Cheaters in the steam community: a social network perspective

The video explains a good deal about the nuances of cheating. In this post, we will look at those that intentional broke the code, what happens to them and their friends after they are branded as cheaters in Steam.

A cheater is a loathed label in society, anyone caught or suspected of cheating faced severe consequences from established written rules (e.g. fines or bans) to conventional unwritten ones (e.g. ostracism). The motivation for such punishment can be varied, but the most common one is fairness, a fairness that everyone is playing by the game and social rules. The motivations for cheating are quite varied and it does not simply focus as a moral issue, it could be a result of a combination of peers, group norms or attitudes towards cheating.

The paper reports on cheaters from a social network analysis which means a huge deal of public data gathered from the internet. Jeremy Blackburn (University of South Florida) is a computer science PhD student specializing in social network analysis. His co-authors include Nicolas Kourtellis, John Skvoretz, Matei Ripeanu and Adriana Iamnitchi. The paper was published in ACM Transactions on Internet Technology, a publication in computer science of which is outside of my expertise.

You can read the article from this link here. This is crossposted at my blog at VG researcher.

The authors made a good argument about studying cheating behaviours on a large scale as it can provide evidence to social sciences and psychological findings of other dishonest behaviours that usually are done on a small scale. A player’s position in a social network can reveal how anti-cheating policies work or affect others. Although, the study does not reveal information about the providers of cheat tools as reported in PC Gamer (Maiberg, 2014), it certainly reveals information from cheaters.

Method

Their target population are Steam users as Steam have a huge market share of the PC digital market. Steam users’ profile can be viewed publicly (even mine) on the internet, that means they can use a web crawler to gather information on a large scale. On a tangent, Ars Technica also did a web crawl on steam and checked out the most popular games (Orland, 2014). More importantly, Steam has an anti-cheating system called Valve Anti-Cheat (VAC) which for the purpose of the social network analysis marks cheaters with a publicly visible VAC ban on their profiles. Marked users are banned from playing VAC-secured servers of the game they cheated in. For example, a user cheated in Team Fortress 2 and is therefore banned from playing VAC-secured Team Fortress 2 servers, but is allowed to play others games like DoTA2. The user is banned after an undetermined delay of detecting the cheat.

They crawled the steam community website between 16 March and 3 April, 2011 and a second crawl between 18 October and 29 October, 2011. They crawled 12.5 million users, although 9 million profiles were public and 1 million were not. The information gathered from the public profiles included nickname, privacy setting, friends list, group memberships, list of games owned, time spent on videogames in the past two week, lifetime gameplay statistics, geographical location (if made available by the user) and the VAC-ban flag.

At the time of 2011 crawls, they had to use third-party database, such vacbanned.com, to find the cheaters and they reported some of their findings in an earlier paper (Blackburn et al., 2012), they found 700 thousands users who were caught cheating. Later on, they were able to obtain more accurate data that can tell them when a VAC ban is imposed and the social dynamics in the aftermath of such ban. With these new methods, the authors started daily monitoring on 20 June 2012 for ban statuses. When a user is newly VAC banned, they monitored the users’ friends list for a 10-day period. They also monitored non-cheater users as a control group.

Results

The analytical techniques for large datasets is beyond my understanding.

The authors analyzed with what information steam profiles can provide.

While summary statistics can help give an idea as to what is going on, outliers can influence them. On average, Steam Community users have around 7 friends. However, around 40,000 users have over 100 friends. For this reason, much of the results in the paper are shown as distributions. For example, the complimentary cumulative distribution function (CCDF), which gives the probability of a random user having greater than or equal to a certain value is used for degree distributions. I.e., it tells us the fraction of Steam Community users that have at least, a certain number of friends.

In addition to plotting distributions, the paper makes use of a variety of social metrics. The degree (number of friends) distribution is quite common and self explanatory, however, the paper makes use of several social proximity metrics that are a bit more complicated.

Neighborhood overlap was used to produce a normalized measure of the number of common friends two users have. Similarly, several geo-social metrics used in the paper capture the real-world distance between users.

Gaming habits: the type of games was divided into single-player-only, multiplayer-only and co-op categories. Steam users in general owned more multiplayer videogames than single-player games. In general, non-cheaters owned more videogames than cheaters, but this difference is much smaller for multiplayer-only games.

