performed on the 10 subcomponent scores. 3 factors were extracted with eigenvalues greater than 1. Together, these 3 factors accounted for 54% of the overall variance. These 3 factors are largely uncorrelated (r’s ~ .10). The chart below shows the factor loadings of the subcomponents on the 3 main components.