Gender Swapping and User Behaviors in Online Social Games

Jing-Kai Lou, Kunwoo Park, Meeyoung Cha, Juyong Park, Chin-Laung Lei, and Kuan-Ta Chen

PDF Version | Contact Us


Modern Massively Multiplayer Online Role-Playing Games (MMORPGs) provide lifelike virtual environments in which players can conduct a variety of activities including combat, trade, and chat with other players. While the game world and the available actions therein are inspired by their offline counterparts, the games' popularity and dedicated fan base are testaments to the allure of novel social interactions granted to people by allowing them an alternative life as a new character and persona. In this paper we investigate the phenomenon of "gender swapping," which refers to players choosing avatars of genders opposite to their natural ones. We report the behavioral patterns observed in players of Fairyland Online, a globally serviced MMORPG, during social interactions when playing as in-game avatars of their own real gender or gender-swapped. We also discuss the effect of gender role and self-image in virtual social situations and the potential of our study for improving MMORPG quality and detecting online identity frauds.

1  Introduction

Massively Multiplayer Online Role-Playing Games, or MMORPGs for short, are the most popular form of serious online gaming in the Internet era. The illustrious developments in modern computing and communication technology have allowed millions of people to dive into these games and enjoy their so-called "second life" inside a fantastical world. The most alluring aspect of MMORPGs is that one can escape the dull, stressful reality and roam freely while fulfilling their fantasies of an alternate identity. Interestingly it is essential that the world in an MMORPG is still very realistic, because it is the strangeness and novelty married with a sense of familiar realism that makes these games popular and enjoyable to millions of gamers worldwide [26]. In MMORPGs, as in real life, the fun of real-time interactions with other players online-via conversation, commerce (trade), cooperation, and battle, as in real life (some literal, others figurative)-are said to be the most attractive feature.
From the research point of view, the realistic nature of the interactions opens up the possibility of viewing the world inside an MMORPG as a laboratory for observing individual and social behaviors of humans in as fine details as the game design would allow. In general, the proliferation of online arenas where people socialize, including MMORPGs, has given rise to formidable changes for user behavior studies like psychology and sociology, where until recently nearly all research on human behaviors have had to rely on one-time self-reported data that were not easily replicable and were relatively small in scale [18]. Large-scale detailed human dynamics data found in online digital environments such as MMORPGs, therefore, are fundamentally transforming the fields by allowing researchers to easily overcome traditional constraints and conduct studies on an unprecedented level of scale and precision.
MMORPGs are enticing to user behavior research as they provide, possibly more than any other online arena, a lifelike hyper-realistic environment, where people can perform various social actions. For this reason MMORPG data are becoming increasingly popular for examining the detailed patterns of people's social behaviors; notable recent studies include an analysis of the structural equivalence of the interaction networks in Pardus [27], and a comprehensive surgery of correlations among distinct kinds of interactions (for example, friendship, party invitation, and trade) between users of AION [26], both well-known global MMORPGs [4,1].
The aforementioned works on MMORPG user data focus on the global topology of the social networks constructed from players' interaction logs. In such a network study, the complete set of interactions is represented as a giant mathematical graph, in which the players are the nodes and the interactions are the links connecting the nodes. Naturally the players, the subjects of observation, are characterized based on graph-theoretical quantities such as their degree and the number of interactions they participate in.
While such Social Network Analysis (SNA) has a rich history in sociology and has recently found much success in various subfields of science and engineering [32,22,6], it often leaves out another essential feature of the nodes (humans): the individual characteristics of the subjects separate from network data per se. The demographics of the players including gender, age, and geographical location are a good example, from which one could examine the potential relationship between the players' social identity or status and their interaction patterns [24]. With the additional independent information, we can expect the user behavior analysis to be enriched with novel, intriguing findings should the social behavior patterns be correlated with the node characteristics [25].
Figure 1: A partial snapshot of the Chat network in Fairyland Online. Two genders-one for the in-game avatar and another for the player-are indicated: Blue (yellow) indicates a male (female) avatar, whereas the square (circle) indicates a male (female) player.
Of the many possible player identity parameters, in this paper we focus on the players' genders and their impact on social interaction behaviors found in Fairyland Online, a globally serviced MMORPG. Gender and its role in life are undoubtedly one of the most celebrated issues in history, sociology, and biological sciences, as it is widely accepted that gender is one of the most salient features that people use to categorize and process social stimuli [7,33,20,19]. A virtual environment like MMORPG, tantalizingly, adds an intriguing twist to this phenomenon: Players are free to choose in-game avatars of either gender, so that "gender swapping" occurs when a player chooses to play as an avatar of the opposite gender. Therefore we investigate the behavioral implications of the four distinct player-avatar combinations: two "straight" (male player-male avatar, female player-female avatar) and two "gender swapped" (male player-female avatar, female player-male avatar).
Figure 1 serves as the motivation for our present work. It is a local snapshot of the Chat network in Fairyland Online. There, each node represents an avatar, while its color and shape indicate the avatar's gender (blue: male, yellow: female) and the player's gender (square: male, circle: female) who controls it. As players are free to choose their avatar's gender, gender swapping may give rise to interesting interaction patterns and advantages in social treatments across different types of friendship links: straight and straight, gender-swapped and gender-swapped, and straight and gender-swapped. This paper investigates the demographic and practical implications of the phenomenon.
Our paper is organized as follows. First, we review previous studies on the issues of gender, including gender swapping, in MMORPGs. It has been found that male and female players have differing goals and motivations for playing online games, with male players being more achievement-oriented and female players being more social relationship-oriented [8]. Despite such primary differences in motivation, the picture becomes much more textured and interesting when males play as female avatars and females play as male avatars. While one's sexual orientation is shown to be moderately correlated with gender swapping it is still common across all orientations, with the majority of gamers having experienced it at some point. A more practical motivation for gender swapping can also be interesting: the player swaps genders in the hopes of getting the social treatment that they think the opposite gender enjoys [36].
Second, we study the patterns of user interaction behaviors in conjunction with the offline demographics and the online gender configurations. We first show that except for a rather prominent player base of male adolescents, Fairyland Online is enjoyed by males and females belonging to a wide age bracket. We also find that gender swapping is widespread, with over 40% of players owning at least one avatar of the opposite gender. An interesting question to ask is, then, whether the players' behavior or interaction patterns vary under different gender configurations. We pursue answers to this question in two directions, first by exploring their correlation with individual play styles, then by examining how they affect the dynamics of player-player interaction. We report the variations and differences in the speed of level-ups, and patterns in private conversation, friendship formation, and the trade of items according to the gender configurations.

