Cheating in Online Games: Reading Friendship Data

Identify what each record represents

Cheating in online games can be studied through detected violations, social connections or observed matches. Those records answer different questions. A friendship edge says two accounts are linked; a ban flag records a platform classification in a specified dataset. Neither record supplies a conversation between the players.

Before discussing how behavior spreads, separate the behavior of interest from the event that makes it visible. A detection recorded later than an action can change the apparent sequence. This matters even when a graph contains millions of accounts.

A historical Steam network dataset

Blackburn and colleagues’ 2012 paper mapped 12,479,765 Steam accounts through a March–April 2011 crawl following friendship links. Among discovered profiles, 720,469 had Valve Anti-Cheat (VAC) flags. An October re-crawl identified newly observed flags.

Following links builds a connected picture, but access to profile information determines what can be observed. The number of accounts is therefore one part of the method. Coverage, missing information and the meaning of the label remain important when deciding which population a result describes.

Separate clustering from influence

Marked accounts had higher proportions of marked friends. Time-grouped ban observations also showed an association with previously recorded marked friends. The authors acknowledged difficulty distinguishing selection of similar friends from social influence.

Imagine three possible histories for an invented cluster. People with similar attitudes might seek one another out. One person might influence an existing friend. Several people might encounter the same outside community independently. Each history could produce connected accounts with similar behavior. Distinguishing them requires information about earlier behavior, relationship formation and shared circumstances. A picture of the final cluster cannot supply all three histories.

Keep friendship and match interactions distinct

A separate Team Fortress 2 server sample contained 10,354 players, including 93 marked for bans in another game. Comparisons of distinct cooperative and opposing partners did not detect significant differences between marked and unmarked players.

Those null comparisons leave attitudes unresolved. A log can show that two people interacted without recording whether either knew about a ban or approved of the other’s behavior. Counting distinct partners also differs from counting repeated encounters with one person. Frame each conclusion around its recorded measure: a partner-count comparison leaves questions about acceptance, exclusion or trust unanswered.

Check when the observation became available

Ban dates meant on-or-before first observation, while the temporal analysis reused a friendship snapshot despite relationships changing over time.

In an invented timeline, account A breaks a rule on day one but its flag first appears on day ten. Account B’s flag is seen on day six. Sorting those flags places B first without establishing which behavior happened first. A friendship visible on day twelve also leaves its start date unknown. This invented timeline makes the missing timing information explicit.

Ask a question the network can answer

When reading a claim about gaming networks, identify the connection being counted, the account status being observed and the observation period. Then distinguish a result from the large profile graph from one obtained on a single game server. These choices define the comparison before any explanation of influence is added.

The full paper and original conference listing identify the same 2012 research. Its historical records offer a concrete way to discuss associations and their limits. A useful account states which pattern appeared, which alternatives remain possible and what additional observations would distinguish them. That keeps a network result informative without turning a friendship link into a complete explanation of someone’s actions.

Keep exploring

FOLLOW THE SOURCE

Sources & further reading

Have a correction? Tell our editorial team.

✦ NICE. YOU LEARNED SOMETHING.