DEVBro
← Writing
Essay · 21 min read · Jun 19, 2026

Toxicity and blame: why players turn on each other

Toxicity in multiplayer games isn't a people problem, it's a design problem. Here's the psychology under the blame reflex, the design choices that load the gun, and what one developer can actually do about it.

D DEVBro
game-designplayer-psychologytoxicity

A deeper look at one section from What makes a game a good game?: the part that wouldn’t leave me alone.

You queue up. You lose. Somewhere around the loading screen of the next match, a small, ugly voice in your head has already finished its work: that was the jungler’s fault. You didn’t choose to think it. It arrived fully formed, complete with evidence. This is the most universal experience in multiplayer gaming, more universal than winning, more universal than fun, the reflex to find a name for the loss and to make sure that name isn’t yours.

In the broad essay I wrote about what makes games good, I argued that toxicity is a design problem, not a people problem. I still believe that. But “it’s the design” is the conclusion, and conclusions are cheap. This article is the working: the psychology underneath the blame reflex, the specific design choices that load the gun, and the concrete things I can actually do as one person building a small game. I went looking for the science, and the science is better than I expected.

A horizontal row of identical dots, with one bold coral arrow singling out one of them.

Part 1. The psychology of blame: your brain is protecting you, not the truth

Two mirrored arrows: one looping inward to its own dot, one flung outward at a separate dot.

Start with the uncomfortable premise. Blaming your teammate is not a character flaw. It is the predictable output of two of the best-documented biases in social psychology, and you run them automatically.

The first is the fundamental attribution error. The term was coined by Lee Ross in 1977, building on a 1967 experiment by Edward Jones and Victor Harris. The error is this: when other people fail, we explain it with their character (“he’s bad, he’s selfish, he doesn’t care”), and we systematically under-weight their situation (“he was getting camped, his lane was a losing matchup, his framerate tanked”). When your support misses a skillshot, your brain doesn’t reach for “difficult angle under pressure.” It reaches for “this person is trash.”

The second bias is the engine that makes it personal: self-serving bias. We attribute our successes to internal factors like skill and effort, and our failures to external ones like luck, teammates, and the game. The two biases interlock into a perfect machine for conflict. The same dropped play is, when you do it, the situation’s fault, and when they do it, their fault. Everyone in the lobby is running the identical software, which is why a single bad teamfight can produce five players who each genuinely believe they were carrying four idiots.

Now add the fuel. The frustration-aggression hypothesis, proposed by Dollard and colleagues in 1939 and reformulated by Leonard Berkowitz in 1989, says that frustration (a blocked goal) produces negative affect, and negative affect produces aggression. A competitive game whose entire purpose is to block your goal is, structurally, a frustration generator. Researchers have tested this directly in games. Breuer, Scharkow, and Quandt’s 2015 study “Sore Losers?” found that losing a video game increased aggression, and that the effect was mediated by negative affect, exactly Berkowitz’s prediction. You don’t get angry because the game is violent. You get angry because you lost, and losing feels bad, and the bad feeling needs somewhere to go.

Where it goes is the last piece: scapegoating and displacement. Berkowitz’s tradition connects directly to the classic theory of scapegoating. When the real source of frustration (your own misplay, variance, a smurf on the enemy team) is too painful or too unavailable to attack, aggression displaces onto a safer, closer target. In a five-person team, the safer target is sitting right there on your scoreboard. Blaming a teammate is cheaper than blaming yourself, because the alternative, “I am not as good as I thought,” is an attack on the ego, and the ego will burn down the whole lobby before it accepts that.

This is the thing I keep coming back to: in a team game, blame has a human address. That’s the whole problem in one sentence.

Part 2. Why online makes it so much worse

A single coral figure half-hidden behind a row of evenly spaced ink bars, like a screen.

If the biases above are universal, why is a ranked queue so much nastier than a pickup basketball game where people also lose and also get frustrated? Because online environments strip out the brakes.

The canonical explanation is John Suler’s online disinhibition effect (2004, CyberPsychology & Behavior). Suler identified six factors that loosen our normal restraint online: dissociative anonymity (“you don’t know me”), invisibility (“you can’t see me”), asynchronicity, solipsistic introjection, dissociative imagination, and minimization of authority. He split the result into benign disinhibition, people being kinder and more open, and toxic disinhibition, people being crueler. The same conditions that let a shy person open up let an ordinary person say things to a stranger they would never say to a face. In gamer folklore this is “John Gabriel’s Greater Internet Fuckwad Theory” (normal person + anonymity + audience = jerk), and Suler is the academic spine under the joke.