2  Related Work

Early MMORPGs were played mostly by males [35,16]. Only later did the female player base gradually broaden as online games saw wider acceptance [15]. This led to research on gender-related issues in games, the major works of which we review here. There appear to be two broad lines of research: The first is on the gender differences, highlighting the variations in gaming motivations and behaviors of the male and the female gamers. The second is on the phenomenon of gender swapping, focusing on how and why players choose avatars of the opposite gender to their own.
Work on the Gender Differences in Gaming
Yee [36], based on a survey to understand the motivation for playing MMORPGs, reported that female players like to play games to form relationships, i.e. meet new people and socialize. Males, on the other hand, were found to value achievement and successes in missions and stronger avatars. According to Bartle [8], players can typically be categorized into distinct types, and male players often chose the role of achievers, and female players chose to be socializers. Various other studies found similar gender differences in motivation, playing style, and player type. Lo et al. showed that male players play longer sessions and achieve higher levels [21]. Cole et al. found that males are motivated by elements of curiosity, surprise, and discovery, while most females played games for their therapeutic function [13]. Most recently, Szell and Thurner showed how females and males differ both individually and social network-wise [28].
Issues of Gender Swapping
Players exist as avatars inside the game, and all activities are carried out as avatars. Players can freely choose their avatar's gender, appearance, and occupation. Gender swapping occurs when a male (female) player choose to play as a female (male) avatar. The phenomenon was first reported in Multi-User Dungeon (MUD) games [10], where researchers found widespread gender swapping. Griffith et al. [16] questioned Everquest (a well-known MMORPG) players, which shows that two third of the players had experienced gender swapping. Hussain and Griffiths showed a few years later that a majority of male and female online gamers have chosen at least one avatar of the opposite gender [37]. Huh and Williams's survey shows that homosexual players are likelier to gender swap than straight players, as they identify more readily with avatars of the opposite gender [17], suggesting that people consider their avatars as an expression of their identity.
Researchers have also questioned why people engage in gender swapping. According to [37], male users engage in gender swapping because they want to enjoy females' advantages in society; they expect that a female avatar to be treated more courteously by male avatars. Wang also showed that male players tend to show more prosocial behavior towards females [29]. Females, on the other hand, showed more pronounced behavioral changes when they swapped genders, by showing acting in a more masculine manner than the average male player [17]: while socializing as female avatars, females were shown to focus heavily on achievements in related activities (combat, hunting, etc.) as male avatars [8].
While these previous studies brought to light valuable insights into understanding gender differences and gender-swapping behaviors in online games, many of these show the potential limitations of survey-based research, namely sampling biases and the danger of self-selection biases [36,17]. Conducting surveys over a small number of people can identify a misleading tendency of the population. Even when surveys are conducted over thousands of subjects [36,17], there is the danger of the self-selection bias of the samples. Also, these studies have been unable to correlate various demographics with gender swapping, which may tell us more about the patterns and the underlying reasons for such behavior.

3  Fairyland Online

In this section, we describe key features of Fairyland Online and introduce the data set analyzed.