In terms of lifetime hours of gameplay, cheaters tend to spend fewer hours on single-player games and co-op games.

Friends list: A pattern emerged in that cheaters are more likely to be friends with other cheaters. 70% of non-cheaters don’t have cheater friends, but 70% of cheaters have at least 10% who are cheaters. Interestingly, 15% of cheaters have over half of their friends being cheaters. They found a correlation in that more friends also means more time and money (by buying games) spent in videogames.

When the VAC ban happens: Their 2012 report indicated that from the crawl data on April 2011 and October 2011, 10% of cheaters changed their profiles to private and 87% were unchanged. In comparison to non-cheaters, only 3% changed their profiles to private. They found that 43% of cheaters had net loss of friends, 13% gained more and 43% were unchanged. Non-cheaters in comparison, 25% had net loss of friends, 36% gained more and 39% were unchanged. However, they only looked at two snapshots taken 6 months apart.

Their more recent dataset examined newly branded cheaters and monitored their friends list over a 10-day period. Cheaters lost friends the most on within the first days of the VAC ban.

Where are the cheaters at?: Steam users can optionally set their geographical locations on their profiles, please bear with the following with this fact in mind. The data revealed that 55,000 cheaters were from the Nordic countries (e.g. Denmark, Norway, Sweden) whereas 39,000 cheaters were from the U.S.

Do cheaters congregate?: Even though steam users can set their geographical locations to any place, it seems that they friend others who are geographically close to each other, cheaters tend to friends geographically close users and much close and more often than non-cheaters do. Interestingly, cheaters tend to have non-cheater friends who are geographical far from them. The authors acknowledged that users can set their geographical locations to any place. They argued that Steam acts like a social networking service and given that one of the issues of multiplayer gaming is latency or lag, it is reasonable that users will play on servers with low lag and therefore are geographically close to them which also brings them into contact with geographically close potential friends.

Data on friendship overlaps, that is how many friends have mutual friendship with each, indicate that cheaters have greater friendship overlap with other cheaters than with non-cheaters and vice versa.

Steam groups are user-generated and the data crawls indicated that over 5 million users were members to at least one of over 1 million groups. However, only 1% of groups have more than 150 members whereas 90% had less than 31 members. They found that 65.4% of cheaters are member of at least one group in comparison to 58.2% of non-cheaters. An interesting statistic is that cheaters are more likely to be members of more groups than non-cheaters do.

The authors examined the centrality of steam users, that is their social influence in the network. The results that I could understand is that among the top 1% of the most central steam users, cheaters make up less than 7% of them.

How does cheating spread over time?: At the time a user is given the VAC ban, they had on average 5.97 (SD = 9.31) friends who were cheaters whereas the non-cheating/control user had on average 1.25 (SD = 3.23) friends who were cheaters. This indicates the influence of cheating friends could have an impact on users behaviours. Interestingly, they monitored a network where at first there were 28 cheaters which soon spread across the users that eventually a month later, there were 279 cheaters.

Discussion

The take home message is that the cheating behaviours of a user can influence the likelihood of cheating of other users who are connected to them. In other words, who you are friends with can influence your future behaviours and your behaviours can affect your friendship with others. There are good deal of descriptive information that we can infer from this social network analysis. Cheaters are quite social gamers in that they own and play more multiplayer videogames than non-cheaters, they are more likely to part of a group which means a social identity.

The paper was written from a computer science and social network analysis perspective, so I did my best in understanding the paper. I offer my interpretation as a social scientist.

The study supports the works of Vivian Hsueh-Hua Chen (Nanyang Technological University) who published several survey studies on cheating behaviours (see Duh & Chen, 2009; Chen & Wu, 2013; Wu & Chen, 2013). The gist of Chen and her colleagues’ findings is that within an online environment where people are anonymous which affect their identities to be loosely defined, they would progressively adopt the group norms as they spent more time in that environment. In other words, the more time spent playing, the more a player adopts the mannerism and attitudes of other players they interact with. In relation to cheating in online games, seeing a lot of cheating and seeing their friends and significant others cheat would send a message to the player that it is a negative, yet a normal thing to do. Perhaps, some players might feel uneasy that they have not cheated like the rest of their group members or with gamers. An important point is that the cheating had a rewarding outcome, seeing their friends succeed through cheating means that it might be worth the risk to reap the benefits. Another important point is that the relationship works both ways and Blackburn’s data supports this argument that your behaviours can affect your social network and you are affected by your social network which would work into a reciprocal relationship.

