Who plays, how much, and why? A behavioral player census of a virtual world



Download 111.19 Kb.
Page1/3
Date conversion25.04.2018
Size111.19 Kb.
  1   2   3


Who plays, how much, and why? A behavioral player census of a virtual world

Dmitri Williams

University of Southern California
Nick Yee

Stanford University


Scott Caplan

University of Delaware



Contact information:

Dmitri Williams

Annenberg School for Communication

University of Southern California

3502 Watt Way, ASC 121E

Los Angeles, CA 90089

dmitri.williams@usc.edu

Who plays, how much, and why? Debunking the stereotypical gamer profile

Abstract

Online games have exploded in popularity, but for many researchers access to players has been difficult. The study reported here is the first to collect a combination of survey and behavioral data with the cooperation of a major virtual world operator. In the current study, 7,000 players of the massively multiplayer online game (MMO) EverQuest 2 were surveyed about their offline characteristics, their motivations and their physical and mental health. These self-report data were then combined with data on participants’ actual in-game play behaviors, as collected by the game operator. Most of the results defy common stereotypes in surprising and interesting ways and have implications for communication theory and for future investigations of games.

Who plays, how much, and why? A player census of a virtual world

Americans, both young and old, play video games at an ever-growing rate. Today’s adults, unlike those who dropped game play in the 1980s due to public shame (Williams, 2006a), play more than previous generations. 40% of adults are now regular players, compared to 83% of teenagers. The average player age is now 33, propelling the industry to $7.4 billion in U.S. sales in 2006 (Entertainment Software Association, 2007). The result is a population engaging in this medium as an acceptable mainstream activity. And as the Internet has become a larger part of everyday life (Wellman & Haythornthwaite, 2002), so too have networked games. 67% of teens now regularly play some game online (Rideout, Roberts, & Foehr, 2005). Online games bring people together as they use their cell phones, their computers and their gaming consoles to access not only games, but other people. The most social and high-profile of these spaces are persistent virtual worlds—games that are always on, in which players maintain a regular character who grows and changes, and in which many players participate in long-term social groups (Griffiths, Davies, & Chappell, 2003; Yee, 2006). These worlds, called “massively multiplayer online games,” or MMOs, are vibrant sites of community (Steinkuehler & Williams, 2006). Yet even as participation in them rises to several million players in North America alone, research on their players and their impact has remained relatively scant. Ethnographic and experimental investigations of player “guilds” and communities have explored the social dynamics and relationships that exist (Taylor, 2003, 2006; Williams, Caplan, & Xiong, 2007), but systematic and generalizable research has remained elusive, largely due to the difficulties of securing access to players within the walled gardens of for-profit companies. The current investigation represents the first case of access to proprietary in-game player data. With the active cooperation of a major game operator, the current study surveyed players and unobtrusively collected in-game data on their behaviors. This combination of demographic, attitudinal and behavioral data generated a true player census of a virtual world. The paper presented here explicates players’ demographics, their playing patterns as related to those demographics, and their motivations for play. Many of the results defy both stereotype and theoretical predictions.

Prior research on virtual worlds

As with any new medium (Wartella & Reeves, 1983, 1985), questions about the impact of video games have fallen into three categories that occur in order (Williams, 2003): What does it replace? What are the health impacts? And, what are the social impacts? While the third question is beyond the scope of the present study, public concern over displacement and health issues continues to be a policy debate and an area of interest for communication researchers. The most basic concern is that consumers of the new medium will use it to the detriment of human relationships. As Putnam has theorized with the rise of television (Putnam, 2000), use of the new technology may come at the expense of personal relationships and community involvement. As use of the isolating technology rises, Putnam argues, users spend less time with their friends and family. Similarly, some have speculated that since games are sedentary activities, as television and prior media have been, then they might be present a risk to health as well (Rideout et al., 2005).

Early reports on video games featured several claims of direct physical health impacts, including “Pac Man elbow” (Skow, 1982), “video wrist” ("Donkey Kong Goes to Harvard," 1983), and “Nintendinitis” (Adler, Rogers, Brailsford, Gordon, & Quade, 1989). Mental health concerns have also made appearances in both cultural outlets and in research. Movies such as Nightmares (Sargent, 1983) featured characters, usually young males, drawn inexorably to the devilish glow of arcade screens. Subsequent research has explored gamers’ mental health (Fisher, 1994; Salguero & Moran, 2002), especially within virtual worlds (Plusquellec, 2000). With such portrayals and concerns, it is not surprising that there are many popular stereotypes of video game players. As portrayed in print media, game players are stereotypically male and young, pale from too much time spent indoors, and socially inept (Williams, 2003). As a new generation of isolated and lonely “couch potatoes,” young male game players are far from aspirational figures.