Behind anonymity sits an older idea: deindividuation, popularized by Philip Zimbardo (1969), who argued anonymity reduces self-awareness and accountability and disinhibits antisocial behavior. But I want to be precise, because the research is more interesting than the cliché. The 1998 meta-analysis by Postmes and Spears, and the Social Identity model of Deindividuation Effects (SIDE), found that anonymity does not reliably make people aggressive on its own. It makes people conform harder to whatever the salient group norm is. Anonymity in a group with a cruel norm amplifies cruelty; anonymity in a group with a prosocial norm amplifies kindness. That single nuance is the most hopeful finding in this entire article, and I’ll come back to it, because it means the lever isn’t “remove anonymity,” it’s “set the norm.”

There’s empirical backing specific to games. Kordyaka and colleagues (2020), building “a unified theory of toxic behavior in video games,” concluded that the online disinhibition effect was the single most meaningful predictor of toxic behavior. And a 2022 interview study of Honor of Kings players, “After All, They Don’t Know Me,” found players themselves attributing their toxicity directly to anonymity. The players know. They’ll tell you.

The last online-specific accelerant is what I called, in the original essay, “the other player as obstacle.” Online competitive design reduces a human being to a number on a scoreboard who is standing between you and a rank. You are not cruel to your jungler because you’re cruel. You’re cruel because the interface never showed you a person in the first place.

Part 3. How game design amplifies toxicity

A row of ranked vertical bars, the single shortest one in coral, marked out from above.

Here’s where it stops being psychology and starts being my responsibility. The biases are constants. Design decides whether they fire.

Competitive ranked ladders. A visible, public skill number makes every loss a documented demotion. The frustration-aggression literature predicts this directly: losing produces aggression. Kwak, Blackburn, and Han’s landmark CHI 2015 study, built on a Tribunal dataset of “over 10 million user reports involved in 1.46 million toxic players and corresponding crowdsourced decisions” in League of Legends, found that reports overwhelmingly came from losing teams. The win rate of accused “intentional feeders” was under 15%, far below the 50% a fair matchmaker targets. Losing and being blamed are the same event.

Match length. This was my own pet observation in the first essay, and the data supports the intuition. The Activision/Caltech study (Morrier, Mahmassani, and Alvarez, PLOS ONE, June 2025), built on proprietary Call of Duty: Modern Warfare III data, found that exposure to toxicity measurably depresses short-term engagement, and that the effects vary sharply by match outcome: toxicity bites hardest after a loss. A forty-minute match you knew was lost at minute ten is thirty minutes of compounding negative affect with a human target in chat range.

Blame-assigning interfaces. This is the one I think designers underrate most. A death recap that names your killer, a scoreboard that ranks teammates by performance, a post-match screen that shows who had the worst stats. None of this is neutral information. It is blame infrastructure. It takes the diffuse feeling “we lost” and resolves it into “this specific named human lost it.” Kwak and Blackburn’s linguistic analysis (2014/2015, drawn from over 590,000 toxic cases) found that complaints and insults about teammates’ performance were a defining feature of toxic matches. The scoreboard hands you the target and the ammunition.

Forced interdependence. Riot’s own behavioral team has described League as “a high stakes game where success is heavily dependent on the actions of a team of (in most cases) strangers,” and called that “the perfect recipe for fun and competition, but unfortunately, it’s also the perfect recipe for folks trying to hurt each other.” When one stranger’s mistake can erase your effort, every stranger becomes a liability you’re entitled to police.

Open communication tools. Free text and open voice are harassment channels with a thin coat of utility. The Kwak/Blackburn work tracked how toxic language escalates as the match progresses. And toxicity is contagious. Shen and colleagues’ “Viral Vitriol” study of World of Tanks (2020) and the 2025 Activision study both find that exposure to a teammate’s toxicity raises the probability that you become toxic in the same match. One flamer is a vector, not an isolated case.

Zero-sum framing. When the only outcome states are “I win, you lose,” every other player is in your way by definition.

The case study I used before still says it best. Compare three games by asking one question: when you lose, where can the blame go? In League of Legends it goes to a teammate, a human target, maximum toxicity. In chess it goes to you and only you. No scapegoat, zero interpersonal toxicity, but a brutal ego cost. Tellingly, the main toxicity left in online chess is accusing the opponent of cheating, the last available blame-shift. In Hearthstone it goes to the dice: “I got unlucky” protects the ego without creating a human victim, which is part of why card-game lobbies stay calmer. The available blame target is a design choice, and it sets the toxicity dial.

Part 4. How game design reduces toxicity (the part with evidence)

Scattered dots gathered and merging into one solid coral ring at the center.

The good news: every studio that took this seriously found the same thing. You can move the numbers. Here are the interventions that actually worked, with the receipts.