3.1  The Game Features

Fairyland Online (">"> is one of the longest-serviced MMORPGs in the world. The game's storyline revolves around several well-known fairy tales, and players engage in turn-based combats and random encounters. Developed by Lager Network Technologies, Inc. based in Taiwan, it was officially launched in February 2003 in the Taiwanese market. Attracting more than 200,000 subscribers in less than two months following the launch, the game expanded to Hong Kong, China, Thailand, and South Korea [3]. Figure 2 is a typical snapshot of the game, showing various gauges and avatars.
Figure 2: An in-game snapshot of Fairyland Online that sports a cartoonish world inspired by popular fairy tales.
Similar to most other MMORPGs [14], the game is hosted on multiple servers that each serves a separate realm or a world. Players join one of the realms and "live" there. Players play as avatars of their choice, chosen from three different races (human, elf, and dwarf), two genders (male and female), and several appearance profiles (outfit, hair style, etc.). Although one user account can possess multiple different avatars, at a given time a user can play with only one avatar.
Race and gender are two prominent avatar features. Race affects the strengths and weaknesses of the avatar: For instance, human is the most versatile race with no apparent strength or weakness. Elf, on the other hand, is outstanding at magic, but tires easily and cannot sustain their magical abilities for long. Dwarves are earthly and dwell in seclusion but possess much physical strength, which makes them suitable to be developed into warriors. Gender, on the other hand, has no effect on the strength or weakness of an avatar, and is merely cosmetic. Figure 3 shows some examples of the avatars in Fairyland Online. As shown in the figure, all three races have human-looking avatars where gender of an avatar is clearly distinguishable from its appearance.
Figure 3: Examples of avatars in Fairyland Online. There are three races (human, elf, and dwarf) and two genders (female and male).
The main attraction of an MMORPG is that players can interact with each other in a variety of ways, much like in real life. Fairyland Online provides a set of communication channels between players: Talk, Whisper, Family, Party, and Radio. Through a talk channel, players can broadcast their messages to everyone in their proximity in the game. Whisper is a private communication channel between any two players. Players can also set up a clan group or a party group and use the family or party channel to talk in groups. Clan represents a friendship social tie between users, while a party represents more goal-driven, short-term social ties (e.g., exploring dungeons or undertaking missions). Lastly, a Radio channel is an invitation-based chat room, where players can invite others to their radio for a chat.
Besides the communication channels, users can also trade or exchange items. Communication often precedes trading, indicating that people barter or bargain. Players can trade items, or sell and buy items to and from the game system itself with gold.

3.2  Data

Lager Network Technologies, Inc. gracefully provided us with the data from Fairyland Online. The data contain the complete log of player profiles and player actions between March and December of 2003. Chat contents were not provided, and the actual identities of users were anonymized for privacy concerns.
For this study, we limited ourselves to the users residing in Taiwan for demographic consistency. The data describes the following information on 4,512 players and 17,610 avatars:

4  Demographic Determinants of
Gender Swapping

Our analyses are motivated by the following questions: Which real-life factors lead a user to swap genders? Do real-life genders matter? Does one's age impact a player's choice of an avatar gender? To answer these questions, we first present the demographic characteristics of the players and show the correlation between certain demographic features and gender-swapping probability.

4.1  Demographic Characteristics of Players

4.1.1  Sociological Variables

The player profiles from Fairyland Online provide insight into several interesting socioeconomic characteristics of the game demographic. We were able to identify the following three types of information: gender, age, and geographical location.
First, in terms of gender, there exists a fair mix of male and female players in the game. Out of 4,512 players in the dataset, 1,819 (40.3%) were female. This is relatively higher than other MMORPGs, where fewer are female (16-32%) [17,5]. We believe this is due to the fact that Fairyland Online features a female-friendly cartoon-style game world based on fairy tales.
Second, in terms of age, a vast majority (86.5%) of players are aged between 10 and 30, with 5.4% under 10 and 7.9% over 30. A similar observation has been made in other MMORPGs, where adolescents and young adults are the main customers [15]. Figure 4 shows the age distribution according to gender. The male population shows a heavy concentration around age 9-13, covering the 2nd to 6th grades in Taiwan. Except for this range, we see no apparent gender difference (p>0.1). The average player age is 18.7, and the median is 17.
Figure 4: The age distribution of players. With the exception of a large male adolescent player base, the distributions are similar between males and females.
Third, in terms of geographical location, we find that players are more or less evenly spread across the urban and the rural areas in Taiwan. Based on the zip codes in player profiles, we mapped the users' locations onto 21 municipalities, including Taiwan's five major cities (Taipei, New Taipei, Keelung, Taichung, and Kaohsiung). Figure 5 shows the distribution of players across Taiwan. Regions are divided into rural or urban areas, where each region has a pie chart showing the male-female ratio in that region. The radius of each pie chart reflects the number of players of the corresponding region.
Figure 5: The geographical distribution of players. The circle size indicates the population of the corresponding region.
We see that while many players are from urban areas, quite a large fraction of players living in the rural areas, shaded in green color in the map (31.25%), also enjoy the game. The pie charts show the ratio of gender in the key areas. Except for Taipei City, the gender ratio is nearly constant across all regions (for Taipei, p>0.1, for all regions, p<0.01). We see a slightly higher fraction of male players in Taipei City than in other places.

