To Bartle or Not to Bartle?: Capturing Player Behavior in Social Games

From horoscopes to the Hogwarts Sorting Hat to Facebook quizzes, we humans seem to have a certain fascination with categorizing ourselves in terms of group identities and fitting ourselves to archetypal models. Maybe this is because of the various psychological perks that come with coding oneself within such sets of labels: we’re provided with a sense of belonging, we’re given a chance to externally corroborate internal thoughts about our “true” selves; we’re able to find meaning by aligning ourselves with an ideal. For instance, I have no doubt in my mind that my essence has a 73% correlation with the cunning vigilantism of Batman or that of all the characters from The Wire I am most like Omar. I know these things about myself; they're obvious, and simple personality tests can verify them in a manner for all the world to see.

Of course, a twenty question test made up by some fanboy and posted online lacks a certain scientific rigor and offers little use beyond helping me to confirm my own positive assumptions about myself. However, there certainly exist other typologies that have come to establish themselves as valid and/or of tangible, applied value. For example, it is not uncommon for Myers-Briggs personality profiling to be used as an aptitude assessment tool by corporate management and human resource officials.

It is in a similar vein that player typologies are thought to provide value. Beyond simply allowing players to think of themselves within the context of a given gamer style or role, player types may also offer useful insights for designers, providing convenient heuristic hooks on which to hang design features when attempting to ensure that a game speaks to a particular audience or, alternatively, has “something for everyone.”

By far, the most well known framework of player types is Bartle’s condensed set of four (as opposed to his full model of 8 types). Bartle’s conceptualization of achievers, explorers, socializers, and killers is elegant and intuitive, allowing it to be easily grasped by players, designers, and even those unversed in MUDs and MMOs. Perhaps this is why it often gets applied to gamers in general, despite being specifically extracted from observations made in MUD environments in particular. And, unfortunately, the games being considered when these player types are discussed often have little in common with the gaming scenarios from which the framework was originally derived.

For example, I’ve recently witnessed attempts to apply Bartle types when describing the player habits found in social games. I would argue that such an application is mistaken, considering 1) some key distinction between typical MUDs/MMOs and social game environments, 2) the resulting differences in player actions and interactions, and 3) the underlying differences in what is meant when each of these types of designs are considered “successful”.

The most obvious distinction between MUDs/MMOs and social games may be the difference in the play “spaces” themselves. MUDs and MMOs are often large-scale “physical” worlds, geographies populated by players who learn to operate by both hard-wired rules and emergent social norms. Indeed, MUDs and MMOs are sandboxes of actions and interactions, with explicit reward structures for the achievers as well as hang-outs and /emotes for the socializers occupying the space. In comparison, one might argue that until now many social games have been interface-bound sets of mechanics, in which social interactions are pre-coded to reflect a certain type of player goal. That is, gameplay in social games is often about individual achievement and experience, with social elements layered on top in a manner that permit social interactions (gifts, help requests, new player recruitment) to assist in progressing along a particular in-game achievement metric.

In other words, social interaction in social games is utilitarian. Granted, established communication theories like uses & gratifications would argue that socializing in MMOs is also about need satisfaction. However, MMO socializing has a tremendous range of investments and complexity, ranging from raid groups to goofy dancing emotes. That is, MMO designs provide room for social interaction that is expressly accomplishment-oriented as well as that which serves no achievement purpose. Social games, on the other hand, typically do not. Rather, in-game friends are harnessed like available assets from one’s pre-existing network, including people one may have not spoken to in years (and still does not speak to within the process of reciprocal in-game gift-giving).

Such a comparison is not meant to be a vertical one – rather, the point is that we’re dealing with apples and oranges. The virtual spaces on which Bartle’s types are based offer different experiences for different types of users in their designs. In contrast, the majority of social game designs are mostly “diamond” game mechanics that leverage “heart” resources from the player’s social network, conflating achievement and socializing. As such, Bartle’s types do not neatly apply.

The point is that the aptness of a given player typology is based upon a given game’s design. In the case of social games, Bartle types do not fit. However, other typologies more in tune with the particular mechanics and exchanges of these games – say, “competitors” vs. “decorators” – may make for more accurate – and in turn, more useful – hooks on which to hang new design elements.