Hide or soften the blame signal, and shield players from the toxic channel. This is the single most striking result I found, and the one most relevant to a small game. In 2012, Riot’s player-behavior lead Jeffrey Lin (a PhD cognitive neuroscientist, “Dr. Lyte”) ran the company’s first behavioral experiment: cross-team “all” chat, a primary venue for between-team flaming, was switched to opt-in, off by default. In Lin’s own account, cross-team chat had appeared in about 54% of games. After the change, in his words, “the number of games that had cross-team chat only dropped 1% to 53% of all games and we did not see the cross-team chat toxicity shift to any of the other chat channels,” and roughly 79% of players opted back in, yet Riot still saw “meaningful drops in number of ‘Offensive Language,’ ‘Verbal Abuse,’ and ‘Negative Attitude’ complaints.” Overall conversation did not decline (46 to 47% of games had no chat both before and after), so positive talk wasn’t suppressed. Lin’s conclusion: shielding players from toxic behavior prevents it from spreading. (Riot eventually removed /all chat from matchmade games entirely in patch 11.21, in 2021, saying “right now negative interactions outweigh the positives.”) The lesson isn’t “remove chat.” It’s “change the context and the default audience,” and toxicity falls more than usage does.

Prime the norm. Remember the SIDE finding, that anonymity amplifies the salient norm. Riot tested this literally, in the “Optimus” experiment (GDC 2013): they varied the color and content of loading-screen tips across millions of games. The results were real but double-edged. A negative-behavior message shown in red on the loading screen produced a larger drop in toxicity than the same message in white, while a sportsmanship question shown in red actually increased negative attitudes, verbal abuse, and offensive language. Priming works, including in the wrong direction. The norm you set in the first ten seconds is not decoration; it’s an input.

Reward good behavior instead of only punishing bad. Two big natural experiments agree here. League’s Honor system (introduced 2012, reworked 2017 and again 2025) lets players commend teammates and now opponents, tying perks to a reputation that climbs with good conduct. Overwatch’s endorsement system (2018) is the cleanest data point. Blizzard research scientist Natasha Miller reported at GDC 2019 that “there had been a 40 percent reduction in matches that contained disruptive behaviors since implementation,” with 50 to 70% of players handing out endorsements (the Overwatch director had earlier cited a 26.4% drop in competitive matches containing abusive chat within the first month). The design insight Miller stressed was redemption: if you slip, your level drops, which makes you want to earn it back. Carrot, not just stick.

Match by behavior, not just skill. Dota 2’s Behavior Score (a hidden 1 to 12,000 conduct number, with a separate Communication Score added in 2023) sorts players into pools by conduct, so well-behaved players mostly meet each other and chronic offenders mostly meet each other. It’s not perfect. Players complain loudly that the communication-mute system mislabels normal callouts and gets gamed by mass-reporting. But the structural idea, that your conduct shapes your matches, is sound, and it makes good behavior self-reinforcing.

Restrict the communication surface. This is where small games can win cheaply. Hearthstone ships with no text chat between strangers, only six preset emotes per hero, plus a one-click “Squelch” to mute an opponent. Designer Ben Brode and director Eric Dodds have said an early “Lucky” emote was cut because “it was used for evil more than good.” The honest caveat: it doesn’t eliminate toxicity (players still weaponize “Well played” sarcastically; the community calls it “BM,” bad manners, studied academically by Arjoranta and Siitonen in 2018), but it caps the ceiling of harm. The worst thing a Hearthstone opponent can do to you is spam a cartoon voice line you can mute in one click. Compare that to open voice chat.

Rocket League shows the same principle and its limit. Its quick chat system (preset phrases like “Nice shot!”, “What a save!”, “Thanks!”) was built so players communicate without typing, but players famously weaponize “What a save!” spammed sarcastically to tilt opponents after a goal. Psyonix’s answer was control rather than removal: via the Chat settings you can disable Quick Chat, Match Chat, and Party Chat independently, or mute individuals. The takeaway is that even a restricted vocabulary will be turned into an insult, so pair limited comms with frictionless, per-player muting.

Frame it cooperative and remove the individual scoreboard. This is the design that beats psychology instead of fighting it. Deep Rock Galactic (Ghost Ship Games) is the proof: “Overwhelmingly Positive” on Steam, with 97% of its 170,000+ English-language reviews positive (and 376,000+ across all languages), and a community famous for kindness despite a co-op design where one player’s mistake can wipe the team. How? Everyone wins or loses together, with no humiliating individual scoreboard to assign blame from. The four classes only succeed interdependently. Communication leans on pings instead of open chat. And the game gives you a one-button, almost ritual way to be nice, the “Rock and Stone!” cheer mapped to its own key, which turns encouragement into a reflex. The blame target is removed by construction: there’s no scoreboard line that says you carried, so there’s no scoreboard line that says they didn’t.