4.1.2  Gamer Profiles

As mentioned earlier, a player can own multiple avatars, each with its own level and friend list. In order to measure the overall characteristics of the in-game player profiles, we examined the following features: the active hours of the day and the number of avatars owned. We also examined other variables such as the game level and trading actions, which we describe in the next section.
The first measure indicates at which hours of the day a user typically plays the game. To infer this, we defined a session, as the game log does not indicate when a user is actually playing. This happens to be a common issue in many web logs including Internet browsing and Internet TV watching [11], since users simply leave their computers on and resume online activities after a prolonged period of time. For our purposes, we defined a session as a time window containing consecutive actions of an avatar separated by no more than ten minutes.1
For each hour of a day, we could then identify how many avatars and players were active. Figure 6 displays the number of active players from a random day within the trace period. The plot is divided to show the pattern of the gender and the age of active players. The number is the smallest around 5AM, while it is the largest around 8-9PM. There was no particular gender-specific pattern (p>0.1). Age-wise, we see an increase (from 21.0% to 31.8%) in the activity of players aged 4 to 16 between June and August, which corresponds to the summer vacation in Taiwan.
Figure 6: Hourly fluctuations in the online population.
The second measure, the number of avatars owned, also indicates the activity of gamers. Figure 7 shows the number of avatars per player. The distribution is skewed, meaning that the majority of players own a small number of avatars, whereas some players own more than 10 avatars. In fact, 74.7% of players owned no more than 4 avatars. We find that gender does not play a role in determining the number of avatars owned (p>0.1).

4.2  Prevalence of Gender-Swapping

Players are free to choose the gender of their avatars, resulting in the phenomenon of gender swapping [10]. Here, we present the prevalence of gender swapping in Fairyland Online and its correlation with certain demographical features.
First we see that over a third of players owned at least one gender-swapped avatar. Table 1 shows that female players are more likely to own gender-swapped avatars than male players (p<0.001). This is consistent with past research that a sizable fraction of players swap genders, and females tend to do so more than males [37]. The geographical location, living in urban or rural regions, was found not to be correlated with gender swapping (p>0.1).
Figure 7: The distribution of the number of avatars owned by each player. The distribution is right-skewed.
Region Male player Female player Total
Urban 675 (37.4%) 537 (35.9%) 1212 (40.7%)
Rural 282 (44.6%) 254 (36.0%)  536 (35.6%)
Total 957 (37.0%) 791 (45.3%) 1748 (40.3%)
Table 1: The number and the fraction of players (against the total) with gender-swapped avatar(s) in the urban and the rural areas of Taiwan.
We then analyzed gender swapping tendencies and the number of avatars owned. Figure 8 shows the prevalence of gender-swapping as a function of the player age and the number of avatars owned. It shows the gender-swapping ratio, defined as the number of gender-swapped avatars divided by the total number of avatars owned. The upper plot in Figure 8 shows that older players are more likely to swap genders than younger players. The bottom plot in Figure 8 shows that the number of avatars has a positive correlation with the ratio of gender swapping. The prevalence of gender swapping among the older and more experienced players suggests that there may be intriguing behavioral motivations and patterns, which we discuss in the subsequent section.
Figure 8: The dependence of the fraction of gender-swapped avatars owned by a player on players' age and the total number of avatars owned.

4.3  Summary

In this section we examined the demographic features of players to understand which offline and online traits correlated with gender swapping. We found that gender-swapping ratios were independent of where people live and when they played the game. The ratios, however, showed a strong correlation with players' age and how active they were. The fact that older players swapped genders more often means that gender swapping may reflect some mature psychological aspects. That the gender-swapping ratio was higher for players with more avatars may indicate that experienced players know the benefits of gender swapping.

5  Effect of Gender Swapping
on In-Game Behavior

To understand how gender swapping impacts a player's behavior, we study four major activities of Fairyland Online: avatar level-ups, private communication, friendship formation and preservation, and trade. Among these, avatars' game levels represent personal achievements and progress, while the latter three represent social actions. For each of these, we verify the extent to which behaviors of gender swappers deviate from those of non-swappers. We introduce the following notation for the four player-gender combinations, the "gender tuple":
A.B = Player of gender A controlling avatar of gender B.
For instance, M.M and M.F indicate a male playing as a male and a female avatar, respectively.