Yet, regardless of the aptness of a typology for a game design’s player base, the fact remains that such labels are ultimately just categorical measurements of behavior. Like the Facebook personality tests, they are descriptive in nature, merely offering taxonomies of what already exist in the world. However, many developers, particularly those of social games who rely on micro-transactions rather than one-shot purchases or extended subscriptions for revenue, are likely interested in more than demographic data. Beyond describing player behavior, such developers are concerned with predicting it. For example, as noted above, social game players leverage network friends as social capital to put towards achievement, assets which are exchanged with and valued in terms of time and real world money. Social game designers are interested in how much of each of these currencies a given player is willing to expend and how they may fine tune a design to reach the desired proportions. Such insights cannot be extracted by typologies alone, and instead require telemetering, which offers quantitative assessment along continuous variables of interest.

Therefore, to answer the title question above, we might conclude that, typologies can be useful when trying to categorize and label your players and to understand how a new design may or may not afford particular player experiences. That is, typologies are good for describing user meaning in a game. In contrast, if a developer is less interested in creating experiences and more concerned with predicting and incenting particular patterns of behavior, designs might be better informed through telemetry-based player profiling, similar to the data tracking underlying non-game choice environments like Amazon and Netflix.

...But where's the fun in that?

To Bartle or Not to Bartle?: Capturing Player Behavior in Social Games by Jim Cummings, unless otherwise expressly stated, is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

4 Responses to “To Bartle or Not to Bartle?: Capturing Player Behavior in Social Games”

  1. Nice post. I've been thinking about similar issues.

    I stayed with your thread until 2nd to last paragraph, then I got a bit confused. Agree that Bartle's typology has its roots in a particular task context and that additional or alternative typologies could and probably should be brought to bear when considering behaviour in new contexts.

    Not sure however about where you are going with the point about telemetry. Is the main issue whether the measure is a continuous variable, rather than categorical, (and therefore the personalisation can in theory also be delivered in a continuously adjusted or at least highly granular way)? Or is it to do with the fact the classification and consequent personalisation naturally falls out of user behaviour within the system, rather than an out of context assessement? Or is it both? Or, indeed, none of the above...?

  2. Jim Cummings says:

    Thanks for the thoughtful response, Heather.

    To answer your question: I think there are a couple issues. :)

    The first is that any player typologies should be contingent upon the design. Bartle types are often employed as heuristic short-hands for describing players within gaming context that have little in common with MUDs and MMOs. Social game players should be described within the context of their environments, which may or may not fit under a single “social game” typology. That is, not only do the players of Farmville and Gardens of Time require a different typology than that used for MUD/MMO players, but their designs and players may be distinct enough from one another to merit separate typologies from one another.

    The second issue concerns the intention behind measuring player behavior. I’d argue that typologies are useful if one wants to qualitatively organize patterns of player behavior within an environment, and perhaps use that information to understand different “types” of players and their goals. That is, typologies, as qualitative descriptions, get at the meaning of the player experience. However, if one is more interested in quantifying player behaviors and using them to predict the likelihood of future actions (as is the case for many social game developers), telemetry may be more useful. That is, the more granular and less subjective measures of telemetry permit likelihood estimates of codified future behaviors.

    In other words, typologies = descriptive, related to social meaning; telemetry = predictive, related to objective measures.

    Of course, these two tools for capturing player behavior can be used together. Take for instance this recent study: http://www.engadget.com/2011/06/14/shocker-gamer-behavior-is-actually-quite-predictable/ . The researchers have defined what they call “cliques” or groups of WoW achievements that cluster in a statistically significant way. Though I haven’t read the full manuscript, it sounds as if this study used telemetry to produce a taxonomy of labels. Note, however, that not only are these labels used to describe types of behaviors rather than types of players, but also that this taxonomy is currently bound to the design of a single game. Were the researchers to try to implement their approach to a game widely different than WoW, they’d likely get a different set of types.

  3. Thanks for clarifying & for the engadget link, which looks really interesting. I think I was confused a bit as you are telemetry in a rather different sense from the sense I am used to - for me it just means remote measurement....

  4. p.s. you might be interested in Radoff's new book "Game On" - I just ran into it myself and wish I had time right now to read more than the toc....;-)

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>