Although stereotypes are common, there is a scarcity of systematic research on game players. One question for scholarly researchers to answer is whether the popular stereotypes of video game players are accurate. The initial research on these populations suggests that players are not all isolated teenage males. While not able to draw on random or stratified samples, two self-reported survey projects of MMO players (Griffiths, Davies, & Chappell, 2004; Yee, 2006) found evidence that the players are older than was previously thought and likely more social than stereotype suggests. Additionally, a range of ethnographic and interview projects have discovered family-like dynamics among active game groups (Taylor, 2003, 2006), active ecologies of players (Castronova, 2005), and complex hierarchies of players in large guilds (Williams et al., 2006). Experimental investigations have explored communication differences and interpersonal variations due to virtual distance or representations (Pena & Hancock, 2006; Williams et al., 2007; Yee & Bailenson, 2007) or type of game played (Smyth, 2007). Yet in all of these cases the work has been limited by access to game populations, with the researchers relying on convenience samples or subject pools. Game companies have been largely closed to researchers due to time, legal, resource and focus constraints. An exception is Kafai et al’s (Kafai, Feldon, Fields, Giang, & Quintero, in press) collaboration with a game developer to test science learning. Kafai’s team was able to work closely with a game maker specializing in adolescents and science education. While studying that target population, the team was able to bring adolescents into social lab settings, and also to marry those data with unobtrusive behavioral measures from within the game world. Such access has never occurred for any of the mainstream game titles.

Instead, the common approach to studying games has been to have single-player lab sessions with self-reported behavioral measures. The dominant model for this research has been the General Aggression Model (GAM) (Anderson, 2004), which extends cognitive neoassociational approaches (Bandura, 1994; Berkowitz & Rogers, 1986) into a larger framework that incorporates behaviors and several routes to aggressive behaviors. This paradigm offers minimal depth when explaining how effects occur in social spaces, or how player motivations might moderate the effects process. Thus, mapping out the social and motivational factors of play within these dynamic spaces is a key necessary condition for theoretical progress, begun by Sherry and colleagues (Sherry, Greenberg, Lucas, & Lachlan, 2006) and continued here.

In the study of virtual worlds, one well-known player taxonomy is Bartle’s four Types (Bartle, 1996)—Achievers, Socializers, Explorers, and Killers. While the model is widely referred to, it was developed without statistical data. Thus, the motivations suggested by the Types may not factor out as proposed. For example, Bartle suggested that Explorers enjoy discovering the geographical boundaries of the world as well as analyzing the numbers and rules underlying the game mechanics, but these two motivations may not be highly correlated.

Nevertheless, Bartle’s taxonomy does provide a starting point for exploring player motivations in online games. In an attempt to create a taxonomy of player motivations based on empirical data, Yee (2006) generated a set of motivation items based on qualitative responses from earlier open-ended surveys eliciting motivations for play from online gamers (N. Yee, 2005). These were then added to the motivation items suggested in Bartle’s typology. Five factors emerged from the analysis—Achievement, Relationship, Manipulation, Immersion, and Escapism. The first three factors mapped on to three of Bartle’s Types while the motivations of Immersion and Escapism were not previously noted in Bartle’s typology. More importantly, Bartle’s suggested Explorer type did not coalesce as a factor; geographical exploration was not correlated with an interest in analyzing game mechanics.

While this factor-analytic framework was based on statistical analysis, some motivations that players have articulated in the mentioned works did not emerge as factors. For example, geographical exploration didn’t load on any of the five factors. To this end, Yee (2007) used additional qualitative data from open-ended responses to expand on the existing inventory of motivation items. This expanded inventory yielded 10 factors. The large number of factors hinted at a higher-order factor structure. Indeed, a second-order factor analysis yielded three factors. In other words, these three second-order factors summarized the 10 first-order factors.

The three second-order factor structures were: Achievement, Social, and Immersion. Under the Achievement structure were the factors related to: 1) advancement, 2) analyzing game mechanics, and 3) competition. Under the Social structure were the factors related to: 1) chatting and casual interactions, 2) developing supportive relationships, and 3) teamwork. And finally, under the Immersion structure were the factors related to: 1) geographical exploration, 2) role-playing, 3) avatar customization, and 4) escapism.