5.1  Levels: Individual Achievements

Avatar levels represent the measure of personal development and progress in a game that is often the primary objective of gamers in an MMORPG [36,13]. A higher level is obtained when players complete missions successfully and collect enough points. Primarily the avatar level indicates the player's skill and the time they have put into the game. We asked, does gender swapping give the player any advantage in attaining higher levels?
Figure 9: The difference in male and female avatar levels, for female (top) and male (bottom) players.
Figure 10: The total session time played as an avatar, as a function of the final avatar level shown for all four player-avatar gender combinations.
To see the impact of gender swapping on game levels, we studied game players who have played with avatars of both genders, ignoring all other players who own avatars that are of one gender. This left us with 1,386 players. Among those, we extracted the highest levels of their male and female avatars (as they may own multiple avatars with the same gender). We then calculated the level difference between one's highest-level male and female avatars, plotted in Figure 9. For females (upper panel), their strongest female avatars' level tends to be slightly higher than that of their strongest male avatars' level and vice versa (lower panel), indicating that more players achieved the highest levels with their "straight" avatars. The peaks are close to zero, indicating that for many players the difference is rather minute.
While this may seem to indicate that gender swapping correlates negatively with avatar levels and in practice is not particularly advantageous, Figure 9 is likely biased as players spend varying amounts of time with different avatars, which is the strongest indicator of one's level. We find, in fact, that in Fairyland Online players are prone to playing longer with avatars of the same gender as their own. To quantify the effect, we examined the rate at which users leveled up with and without gender swapping. Figure 10 shows the relationship between the final avatar level and the total time played in Fairyland Online for the four different gender tuples for the 1,386 players. Here the x-axis is the final avatar level, and the y-axis is the total amount of time played by the avatar (starting at level 1).
From Figure 10 we can make the following three observations. Firstly, leveling up takes an increasing amount of time as one's level rises (seen from the shape of the curve). This is a common feature of many games that incorporate the level system. Secondly, male players level up more quickly than female players. To verify this, we computed the total time played until one reaches a given level, and inferred the average time spent at each level. We then tested if there was any meaningful difference in the average time spent between the male and female players at each level using the Kolmogorov-Smirnov (K-S) test. We found sufficient evidence to support that the two average time distributions are different (D=0.0611 and p<0.01). This is also in line with previous study that male players are more efficient "achievers" in the virtual world than female players [36]. Thirdly, both male and female players achieve a given level more quickly via a male avatar than via a female avatar. This finding is especially intriguing for female players, because this suggets a concrete benefit of gender swapping. For instance, F.F players (i.e., female players logged on with female avatars) played on average 19.24 hours more to reach level 50 than F.M players. We again used the K-S test and confirmed meaningful difference between F.F and F.M players (D=0.1824 and p<0.001). Since the gender of an avatar is a cosmetic factor with no difference in abilities in Fairyland Online, this means that female players exhibit a rather drastic transformation into male-like aggressive, achievement-oriented characteristic when pretending as a male avatar [8].

5.2  Whisper: Private One-to-One Messaging

Here and in the two subsequent sections we study user-to-user interactions. The first one is "Whisper", a private one-to-one communication function. While players can whisper to any user in the game, the messages are revealed to the sender and the receiver only, as opposed to public broadcast. Whispering thus can reveal the patterns of affinity between users.
In order to determine the effect of gender in whispering, we examined the probability that a given user of type A.B sends whispers to another user of type A′.B′ from the entire data. Since whisper interaction is directional, there are 16 possible gender combinations (A.B→ A′.B′). For comparison, we also computed its null values, the expected probability of whisper when we assume that the sender of a whisper selects its receiver at random regardless of their gender combination, by randomly matching 10,000 pairs of avatars in the network.
Table 2 displays the magnification factor of the probability of whisper (i.e., the real probability divided by its random expectation). Therefore a value larger than 1.0 (marked in bold) indicates that the whispers occurred more frequently than expected. The table shows a trend where a whisper is likelier to take place between avatars of different genders, whichever the players' genders are. For instance, a male player with a male avatar (M.M) is 1.26 to 1.34 times likelier to send a whisper message to a female avatar (both M.F and F.F) than expected by chance. In contrast, the probability dropped roughly by a third to a half when the other player had a male avatar (M.M or F.M), irrespective of the true gender of the other player.
Despite this interesting trend we may need to take caution, as whisper does not necessarily imply a substantial relationship; since one can send a whisper to any random player, some whispers may only be a failed, unreciprocated attempt to strike up a real, substantive conversation. Hence to find patterns of a sustained, concrete conversation that took place for some measurable time, we chose and focused on all avatar pairs that exchanged no less than 35 whispers in total, of which there were 36,820 avatar pairs. As they were reciprocated whisper pairs, we ignored the sender-receiver relationship, and examined how the gender configurations affected the length of the conversation, presuming that a longer conversation meant a more substantial relationship.
Receiver type
M.M 0.64 1.34 0.50 1.26
Sender M.F 1.30 0.76 1.07 0.59
type F.M0.51 1.18 0.83 1.99
F.F 1.25 0.57 2.09 0.87
Table 2: The frequency of whispers divided by random expectation across all possible gender tuples.
We compared the duration of conversations between three types of player pairs: female and female, female and male, and male and male. When we considered all conversations (involving at least one whisper message), we found that the median and the mean durations were 3.1 minutes and 33.7 minutes, respectively. The large discrepancy between the two indicates a possible right-skewed distribution of conversations (e.g., certain conversations lasting several hours or more). Looking at the 36,820 pairs (with 35 whispers or more), we found that the median and the mean duration were 32.6 minutes and 125.9 minutes, respectively.
In Figure 11 we show the relative duration of the long conversations for the different types of player gender pairs. We normalized the duration so that the highest bar is 100. For each player pair type, we show two bars: one where the avatar genders are the same and another where avatar genders are opposite. The figure shows that players engage in a lengthier conversation when their avatars are of different genders regardless of the true genders of the players. In fact, the conversation duration increased 1.62- to 1.77-fold when the avatars were of the opposite sex. Controlling for avatar genders, we also find that conversation involving a female player lasts longer. This supports a well-established notion that females tend to socialize more actively than male players do in online games [17].

