Cognitive Biases in Games

Summary

Players do not make decisions as rational actors who enumerate all options and select the one that maximises expected value. They use mental shortcuts (heuristics) that allow fast decision-making at the cost of systematic, predictable errors called biases. Hiwiller (Players Making Decisions, Chs. 26–27) applies decades of behavioural economics and cognitive psychology research — primarily Kahneman and Tversky — directly to game design: these biases are not curiosities to be aware of, but levers a designer can pull deliberately or, if not careful, trigger accidentally.

(Hiwiller, Players Making Decisions, see source-players-making-decisions)

Attribution errors

The fundamental attribution error

The fundamental attribution error is the tendency to explain other people’s behaviour using dispositional factors (their personality, skill, or intent) while explaining one’s own behaviour using situational factors (circumstances, chance).

When a player loses to an opponent, they tend to attribute the opponent’s success to some inherent advantage or game-breaking strategy (dispositional). When they win, they attribute it to their own skill (situational: “I played that well”). This creates an asymmetry: in multiplayer games, the same outcome reads as unfair imbalance to the loser and well-earned victory to the winner.

Design implication: if a player loses, the designer wants the player to attribute the loss to their own play decisions, not to the game system or to chance. When a player blames the game, they devalue the experience and are more likely to churn. Legible feedback (clear cause-and-effect between actions and outcomes) shifts attribution back toward the player’s own choices.

The self-serving bias

The self-serving bias is the tendency to attribute successes to skill and failures to bad luck or external factors. Players who win coin-toss-equivalent events believe they are skilled; players who lose the same events believe the random number generator is broken.

This is directly relevant to Sid Meier’s Civilization: when players with a 3:1 army advantage lost, they blamed the game. When they won against a 3:1 disadvantage, they praised their own strategy. The same odds produced different attribution in the two conditions (see also randomness-in-games).

Misunderstanding randomness

The gambler’s fallacy

The gambler’s fallacy is the belief that random processes “remember” their history and will self-correct. If a coin has come up heads five times in a row, the gambler’s fallacy says tails is now “due.” Statistically, each flip is independent; the coin has no memory.

In game design: if a player has missed the 25% MegaBoss encounter fifty times in a row, the gambler’s fallacy suggests it is “due.” It is not — the 25% probability applies independently to each session.

Design implication: players will routinely misinterpret independent random events as part of a pattern. When players lose several times in a row on an objectively fair system, they will perceive the system as broken. Options:

1. Implement a pity timer (guaranteed drop/event after N failed attempts) — this actually does violate independence, deliberately, to match player expectation
2. Communicate probability clearly with visible history (“25% chance each encounter”)
3. Accept that some players will always perceive fair systems as unfair

Hot hand fallacy

The hot hand fallacy is the belief that a player or system on a “streak” is more likely to continue succeeding. Research shows that hot-hand streaks in sports are statistical artefacts of random sequences — the human mind patterns-matches on noise.

In games: a “hot hand” mechanic (increasing hit rate after consecutive successes) is not modelling a real statistical phenomenon; it is a deliberate design choice to add momentum and risk/reward. Use Monte Carlo simulation (see probability-for-designers) to verify that the mechanic does not converge to extreme values.

Compound probability errors

Players consistently underestimate the combined rarity of compound events. The “Billion Dollar Bracket” (correctly predicting all 63 NCAA tournament games) seems achievable to most players — the odds are 1 in 9.2 quintillion. Streak-for-the-Cash (getting 20 correct sports picks in a row at 50% each) seems achievable — the odds are roughly 1 in 1,000,000.

Players anchor on the simple version of each event (one game = 50/50) and fail to appreciate how rapidly compound probabilities shrink. This bias directly affects:

• Perceived difficulty of achievement rewards
• Player willingness to attempt skill-based long-shot events
• The “feel” of loot drops with compound rarity conditions

Anchoring and adjustment

Anchoring is the cognitive tendency to over-weight the first number encountered when making an estimate, and to adjust insufficiently away from it.

Researchers found that writing down the last two digits of one’s Social Security number before bidding on items in an auction affected how much participants bid. High numbers → higher bids; low numbers → lower bids. Even implausible anchors (“Is a whale larger or smaller than 0.2 m?”) affect subsequent estimates.

Application to virtual currency design: if a game presents its most expensive currency bundle first ($9.99),subsequentbundlesfeelcheapbycomparison.Ifitpresentsthecheapestbundlefirst($0.99), the most expensive bundle feels expensive. The same prices, presented in reverse order, produce different purchase behaviour.

Even including a bundle the designer does not expect anyone to buy (e.g., a $99.99"founde{r}^{\prime }spack")anchorsallplayerstoahigherperceivedpricerange,makingthe$9.99 bundle feel like a reasonable middle option.

Hiwiller explicitly flags this as a technique that is “used nefariously quite often in games.” Compare dark-patterns for the ethical framing.