Or remove the human target entirely. The purest version is thatgamecompany’s Journey (2012), designed by Jenova Chen. You meet one other real player in the desert; you cannot see their name, gamertag, or chat; you can only emit a wordless musical chime; names are revealed only after the credits. Chen’s stated goal was “to show the positive side of humanity.” With no way to insult and no identity to attack, players overwhelmingly help each other. The studio carried the philosophy into Sky: Children of the Light (2019). Anonymity, the thing Suler blamed for toxicity, becomes prosocial the instant you remove the tools of cruelty and set a gentle norm, exactly what SIDE predicts.

One more sobering data point to keep us honest. The Anti-Defamation League’s ongoing research found, in its Hate is No Game 2023 report (fielded with Newzoo across 1,971 US gamers aged 10 to 45), that “76 percent of adult participants (aged 18-45)” and “75 percent of teens (ages 10-17)” reported experiencing harassment in online multiplayer games; its 2025 follow-up found hate or harassment in nearly half of the multiplayer sessions its researchers tested. And the original villain narrative is wrong: Riot’s Lin found that 87% of abuse comes from otherwise neutral or positive players having a bad day, while persistently bad-behaving players produced only about 13% of the harassment. You’re not designing against monsters. You’re designing against ordinary people, including past-me, on their worst fifteen minutes.

Part 5. The solo dev takeaway: concrete principles

I’m one person building small things. I can’t run a 24/7 trust-and-safety team or train a voice-moderation model like Activision’s ToxMod. But the research above is almost entirely about structure, and structure is exactly what a solo dev controls completely and for free. Here’s what I’m actually taking to the prototype.

  1. Decide the blame target on purpose. Before anything else, ask my own question: when a player loses, where does the blame go? Design the answer. Route it to variance, to the environment, to “us together,” anywhere but a single named human. If I ever put a competitive scoreboard in, I’m choosing where blame lands whether I mean to or not.

  2. Default to cooperative or solo framing; make “multiplayer” enrich, not depend. This is the same conclusion I reached about critical mass in the original essay, and it doubles as the strongest anti-toxicity move available. Deep Rock Galactic and Journey aren’t kind by accident; they’re kind by architecture. In Dashline, the other players are obstacles by design, so the antidote is short 90-second rounds (frustration can’t compound) and bots filling empty arenas (no cold-start rage).

  3. Kill the individual blame scoreboard, or hide it. If players need stats, show them team outcomes or private personal stats, not a public ranking that lets four people identify the “worst” one. The death recap that names a culprit is the most toxic UI element in gaming, and I’ll think twice before adding its equivalent.

  4. Restrict the communication surface, and make muting one click. Default to pings, presets, and emotes over free text and open voice. Borrow Hearthstone’s lesson (cap the ceiling of harm) and Rocket League’s correction (let people mute an individual instantly). If I can’t moderate it, I shouldn’t ship it open.

  5. Set the norm in the first ten seconds, and reward the behavior I want. SIDE says anonymity amplifies the salient norm, so I get to choose the norm. A one-button “nice” gesture (“Rock and Stone!”), a visible commend or endorsement that grants small perks, a warm tip on the loading screen: these are nearly free, and they measurably moved Blizzard’s and Riot’s numbers. Carrot beats stick; reward beats punishment.

  6. Shorten matches and soften losses. Long, foregone-conclusion matches are toxicity incubators. Short rounds, surrender options, and comeback mechanics all reduce the dose of negative affect that has to go somewhere.

  7. Remember it’s mostly good people having a bad day. Design for redemption (let reputation recover), not just for banning. The relentless minority you can’t fix; the 87% you can nudge.

A good game, I wrote last time, meets a real human need cleanly. Connection is one of those needs, Self-Determination Theory’s “relatedness.” Toxicity is what happens when you offer people connection and then hand them a knife and point at a target. The fix isn’t to lecture players into being nicer. It’s to stop handing out the knife. That part, at least, is entirely within one developer’s control.

Sources

Grouped by theme. Where a figure comes from a company about itself (Riot’s complaint drops and 87%/13% split, Blizzard’s 40%, Dota’s behavior-score claims), I’ve flagged it in the text. Treat those as directional, and the peer-reviewed studies and the ADL survey as the firmer ground.

Psychology of blame, attribution, and aggression

Online disinhibition, anonymity, deindividuation

League of Legends toxicity data and Riot’s interventions

Overwatch, Dota 2, Valorant, Activision

Restricted-communication and cooperative/positive designs

Industry data on harassment

A note on the numbers: several of the strongest figures here are self-reported by the companies that benefit from them (Riot’s complaint drops and 87%/13% split, Blizzard’s 40%, Dota’s behavior-score claims). I’ve flagged those in the text. Treat company stats as directional evidence, not independent proof, and treat the peer-reviewed studies (Breuer; Kwak/Blackburn/Han; Morrier/Activision; Shen; Arjoranta/Siitonen) and the ADL survey as the firmer ground.

#game-design#player-psychology#toxicity#multiplayer#solo-dev
← Older
Dashline: the first prototype, against my own advice