5.3  Friendship: Sustained social ties

The next interaction we analyze is friendship. In Fairyland Online, each avatar maintains a "friend list" of other avatars. Unlike whisper, a friendship tie is bidirectional in that both avatars must agree to be friends. Once friends, users can easily locate and contact each other inside the virtual world. We examine whether gender swapping affects the chance of people becoming friends.
To compute the effect of avatar gender on the chance of forming a friendship tie between different avatars, we first note that there are 10 possible gender configurations in a bidirectional link (player1.avatar1 and player2.avatar2). Then we count all friendship links between gender types, and compare it with the randomly expected number of friendship links without considering genders. The resulting magnification factor (i.e., observed frequencies divided by random expectation) are shown in Table 3. The table is symmetric, as friendship is bidirectional. The larger-than-expected pair types (the values larger than 1.0) are marked in bold.
Figure 11: Duration of long whispers (i.e., conversations).
M.M 1.03 1.08 0.82 0.97
M.F - 1.07 0.84 0.77
F.M - - 1.08 1.26
F.F - - - 1.23
Table 3: The prevalence of friendship divided by random expectations across all possible gender tuples.
The table shows a higher probability of friendship formation between the same-gendered players, so that a male player is likelier to befriend another male player than a female player, and vice versa. Interestingly, controlling for player genders, we found that the players were slightly likelier to form a friendship tie when their avatars had opposite genders. For instance, two female players were 1.26 times more likely to become friends with each other when their avatars differed than when they did not. Since players are not told of the real gender of other players and can only see the avatar gender, this appears to indicate the possibility that a subconscious, implicit selection process may be at work [26]. The pattern of people being linked more often when they are of the same gender but of different avatar genders was observed also in the whisper network.

5.4  Trade: Exchange and Sales of Items

Lastly, we study the patterns of Trade, namely the exchange and sales of items. The economic activity of item acquisition and currency transfer is a staple of MMORPGs, as players need various items to strengthen, heal, or embellish their avatars. Since items are often directly related to one's performance, trade is a highly strategic activity and can incur other activities as its precursor or aftermath: private conversations often precede trading and social ties can be formed afterwards [26]. Highly valuable items can fetch such a high price that some players become serious businessmen dedicated to item acquisition and sales, often making a handsome sum of real money by selling them online or offline [31].
We investigate whether a player's or an avatar's gender has any impact on trade. Fairyland Online provides a wide variety of items for use and trade. Furthermore, some items are indispensable in that players need those items to qualify for a level-up (a Fairyland Online-specific example would be a bunny required at a certain level that can be lured with a particular herb). All these needs are motivations for trade, which can take a form of item-to-item bartering or straight-up sale for gold.
Our data contain all trading interactions, allowing us to observe details of the trade activities including the item traded and its price. To see the relation with gender, we first extracted the most traded items during the ten-month observational period and built a list of its market value (price), the median gold fetched in sales. We intentionally omitted bartering (item-for-item) because the market value of an item was difficult to determine from such data. Table 4 shows the five most traded items along with their median market prices.
Market price Expected transaction gain (sell − buy)
Item Item name Description (median) F.F F.M M.F M.M
fig/items/greaterhealingpotion.png Healing Potion Used after combat 1137.5 125.0 −100.0 −133.0 66.6
fig/items/bloodcirculator.png Blood Circulator Used for fast regeneration 959.5 160.7 −34.5 25.0 −100.0
fig/items/iron.png Iron Ingot Used for smithing 125 91.3 −17.5 36.8 0.01
fig/items/applecider.png Apple Cider Used for quests 416.6 −83.3 −200.0 41.6 0.0
fig/items/cotton.png Cotton Used for making yarn 40.3 152.5 17.5 −154.8 −8.3
Table 4: Five most traded items with their median bartered market price in gold. The expected transaction gains (net profit) are also shown. Gains that exceed the market value by more than 5% points are marked in bold.
We then examined the frequency of trade and the price of each item between different gender combinations. For simplicity, we examined the average selling and buying prices that the items fetched from trades between the four gender combinations (M.M, M.F, F.M, and F.F). The difference between the two, i.e., the average selling price minus the average buying price, would be the expected monetary gain for the selling player. The net gains calculated in this fashion are also shown in Table 4.
We find a few interesting patterns regarding trade in Table 4. First, the selling and the buying prices can vary widely depending on the player's gender as well as that of the avatar: for instance, Healing Potion is sold at a higher price when the seller is a F.F than when they are F.M, even though the real player gender is the same. Second, focusing only on significant net gains (defined to be exhibiting more than a 5% gain over the median market price), we find that female players in general make better, more profitable salespeople than males do. When we control for the player gender, we also saw that female avatars were likelier to succeed in profiting from trade than male avatars. This could imply that male players benefit through gender swapping in trades.
We also tracked select items, which revealed to us some interesting potential mechanisms of price determination. The Healing Potion, for instance, is a popular item among players who were warriors or enjoyed combats, i.e., usually male avatars. Hence, such "achievers" may be willing to buy them above market price if they need to. Then their willingness could be exploited by those who do not need the item, for instance female avatars who are more of the "socializer." This could be a plausible explanation for the sizable discrepancies in the expected net gain of Healing Potion between F.F and F.M, as we saw also in other contexts that female players become drastically aggressive and masculine when controlling a male avatar. A similar dynamic might apply to other items as well, although many other factors may well be at play that need further examination.

