Social Machines
Do you believe in Burkina Faso?

The State of the Art: Audience Models

There is surprisingly little information available about the way people play games, and fewer still audience models. In this field, our DGD1 model (which is talked about extensively in our forthcoming book, 21st Century Game Design, which I am contractually obligated to plug) is practically the only game in town.

True, there is the Bartle type system, but this is a model that applies only to online play. There is Nicole Lazarro's four emotional keys - but this is a model of means of generating an emotional response in a player without use of narrative. I must say, I found Nicole's work hugely satisfying to read - I wish I had a link I could provide for it, and it was certainly influential when we were interpreting the clusters that came out of the DGD1 research. I wish we had more researchers doing this kind of work.

I'm going to talk about how the DGD1 came about, specifically with the goal of explaining where we are now, where we'd like to be, and why we can't get there yet.

The original motivation for the DGD1 came out of an earlier audience model we at International Hobo had cobbled together from very scratchy survey data. We wanted to see what patterns we could find between the axis of the Myers-Briggs inventory and the way people played games, as our informal observation suggested certain hypotheses (I won't go into detail - it's all in the book). There were some issues... firstly, the tendencies that correspond to each axis in Myers-Briggs are just that - tendencies - and all individuals express all these tendencies under certain situations. The notion of a 'type' in Myers-Briggs terms refers to a preferred pattern, and as such, most conventional tests are inadequate to measure this.

This didn't greatly bother me when I was setting up the original survey. I figured, if we got enough results, we'd be able to look at statistical correlations, as individual deviations would smooth themselves out.

What we were looking for were clusters of related behaviours - specifically, was there a way of dividing the results we got back such that they appeared to show patterns of related behaviour. The best clustering result we found was when we used a high score in certain Myers-Briggs axes to group the results. This lead us to clusters which predominantly showed similar patterns of game playing styles - and so we followed up with a set of case studies, which showed the patterns were robust, and statistically linked to Myers-Briggs types, albeit imperfectly.

Briefly, here are the four play styles we found:

  • Type 1 Conqueror play style is associated with challenge and the emotional payoff of Fiero - triumph over adversity. This correlates with what Nicole Lazarro has called "Hard fun". We associate Type 1 play with players who aim to utterly defeat games they play - they finish games they start.
  • Type 2 Manager play style is associated with mastery and systems. Victory for people preferring this play style seems to be the sign that they have acquired the necessary skills, not a goal in and of itself. They may not finish many games that they start playing.
  • Type 3 Wanderer play style is associated with experience and identity. This correlates somewhat with what Nicole Lazarro has called "Easy fun". Challenge is not especially desired, but may be tolerated - what they enjoy is unique and interesting experiences. Stories and mimicry are key draws.
  • Type 4 Participant play style is associated with emotions and involvement. It connects with what Nicole Lazarro calls "The People Factor". Participants seem happiest when they are playing with people, but they also enjoy play which is rooted in emotion. Any game which allows the player an emotional stake is a potential Type 4 game.

Unexpectedly, we found these patterns spread across the Hardcore and the Casual market segments - that is, those players who buy and play many games versus those who buy and play few games. This, we had not anticipated. The Hardcore clusters were universally more Introverted and Intuitive (in Myers-Briggs terms), while the Casual clusters were generally more biased towards Sensing and (to some relative extent) Extroversion.

The two elements combine to provide eight clusters - H1, H2, H3, H4 and C1, C2, C3 and C4.

The case studies demonstrated that the play styles were a robust model, but that Myers-Briggs type alone could not be used to assign a play style to an individual (not that surprising, really).

Furthermore, we did a follow up study in which we did a survey of people who belonged to online communities that identified with a specific type in Myers-Briggs typology. We found exactly what we expected - an approximately Gaussian distribution with the peak in each case centred upon the expected play style. However, this data was laughably incomplete, as not every type had an online community we could survey - and some of the communities refused to allow us to access their members, seeing what we were doing as attempting to sell to them, rather than attempting to understand their play needs.

I was hoping we would go forward and get to work on a DGD2 model - and indeed, our work looking at Temperament Theory and the related skillsets (Logistical, Tactical, Strategic and Diplomatic) suggests a possible way forward, as this connects in an interesting fashion with our current model. But one key ingredient is currently missing: feedback. I feel it inappropriate to begin a new survey until we've had useful peer review and feedback on the work we've done so far, and thus far, this hasn't really happened. We've received a lot of praise and support for what we've done (I think largely because no-one else has done something similar yet), but no feedback that would help guide and inform a next step. I'm hoping that this will come, and that in particular getting the book out to a wider audience will help.

In the long term, I'd hate to think that the DGD1 would still be the only audience model in town, say in twenty years time. That would be very disappointing. What I'd like, personally, is for other people to take forward the motivations that lead to this model and come back with better models. Nothing would please me more than, like Sigmund Freud, to be proved utterly wrong, but in doing so to inspire other people to produce something better along the same lines.