The authors focused on the VAC ban status of steam users, but there are also other bans, community ban and trade ban. Trade ban is for users who violated trading rules in the steam market, such as fraud, running scams, buying and selling a lot in a short amount of time. The curious one is the community ban, there is not much documentation about it. A possible meaning could be that it is used for forum and discussion purposes, perhaps given to users who behaved offensively to other users, whose profile may be offensive or inflammatory or did something in-game that is not related to cheating, maybe trolling. Another less likely meaning is that users were banned from certain groups because they flamed or harassed members of the group. It would interesting to crawl the steam community to examine the relationship between the bans statuses of the cheaters. The reasoning is that cheating is an antisocial behaviour, committing such behaviours would mean that they are thinking with antisocial intent. These antisocial thinking can also mean that committing other antisocial behaviours in steam is also likely than not.

Another idea stemming from a sensitive line of research is harassment and toxic behaviours. One way is to look at user profiles, what sort of language or words do a user present themselves to others. People might present themselves as they were in real life, but sometimes we encounter trolls whose profile matched their behaviours and hostility. We can also extend to which group they belong to, as Chen and colleagues observed that group membership can influence player behaviours. A look into which group users belong to might give some clues as to what group norms they adopt, how would a user who belong to 4chan, griefing or perhaps a hate group might behave in-game.

A more remote research line is the relationship between cheating and trust. I have been acquainted with some research regarding how a lack of trust, a sense of entitlement or a sense of being victimized might increase the likelihood of thinking antisocially which could in turn lead to antisocial behaviours, including cheating. An anecdote is that when a user opened a thread where they expressed that they were treated unfairly about why they were banned and they do not understand why they were banned or they do not see the (potential) harm of their actions when other users pointed them out. This is rather interesting for empathy researchers. I’ll just leave that idea for others to pick up on.

 References

Blackburn, J., Kourtellis, N., Skvoretz, J., Ripeanu, M., & Iamnitchi, A. (2014). Cheating in online games: A social network perspective. ACM Trans. Internet Technol., 13 (3).DOI: 10.1145/2602570

Blackburn, J., Simha, R., Kourtellis, N., Zuo, X., Ripeanu, M., Skvoretz, J., & Iamnitchi, A. (2012). Branded with a scarlet “c”: Cheaters in a gaming social network. In Proceedings of the 21st International Conference on World Wide Web, WWW ’12, (pp. 81-90). New York, NY, USA: ACM. DOI:10.1145/2187836.2187848

Chen, V. H., & Wu, Y. (2013). Group identification as a mediator of the effect of players’ anonymity on cheating in online games. Behaviour & Information Technology, (pp. 1-10). DOI:10.1080/0144929x.2013.843721

Duh, H. B., & Chen, V. H. H. (2009). Cheating Behaviors in Online Gaming, vol. 5621 of Lecture Notes in Computer Science , chap. 61, (pp. 567-573). Berlin, Heidelberg: Springer Berlin Heidelberg.DOI: 10.1007/978-3-642-02774-1_6

Maiberg, E. (April, 2014). Hacks! An investigation into aimbot dealers, wallhack users, and the million-dollar business of video game cheating. PC Gamer, URL: http://www.pcgamer.com/2014/04/30/hacks-an-investigation-into-aimbot-dealers-wallhack-users-and-the-million-dollar-business-of-video-game-cheating/

Orland, K. (April, 2014). Introducing Steam Gauge: Ars reveals Steam’s most popular games. Ars Technica, URL: http://arstechnica.com/gaming/2014/04/introducing-steam-gauge-ars-reveals-steams-most-popular-games/

Wu, Y., & Chen, V. H. H. (2013). A social-cognitive approach to online game cheating. Computers in Human Behavior, 29 (6), 2557-2567. DOI: 10.1016/j.chb.2013.06.032