5.5  Summary

In this section, we investigated how peoples' behaviors change when they swap genders of the avatars they control. Our major findings include that male players achieve higher levels faster than females in general, but female players also do so when they play with male avatars; and players tend to whisper more to opposite gender avatars, but they are likelier to become friends with players of the same gender. Our study also showed that female avatars in many cases profited from trade than male avatars did. The fact that many items were sold at different prices for the male and female avatars at statistically significant levels may imply an intricate relationship between the traded items and avatar genders.
Since there are no built-in capability differences in male and female avatars, the fact that we observe significant behavioral changes implies that people act according to the perception or the image of the gender that they are playing and of the gender of the interacting players. Therefore there may also be advantages in gender swapping, not innate but derived from people's subjective perception and the resulting responses, leading to the intriguing differences in the behavior patterns [17].

6  Implications

In this paper, we have examined the gender swapping behavior in MMORPG. Through several key interaction patterns, we found that male players act and get treated more like a female when playing with a female avatar (e.g., chatted for a longer period of time, enjoyed increased affinity by other M.M players, traded items at a higher price), and vice versa for female players. Our major findings on how gender roles and identity are associated with avatars (rather than players) suggest several interesting applications, of which two major ones are:
Similar to detecting bots and fraud, detection of malicious users and other types of potential criminals (sybils and cheaters) is often conducted by analyzing user behaviors in many web services [30,9,34]. Our study can produce valuable insights in this regard.

7  Conclusion

In this paper, we studied the issue of gender in large-scale MMORPG social interaction networks, and investigated an interesting problem in human behavioral research called gender swapping. While there have been some past studies on the issue, they were based on uncontrolled online opinion surveys, which could have been subject to self-selection bias. Unlike the previous works, we used the complete interaction and profile data of the players from Fairyland Online, thereby easily overcoming the limitations of self-selection bias and small scale. We searched for intriguing patterns of social behavior that were potentially correlated with gender swapping, and found several significant trends in conversation, friendship formation, and economic behavior.
Our biggest finding was that (a) gender swapping is prevalent across player gender and age, and it becomes more prevalent with the players' age, experience, and activity; (b) while in an online setting people try to make friends with the opposite sex, they turn out to be of the same gender in real life, indicating that assuming an alternative identity does not completely mask the real persona; (c) female players tend to act more masculine and achievement-oriented when acting as a male than actual males do; and (d) people's personalities and strategies (such as when trading for items or gold) depend on their assumed identities, an example being that female avatars, regardless of the player's real gender, appear more successful at trade.
With these interesting findings, we believe we have demonstrated the potential of massive user data sets from MMORPGs in studying intriguing and important human psychology and behavior. Free of self-selection bias and limitations of scale, we were able to make two novel observations that had not been shown in previous studies: one is that both male and female players tend to talk more with avatars of the opposite gender; two players tend to become friends with others whose real gender is the same as one's own even though they only know the virtual gender of others.
We foresee multiple interesting avenues for taking our research further, as our ability to observe human behavior and analyze them ever improve both online and offline. Here we propose two possible examples. First, a natural extension of our work would be to consider more types of user interactions including many-to-many interactions, i.e., group dynamics, which can be significantly more complex than one-to-one interactions that we studied here. Since a main attraction of MMORPGs can in fact be found in group activities, exploring how gender affects such interactions would be a necessary next step. Second, if we could define and retrieve even more details of the interactions-the content of a conversation, for instance-there would be an exponential development in how we analyze and understand the full complexity of human social behavior. One example usage of the conversation data would be a study of the role of verbal cues and gender identities, as it is believed that verbal cues can be instrumental in revealing a person's identity, and we would find many applications of the lessons learned from how gender swappers change their language patterns.


The authors would like to thank those at Lager Network Technologies, Inc., especially Philip Chang and Samuel Kuo, for providing the data and the anonymous reviewers for their valuable comments. This work was supported by the National Science Council of Taiwan under the grant (NSC100-2628-E-001-002-MY3), the WCU (World Class University) program under the National Research Foundation of Korea and funded by the Ministry of Education, Science and Technology of Korea (R31-30007), the international cooperation program managed by National Research Foundation of Korea (2011-0030936), the National Research Foundation of Korea funded by the Korean Government (KRF-20100004910), and the Korea Advanced Institute of Science and Technology.