Expected value, risk, and diminishing returns

Expected value

Expected value is the probability-weighted average outcome of a decision. For a 50/50 coin flip that pays $3forheadsand$0 for tails: EV = 0.5 × $3+0.5\times$0 = $\mathrm{1.50.}Arationalexpected-valuemaximiserwouldpayupto$1.50 to play.

Expected value is a practical tool for balancing in-game choices: if two paths have equal expected value, they are roughly interchangeable and neither is dominant. Significant EV imbalances signal dominant strategies.

The St. Petersburg Paradox and diminishing marginal utility

The St. Petersburg Paradox (Nicolas Bernoulli, 18th century): a game that pays $2^n where n is the number of flips before the first tail has infinite expected value, yet no reasonable person would pay more than a few dollars to play it.

The resolution (Daniel Bernoulli, 25 years later): people do not value all gains equally — the marginal utility of a reward diminishes as you already have more. The first $100inastarvingperso{n}^{\prime }spocketisworthmorethanthethousandth$100 in a billionaire’s.

Design implication: as players accumulate resources or power, each additional unit is worth less to them. This is the fundamental mechanism behind:

• XP curves that require exponentially more XP per level (each level is worth less relative to total effort)
• Resource saturation in economy games
• Hedonic fatigue in reward systems (see progression-and-power-curves)

Risk aversion and risk-seeking

People are not uniformly risk-neutral. They are risk-averse in some situations (pay 100 USD for insurance against a 1% chance of losing 1,000 USD, despite negative EV) and risk-seeking in others (buy a lottery ticket despite heavily negative EV).

The pattern, identified by Kahneman and Tversky: people tend to be risk-averse when the stakes involve potential gains (prefer a certain smaller gain to a risky larger one) and risk-seeking when the stakes involve potential losses (prefer a risky chance of no loss to a certain smaller loss).

In games: high-risk, high-reward mechanics attract risk-seekers; safe, predictable progression appeals to risk-avoiders. Designing for both requires offering both paths simultaneously.

Loss aversion and the endowment effect

Loss aversion (Kahneman and Tversky): people feel losses approximately twice as intensely as equivalent gains. Losing $20producesroughlythesameemotionalimpactasgaining$40 would.

The endowment effect: people value things more once they already own them. The Honda/Mini Cooper car dealership story: giving a customer the car to take home for a few days made them feel the loss of returning it more acutely than they had previously desired to own it.

Game applications — intended:

• Metroid Prime: the player is given all weapons at the start, then they are taken away. This loss motivates seeking them out more powerfully than never having had them would.
• Equipment durability systems: losing a cherished item feels worse than the cost of repair suggests.
• FarmVille’s withering crops: crops that die feel like a loss compared to not planting (even though there is no net cost to not having planted). The mechanic drives return visits.

Game applications — exploitative (dark patterns):

• “Your offer expires in 24 hours” — framing an expiring sale as a potential loss rather than a discount drives urgency
• Loot box content reveals that nearly give a rare item — the sense of near-miss activates loss-aversion responses disproportionate to the actual outcome
• Starter boosts that are later removed (first-day double XP that expires)

See dark-patterns for the ethical analysis of these patterns.

“Rather than lose a turn, try to gain a turn. […] The happier everyone is, the less likely they’ll never want to play again.” — Ray Mazza (quoted in Hiwiller, Ch. 26)

Framing decisions

The same choice, described differently, produces measurably different behaviour.

• “Should the US forbid public speeches against democracy?” — 46% agreed.
• “Should the US allow public speeches against democracy?” — 62% said no (logically identical to the above).
• A 50 USD penalty for late registration → 93% registered early.
• A 50 USD discount for early registration → 67% registered early (the same deadlines and prices).

Design applications:

• Framing a decision as “what will you lose if you don’t act?” is more motivating than “what will you gain if you act?”
• Presenting upgrades as “preventing degradation” rather than “improving from baseline” feels more urgent
• Showing elapsed play time on quitting (as Steam does) makes players retrospectively rate the experience more highly than prompting them to consider money spent
• Pricing tiers presented high-to-low anchor the player to the top price; presented low-to-high, the top price feels expensive

“By framing your game’s decisions correctly, you can subtly direct players toward options of your choice.” — Hiwiller, Ch. 26

Attention

Attention limits and the multitasking myth

Human attention is selective and limited. Multitasking is a myth: the brain can consciously process one thing at a time; apparent multitasking is rapid task-switching with a measurable performance cost. Frequent task-switching makes players worse at each individual task and more susceptible to distraction.

Design implication: do not require players to simultaneously attend to multiple independent streams of information. The most critical information should occupy the primary attention channel; secondary information should be available on demand rather than always present.

Attention misdirection

Designers can direct attention by controlling what is salient (lit, moving, loud, novel) and what is not. BioShock’s infamous dentist scare works precisely because it misdirects player attention to a distant fog (expected enemy direction), then delivers the threat from close range. Haunted houses apply the same principle by design: light and sound draw attention to a “look here” element, then the scare comes from the opposite direction.

Negative application: players who are attending to the action are not reading on-screen text. UI text is often functionally invisible to players who are in-game.