The 10 factor model covered all of the motivations listed in Bartle’s typology. In addition, the factor analysis showed that Bartle’s Explorer Type was composed of two uncorrelated motivations—geographical exploration and analyzing game mechanics. The 10 factor model also covered five out of the six motivations in Sherry’s (2006) typology of motivations for video gamers in general (i.e., across genres). The remaining motivation—Arousal, or the constant feeling of an adrenaline rush—is seldom seen in qualitative responses from online gamers because the genre, based on long-term goals and progression over the span of weeks and months, doesn’t lend itself to constant arousal or instant gratification. This is an area in which first-person shooter players’ motivations likely differ from MMO players’(Jansz & Tanis, 2007). The second-order factor structures also show how the underlying motivations can be grouped together. Indeed, it begins to become unwieldy to have 10 entirely independent factors in a model. The second-order structures provide a more parsimonious representation of the disparate factors. These second-order structures also allow us to cover the range of known motivations with fewer factors in an analysis.

Without access to game server data, no study has yet been able to draw a representative or stratified sample from any game to study motivations or basic play patterns and their offline demographic correlates. What is more, measurements of in-game activities have been limited to either self-reports of behaviors or observable ones, rather than measures of actual in-game data obtained by servers. This has been particularly vexing given that the actions and behaviors within game spaces are all capturable in a way that is atypical in social science research. A truly unobtrusive measure of behaviors exists, untapped, on the electronic servers and databases of game companies. These data are often captured at millisecond resolution and contain thousands of potential variables at a level of specificity never imagined by early proponents of unobtrusive measurement (e.g. Webb, Campbell, Schwartz, & Sechrest, 1966). Even limited studies of unobtrusive game data have shown great promise (Ducheneaut, Yee, Nickell, & Moore, 2006). The findings reported here are the first such use of data supplied by a game operator, and the first combination of them with standard survey measures. The result is a unique combination of demographic, attitudinal and behavioral measures.

Research Questions

Despite the stereotypical portrayals, the three waves of concerns identified by Wartella and Reeves (1983, 1985) have become a common way to fear—and to study—new media. Nevertheless, this does not also mean that those concerns are spurious. It is worth examining the demographic profiles and health outcomes of game players simply to add empiricism to an area of study often weighed down by preconceptions and popular stereotypes. Policy makers and parents are also concerned with game play and should not have to rely on industry trade organizations for data.

Thus, the overall goal of the current study was to conduct a proper census of MMO game players:

RQ1: What are the means and distributions of age, gender, race and class among MMO players?

Such a census cannot, of course, exist in a vacuum. It must be paired with national-level data on the same categories so that the MMO population can be compared to the general population. Therefore:



RQ2: How do the demographics of MMO players compare to the general population?

Next, it is important to gain not just a measure of who these people are, but to also discover what they do. Basic play patterns of time can be examined, and, combined with the demographic data, the following question can be asked:



RQ3: Who plays how much?

In turn, play patterns can be examined for their displacement implications. The most obvious example is in media use. If players add a significant amount of playing time, this time must come at the expense of something else. Prior work (Williams, 2006b) suggests that other entertainment or communications media will be displaced or altered. So:


RQ4: What are the media use patterns of MMO players?

In keeping with the questions about the health impacts, it is worth learning about both the physical and mental health of the playing population in comparison to the general population. Players may be the sedentary at-risk populations feared (Rideout et al., 2005), or may exhibit unhealthy levels of mental health. Although no causal inferences can be drawn, it would be important to future research to explore the health of players. Such outcomes, combined with demographics and play patterns, can lay the groundwork for causal studies of the impact of games (positive or negative) on physical health and mental state. Therefore:



RQ5: How does the physical health of MMO players compare with the general population? And

RQ6: How does the mental health of MMO players compare with the general population?

For researchers, establishing baselines for populations offers the obvious practical value of establishing frames and starting points for future work, but a census of behaviors and motivations is equally important for theoretical reasons. Practically speaking, having a framework for discussing and measuring motivations for play among online gamers extends the tools of uses and gratifications theory for online gamers, and provide us with a means to better differentiate users beyond demographic information alone. Furthermore, such a framework provides the foundation to explore whether different subgroups are motivated differently, and whether certain motivations are more highly correlated with usage patterns or in-game preferences or behaviors. It is important to recognize that different users have always had different expectations and uses of the same media (Blumler & Katz, 1974), and that new media do not change this fact (Sherry et al., 2006). Game players make not only the choice of title and genre, but direct the action and make choices throughout the play experience. So while it is still fair to ask what the impact of such experiences are, it is no longer sufficient to stop there. We must also ask why the users do what they do so that we can understand the role their choice and actions might play as moderating variables in any effects model. Understanding the motivations behind video game-play therefore provides us with good predictors of gamers’ usage and genre preferences (Sherry et al., 2006), consistent with a perspective driven by what Ruggiero (2000) calls the “uses and effects” paradigm. But while Yee’s (2005) original framework was validated with a large sample of online gamers, respondents to the survey were self-selected, and thus the framework would benefit from a replication with a known and stratified sample. Incorporating this framework of motivations in the current study also allows us to understand how player motivations intersect with other variables of interest. Thus:


RQ7: What are the motivations of MMO players?