[1] Aion online: The official fantasy MMORPG website.">
[2] Dori Hartley: Cybe crime: That's entertainment!">
[3] Fairyland Online (English).">
[4] Pardus - free browser game set in space.">
[5] Terra Nova: New Daedalus: Demographics of WoW players.">
[6] R. A. Albert-lászló Barabási. Emergence of scaling in random networks. Science, 286(5439):509-512, 1999.
[7] M. Bartini. Gender role flexibility in early adolescence: Developmental change in attitudes, self-perceptions, and behaviors. Sex Roles, 55:233-245, 2006.
[8] R. Bartle. Hearts, clubs, diamonds, spades: players who suit MUDs. Journal of Virtual Environments, 1996.
[9] J. Blackburn, R. Simha, N. Kourtellis, X. Zuo, M. Ripeanu, J. Skvoretz, and A. Iamnitchi. Branded with a scarlet "c": cheaters in a gaming social network. In Proc. of the International Conference on World Wide Web, 2012.
[10] A. S. Bruckman. Gender swapping on the Internet. In Proc. of the Internet Society, 1993.
[11] M. Cha, P. Rodriguez, J. Crowcroft, S. Moon, and X. Amatriani. Watching television over an IP network. In Proc. of the ACM Internet Measurement Conference, 2008.
[12] K.-T. Chen and L.-W. Hong. User Identification based on Game-Play Activity Patterns. In Proc. of the Annual Workshop on Network and Systems Support for Games, 2007.
[13] H. Cole and M. D. Griffiths. Social interactions in massively multiplayer online role-playing gamers. CyberPsychology & Behavior, 10(4):575-583, 2007.
[14] N. Ducheneaut, N. Yee, E. Nickell, and R. J. Moore. "Alone together?: Exploring the social dynamics of massively multiplayer online games. In Proc. of the ACM Conference on Human Factors in Computing Systems, 2006.
[15] M. D. Griffiths, M. N. Davies, and D. Chappell. Breaking the stereotype: The case of online gaming. CyberPsychology & Behavior, 6(1):81-91, 2003.
[16] M. D. Griffiths, M. N. Davies, and D. Chappell. Demographic factors and playing variables in online computer gaming. CyberPsychology & Behavior, 7(4):479-487, 2004.
[17] S. Huh and D. Williams. Dude Looks like a Lady: Gender Swapping in an Online Game. Springer London, 2010.
[18] D. Lazer, A. Pentland, L. Adamic, S. Aral, A. lászló Barabási, D. Brewer, N. Christakis, N. Contractor, J. Fowler, M. Gutmann, T. Jebara, G. King, M. Macy, D. Roy, and M. V. Alstyne. Computational social science. Science, 323(5915):721-723, 2009.
[19] J. W. Lee. Gender Roles. Nova Science Pub Inc, 2005.
[20] L. L. Lindsey. Gender Roles: A Sociological Perspective. Pearson, 2010.
[21] S.-K. Lo, C.-C. Wang, and W. Fang. Physical interpersonal relationships and social anxiety among online game players. CyberPsychology & Behavior, 8(1):15-20, 2005.
[22] M. E. J. Newman. The structure and function of complex networks. Siam Review, 45(2):167-256, 2003.
[23] H.-K. Pao, K.-T. Chen, and H.-C. Chang. Game Bot Detection via Avatar Trajectory Analysis. IEEE Transactions on Computational Intelligence and AI in Games, Sep 2010.
[24] J. Park, M. Kim, and M. Cha. An inconvenient truth: Where you live decides how you are treated online. In Workshop on Information in Networks, 2012.
[25] J. Park and A. lászló Barabási. Distribution of node characteristics in complex networks. PNAS, 104(46):17916-17920, 2007.
[26] S. Son, A. R. Kang, H. chul Kim, T. Kwon, J. Park, and H. K. Kim. Analysis of context dependence in social interaction networks of a massively multiplayer online role-plyaing game. PloS One, 7(4):e33918, 2012.
[27] M. Szell, R. Lambiotte, and S. Thurner. Multirelational organization of large-scale social networks in an online world. PNAS, 107(31):13636-13641, 2010.
[28] M. Szell and S. Thurner. How women organize social networks different from men. Scientific Reports, 3:1214, 2013.
[29] C.-C. Wang and C.-H. Wang. Helping others in online games: Prosocial behavior in cyberspace. CyberPsychology & Behavior, 11(3):344-346, 2008.
[30] G. Wang, C. Wilson, X. Zhao, Y. Zhu, M. Mohanlal, H. Zheng, and B. Y. Zhao. Serf and turf: crowdturfing for fun and profit. In Proc. of the International Conference on World Wide Web, 2012.
[31] Q.-H. Wang, V. Mayer-Schonberger, and X. Yang. The determinants of monetary value of virtual goods: An empirical study for a cross-section of MMORPGs. Information Systems Frontiers, pages 1-15, 2012.
[32] S. Wasserman and K. Faus. Social Network Analysis. Cambridge University Press, 1994.
[33] M. E. Wiesner-Hanks. Gender in History: Global Perspectives. Wiley-Blackwell, 2010.
[34] X. Xing, Y.-L. Liang, H. Cheng, J. Dang, S. Huang, R. Han, X. Liu, Q. Lv, and S. Mishra. Safevchat: detecting obscene content and misbehaving users in online video chat services. In Proc. of the International Conference on World Wide Web, 2011.
[35] N. Yee. The demographics, motivations, and derived experiences of users of massively multi-user online graphical environments. MIT Press Journals, 9(6):772-775, 2006.
[36] N. Yee. Motivations for play in online games. CyberPsychology & Behavior, 9(6):772-775, 2006.
[37] H. Zaheer and G. M. D. Gender swapping and socializing in cyberspace: An exploratory study. CyberPsychology & Behavior, 11(1):47-53, 2008.


1. The data reveals that nearly all (98.4%) of consecutive actions are separated by less than one hour.

Sheng-Wei Chen (also known as Kuan-Ta Chen) 
Last Update September 19, 2017