Attention direction

Positive attention direction guides players to goals without breaking immersion. Halo’s original level designers used lit arrows on floors to direct players through labyrinthine repeated-element levels. Lighting, paths, colour, sound effects, and orientation can all subtly guide player attention. See player-guidance for the full toolkit.

Memory

Working memory capacity

Human working memory holds approximately four chunks — not seven, as the 1956 “magical number seven” paper suggested. More recent research (Cowan, 2001) places the actual limit at four items.

Design implication: when teaching players a new character, mechanic, or level, present no more than four new concepts simultaneously. League of Legends addresses this explicitly: most characters have exactly four unique abilities, keeping the new-character learning cost within working memory limits.

If a player has not yet reached automaticity with existing game concepts, introducing four new ones compounds the load. New mechanics introduced mid-game must account for the existing cognitive load.

Chunking

Information grouped into meaningful chunks is easier to retain. “GAMEDESIGNLIFE” (three meaningful words) is remembered easily; “HKNLZAMDKJPQBXL” (same length, meaningless) is not. This is why teaching game rules through categories and meaningful relationships aids retention more than presenting rules as a flat list. See chunking and fun-as-learning.

Memory aids for tutorial design

Hiwiller’s recommendations for working with memory limitations (Ch. 27):

• Organisation: group related concepts hierarchically — subjects remember hierarchically organised word lists 3.5× better than random lists
• Meaning: use established symbols and terms rather than inventing new ones; familiar chunks allow new concepts to build on existing memory
• Order: information presented first (primacy effect) is remembered more strongly when immediate recall is required. Use primacy for tutorial design — put the most important information first.
• Environment: content is best recalled in the environment where it was learned. Teach mechanics in the context where they will be used, not in separate tutorial screens.
• Repetition: use it to move information from short-term to long-term memory. Having players immediately perform what they just learned is more effective than re-reading.
• Emotion: emotionally charged experiences are retained disproportionately well. Narratively meaningful tutorial moments are retained better than dry instruction.

Text is often invisible

The most common tutorial failure: adding hint text to the screen and assuming players will read it. Players attending to game action cannot simultaneously focus on adjacent text; the attention switch has a cost and players avoid making it. The more text onscreen, the less any of it is read.

Better approach: teach through play, not through text. Let players perform the action while learning it; introduce mechanics through environmental design rather than written explanation.

Colorblindness accommodation

1 in 12 men and 1 in 200 women are colorblind; red-green is the most common form. Games that use colour as the only differentiator between game elements exclude a significant portion of their audience.

Ticket to Ride is the positive example: both colours and symbols differentiate train cards and routes, so colorblind players can distinguish elements without colour. The additional design cost is minimal; the accessibility impact is substantial. See accessibility-and-localisation for the full accessibility framework.

In practice

A checklist for applying bias awareness to game design:

• Is failure clearly attributable to the player’s own choices? (attribution error mitigation)
• Are probabilities clearly communicated, especially for compound events? (gambler’s fallacy mitigation)
• Have you verified that random outcomes don’t produce player-perceptible patterns that don’t exist? (hot hand fallacy)
• Is the first price anchor you show the one that makes your other prices look most reasonable? (anchoring)
• Have you mapped loss-aversion opportunities: what do players stand to lose, not just gain?
• Is the same choice framed as loss avoidance wherever appropriate?
• Does any single tutorial screen require learning more than four new concepts?
• Is the most critical mechanic introduced early and with repetition?
• Are gameplay-critical distinctions communicated through more than just colour?

Open questions

• Loss aversion is a powerful tool and an easy path to exploitative design. Is there a principled distinction between using loss aversion to create meaningful stakes versus using it to manufacture artificial urgency?
• The gambler’s fallacy and the hot hand fallacy push in opposite directions: pity timers (satisfying the gambler’s fallacy) vs streak systems (satisfying hot hand intuitions). Which should be implemented depends on the game’s progression goals — but how does a designer know which player perception is dominant for a given mechanic?
• Attention misdirection requires knowing where the player will be looking. For open-world or sandbox games, designers cannot reliably predict attention direction. How do these techniques apply in unconstrained contexts?

Related

• randomness-in-games — Skill/luck spectrum; mitigation techniques; fairness perception; gambler’s fallacy context
• probability-for-designers — The mathematical grounding for probability claims; Monte Carlo simulation tools
• meaningful-decisions — The anatomy of a choice; how biases interact with decision quality
• reward-systems — Reinforcement schedules; how variable reward exploits anticipation and uncertainty
• dark-patterns — Exploitative applications of anchoring, loss aversion, and framing
• player-guidance — Attention direction techniques in level design
• fun-as-learning — Chunking, grokking, and the cognitive model of learning through play
• game-feel — Feedback timing as attention-management: when feedback arrives determines what the player attends to
• accessibility-and-localisation — Colorblindness, readability, and perceptual accommodations
• self-determination-theory — Autonomous motivation vs externally controlled; loss aversion can undermine autonomy
• source-players-making-decisions