Assigning more blame for identical behavior when the outcome happens to be worse, even though the decision was the same.
Imagine two kids throwing a ball in the house. Both throw it the exact same way. One kid's ball bounces harmlessly off a pillow, and the other kid's ball breaks a fancy vase. Even though they did the exact same thing, the kid who broke the vase gets in way more trouble. We punish people based on how bad things turned out, not just on what they actually did.
Severity Bias describes the systematic distortion in moral and causal judgment whereby the seriousness of a negative outcome inflates perceived blame, negligence, and deserved punishment for the person involved, independent of their actual intentions, knowledge, or decision quality. When an identical risky action results in a catastrophic outcome versus a minor or neutral one, observers retroactively judge the actor as more reckless, less competent, and more deserving of sanction — even though nothing about the actor's behavior differed. This bias operates as a powerful confound in legal, medical, and organizational settings, where identical conduct can be deemed acceptable when outcomes are benign but negligent when outcomes are tragic. The effect is robust across contexts and persists even among professionals trained to separate process from outcome.
The same glitch looks different depending on the terrain. Finance, medicine, a relationship, a team — same mechanism, different costume.
Investors and regulators judge fund managers as more negligent or incompetent when risky but reasonable investment decisions result in large losses versus small gains, even when the strategy was identical. Financial advisors face lawsuits not for the quality of their advice but for the magnitude of losses that luck produced.
Physicians face dramatically different evaluations of identical clinical decisions depending on patient outcomes. Doctors who make standard-of-care decisions are judged as negligent when patients die or suffer severe complications but competent when patients recover — driving defensive medicine and over-testing to avoid post-hoc blame.
Elaine Walster (1966) first demonstrated the severity-responsibility link in her seminal paper 'Assignment of Responsibility for an Accident' in the Journal of Personality and Social Psychology. The concept was further developed by Kelly Shaver (1970) through the defensive attribution hypothesis and synthesized in Robbennolt's (2000) meta-analytic review of outcome severity and responsibility judgments.
In ancestral environments, severe outcomes — a death, a serious injury, a destroyed food cache — demanded swift identification of responsible agents to deter future dangerous behavior. Erring on the side of harsher blame for worse outcomes served a protective function: it discouraged risk-taking that could threaten group survival and signaled strong norms against carelessness. The cost of under-blaming a genuinely negligent actor (allowing future harm) was far greater than the cost of over-blaming an unlucky one.
Machine learning models trained on human judgment data (legal sentencing datasets, performance reviews, incident reports) inherit severity bias by learning to weight outcome severity as a predictor of negligence or fault. AI systems used in legal or insurance contexts may systematically assign higher risk scores or blame assessments to actors associated with severe outcomes, perpetuating the bias at scale without the possibility of case-by-case human correction.
This page taught you the name. The deck turns the name into reflex. 1,100+ swipeable scenarios, 1,100+ defenses, 650+ detection prompts — spaced-repetition Swipe Deck, unlimited Spot-the-Bias Quiz, Defense Playbook, Pre-Flight, My Blindspots, Cheat Sheets, Field Guide e-book. $39.53$59.