Lastly, in keeping with the user-centric investigation of motivation, one key player choice is made when a player chooses a “home” server. To explain, most MMOs do not exist as one large, common space. For reasons of virtual geography, density and computational capacity, players are spread out across multiple parallel versions of most games. These versions sometimes come in slightly different flavors, with minor differences in rules and affordances. The two common differences include the ability to attack other players (a “PvP” or player vs. player server), and a focus on role playing. The game studied here also featured a unique variation dubbed “Exchange” servers which allow using real US dollars to buy virtual items and characters. Thus, players entering the game world for the first time are presented with a choice of playground: normal, PvP, role play or Exchange. This self-selection can be studied by asking:



RQ8: Do players of different types systematically select different servers?

Method

The current study focused on the MMO EverQuest 2 (EQ2) because of its popularity, its representativeness of mainstream MMOs and because of the unique access provided by the game operator. The sequel to the highly successful EverQuest, EQ2 launched in November of 2004. Both it and its predecessor maintained a large share of the North American market until the launch of World of Warcraft, the current US MMO leader. Despite losing its market lead, the EverQuest franchise continues to expand and still attracts several hundred thousand players (Schiesel, 2007). For generalizability purposes, EQ2 represents the mainstay of the MMO market, fantasy role playing. Its basic game rules and goals are nearly identical to World of Warcraft’s and the other several fantasy titles on the market, which altogether comprise 85% of the total MMOs as played.1 Because MMO operators do not release data on their players, there is no way of knowing whether the results here are or are not indicative of other game populations. Indeed, the findings released here are the first public data to be shared by a major game company. The game operator, Sony Online Entertainment, agreed to cooperate with the research team, and to provide access to data from the game’s large back-end databases. Sony further worked with the research team to help field the large survey described below. This level of access and cooperation between a game developer and an academic research team is the first of its kind. It enabled a stratified sample rather than a convenience sample, helped establish trust with the potential survey takers, and most importantly, it allowed the linkage of survey data with unobtrusively collected game-based behavioral data.

Sampling and Procedure

Survey sampling in the world of MMOs requires focusing on the player as the unit of analysis, but being aware of the fact that players can maintain more than one character. These characters may reside on multiple servers. However, Williams et al (2006) found that most players play one character most actively and consider it their “main.” Thus, the first step in establishing the sampling frame was to examine the game databases to determine which of the player’s characters was played the most frequently over the prior year. This “main” character became the sampling unit in the study, and established the equivalent of residency for the player—their main character determined which server they belonged to. These characters then populated the sampling frame, and were listed evenly across the four servers used in the study.

Players with a character within the frame were then invited to participate in the survey if they logged in during the survey window. There were no special efforts made to hide the nature of the survey as it was what it was labeled to be: a general study of who plays EverQuest 2. If they agreed to participate, players were directed to a secure web page. After giving informed consent, the players completed the survey, which took about 25 minutes. Players were not offered a cash or prize incentive for their participation. Instead, they were promised a special virtual item that would be added to their in-game inventory as compensation. This item, the “Greatstaff of the Sun Serpent” was created by Sony for this unique use. It was considered desirable for players of all levels because of its rarity and its potency in combat and proved to be a valuable recruiting tool for the survey. Based on prior survey work by the research team in this area using cash incentives or no incentive (Williams, 2006b; Yee, 2006), the prediction was that the survey would take one to two weeks to fill to a cap of 7,000 respondents, but the total was reached in just over two days.

Participants were identified by an unseen account number in Sony’s databases. These account numbers were used to unobtrusively link the survey data with the game’s databases. The main measure derived from Sony’s databases was total playing time, calculated by averaging the number of cumulative seconds of play over the playing window on each account. These data were then used to compute the players’ average time in the game per week, and were matched back to their entries from the survey. Because players could maintain multiple characters on their account, each character’s playing time was collapsed into one meta-level play value for the player. These calculations occurred within the database and separate from the survey, so subjects were unaware that their time online, their server residence, or their character use were part of the research.



Measures

To answer the research questions, the survey instrument used a variety of standard demographic measures. Players were asked for their age, gender, race, household income, education, and religion. For general demographics, comparative data were derived from the 2000 U.S. census. General population religion was drawn from Kosmin, Mayer, and Keysar (2001). With the large dataset in use here, nearly all tests are significantly significant. Thus, t-tests are reported with Cohen’s d to allow for substantive interpretation. As a rough rule of thumb for Cohen’s d, .20 is considered a small effect, .50 moderate and .80 large.



  1   2   3


The database is protected by copyright ©hestories.info 2017
send message

    Main page