I believe we need audience models if the games industry is going to grow, and I sincerely hope that we inspire somebody to go out and prove us wrong, and come up with an audience model which blows ours out of the water. If that doesn't happen, I guess we'll be forced to do it ourselves.

Comments

Feed You can follow this conversation by subscribing to the comment feed for this post.

Hello Chris.

Keep up the good work. Been cooking something for a while now (what is it? 2 years since our last rant).

We shall rant again soon... just give me a few more weeks... I think you'll like it ;-)

DB

Chris, I am currently writing my final year dissertation. My working title is the "Demographic, Experience & motivation of gamers" You mention in the article above about the limitations of DGD1 but I can find no reference of them in 21st Century Game Design. Would it be possible to contact me with some more information?

Many Thanks Karl Reader

Karl: the limitations of DGD1 are numerous. Here are a few of them:

1. Paper tests for Myers-Briggs type (which was at the centre of this research) are notoriously inaccurate - they just about work at a statistical level, but are a severely inaccurate instrument in general terms.
2. The survey data aimed to be orthogonal by collecting from different website sources (including at least one non-gaming site), but it still seems to me that we did not get a perfectly representational audience cross section i.e. there were still (unidentified) biases in the source data. (Is this always a problem with this kind of research? Perhaps I am too harsh on our own study). For instance, the skew towards Wanderer in the final data suggests a sampling error - either because of the source of our data or because of flaws in the paper test.
3. Much of the model was constructed from the survey data, but the case studies were the more valuable part of the study. It would have been far superior to construct a large case study exercise and skip the survey element if we'd had the resources to do so.
4. The hypothesis affected the focus of the research. Since we were testing for a hypothetical correlation between a specific MBTI pattern and 'hardcore' audience status, this affected the nature of the research. I believe it narrowed our focus somewhat. (This is something that is always going to be a problem with science, though).
5. The clusters derived were effectively drawn onto the data - that is, we derived the cluster conditions, then derived the patterns associated with those clusters. This is probably fair game in a subjective statistical science such as cluster analysis, but still, it is a limitation that the four play styles we describe are slightly arbitrary - because we could have drawn our clusters in a nearly infinite set of alternative combinations. I feel this really lets DGD1 down, and I'm keen to improve upon this in the next research (although it must be said that all clustering solutions face this criticism!)

Bottom line: DGD1 is a brave attempt at a solid audience model, but it is far from definitive. In fact, a difinitive audience model may be difficult or impossible to derive - and it may not even be helpful. Perhaps it is more useful to have many different audience models so that we can learn about the audience in many different ways. If this is the case, then the DGD1 at least establishes the concept of a researched audience model as a valuable addition to the game design process.

If you need more than this, please get in touch with me and I will be happy to answer any specific queries you have. Use the About link from this blog to find my company website, use that contact address - anything you send addressed to me will get forwarded.

Take care!

Hello,

As you stated, the main problem with the MBTI is the low fidelity (and validity, as some research show).

I think of it as an ''heuristical'' personallity discription tool.

Have you taught about using a test like the Big-5 or NeoP-R ? alot of research are have used these in the field of psychology.

Hi Christopher,

I did ponder switching over to Big-5, but after studying it for a while I concluded there wasn't much of an advantage to switching to that over a post-Myers-Briggs model, such as Temperament Theory. I'm not really convinced that the quality of any of these models is so much better than any of the others right now.

I like the idea of seeing these models as heuristic - but I confess, Big-5 seems just as heuristic as MBTI to me, and in terms of providing terms that allow us to talk about personality traits I find Temperament Theory to be easiest to apply (although not without issues).

I think I'm close to converting Myers-Briggs/Temperament Theory into neurobiological terms, though (barring a few remaining issues) which gives an interesting new perspective on the issue.

Thanks for your comment!

If you want to explore the motivations of players then I suggest you could put your findings in parrallele to other research that have been completed in comparable fields, such MMO environnements.

Nick yee has been a pionner in developping motivational models in the MMORPG industry. check him out : http://www.nickyee.com/

I hope the reading will be inspiring,

Keep up your good work.

regards,

Christopher.

Hi!

This post was written before Nick's Daedalus Project was completed, but rest assured I've read all of Nick's papers, and indeed, he and I were in a tutorial at GDC together.

Despite the title of this one, it's far from "the state of the art" any more, since the post is three years old. :)

Thanks for commenting!

Verify your Comment

Previewing your Comment

This is only a preview. Your comment has not yet been posted.

Working...
Your comment could not be posted. Error type:
Your comment has been posted. Post another comment

The letters and numbers you entered did not match the image. Please try again.

As a final step before posting your comment, enter the letters and numbers you see in the image below. This prevents automated programs from posting comments.

Having trouble reading this image? View an alternate.

Working...

Post a comment

Your Information

(Name is required. Email address will not be displayed with the comment.)