Omission Bias¶

Prime #: 1033
Origin domain: Cognition And Psychology
Subdomain: judgment and decision making → Cognition And Psychology

Core Idea¶

Omission bias is the pattern in which, in situations where the expected outcome of doing something is no better than the expected outcome of doing nothing — and sometimes worse — decision-makers systematically prefer inaction, and judge a harm that results from acting as worse than an equal or even greater harm that results from failing to act. It is the mirror image of action bias: where action bias tilts an accountable agent toward doing something because action is more visible and creditable, omission bias tilts the agent toward forbearance because commission is more culpable — a harm one caused by acting feels worse, and is judged and blamed more harshly, than the same harm one merely allowed by not acting. The bias has two reinforcing sources. First, an asymmetry in causal attribution: an act is a salient intervention into the course of events, readily traced to the actor as its cause, whereas an omission blends into the background flow of what would have happened anyway and is harder to pin on the agent. Second, an asymmetry in counterfactual and emotional regret: the harm caused by acting invites a vivid "if only I had not done that," while the harm from omission invites the weaker "if only I had done something," so anticipated regret pulls toward not acting. The result is a non-symmetric preference for inaction that persists even when the underlying expected-outcome calculus is symmetric or favors acting.

The structure has a fixed shape. There is a decision between acting and not acting whose expected consequences are similar, or where action is at least competitive. There is an evaluation environment in which the agent will be held responsible for outcomes — by others or by their own conscience. There is an asymmetry: a bad outcome following an act is more readily attributed to the agent, and judged more culpable, than the same bad outcome following an omission. The consequence is a bias in choice toward inaction that exceeds what the expected-outcome calculus would justify. The essential commitment is that the bias lives in the evaluation layer, not the decision substrate: the consequences of acting and not acting may be equivalent, but they are evaluated differently — the act-caused harm carries an extra culpability penalty the omission-allowed harm does not — and that differential evaluation, not any difference in expected value, drives the choice. This locates a correctable layer: making the agent equally accountable for the consequences of inaction — treating a foreseeable harm one failed to prevent as one's responsibility on a par with a harm one caused — reduces or removes the bias, because it removes the attributional and culpability asymmetry that produced it.

How would you explain it like I'm…

Better Not To Touch

Imagine two ways your toy could break. If YOU push it off the table, you feel really bad. If you just watch it fall and don't catch it, you feel a little bad — even though the toy broke the same either way. Omission bias is when 'I did it' feels much worse than 'I let it happen,' so we'd rather just sit still and do nothing.

Doing-It Feels Worse

Omission bias is our habit of preferring to do nothing, even when doing something would turn out just as well or better. The reason is that a bad result we CAUSED by acting feels worse — and gets blamed more — than the same bad result we merely ALLOWED by not acting. Two things drive it: an action is easy to trace back to you as the cause, while not-acting blends into 'what would have happened anyway'; and acting invites a stronger regret ('if only I hadn't!') than not-acting does. So even when the actual outcomes are a tie, we lean toward sitting still. The fix is to hold yourself just as responsible for the harm you fail to prevent as for the harm you cause.

The Inaction Tilt

Omission bias is the pattern where, when the expected outcome of acting is no better than doing nothing — and sometimes worse — people systematically prefer inaction and judge a harm caused by acting as worse than an equal or greater harm caused by failing to act. It's the mirror of action bias: where action bias pulls toward doing something because action is visible and creditable, omission bias pulls toward forbearance because commission feels more culpable. Two reinforcing sources drive it: an asymmetry in causal attribution (an act is a salient intervention easily traced to you, while an omission blends into the background flow), and an asymmetry in regret (the harm from acting invites a vivid 'if only I hadn't,' while harm from omission invites only the weaker 'if only I had done something'). The key insight is that the bias lives in the EVALUATION layer, not the actual consequences — the outcomes may be identical, but the act-caused harm carries an extra culpability penalty the omission-allowed harm doesn't. That locates the fix: make people equally accountable for the consequences of inaction, and the bias shrinks.

Omission bias is the pattern in which, when the expected outcome of acting is no better than doing nothing (and sometimes worse), decision-makers systematically prefer inaction and judge a harm resulting from acting as worse than an equal or greater harm resulting from failing to act. It is the mirror image of action bias: action bias tilts an accountable agent toward doing something because action is visible and creditable, whereas omission bias tilts toward forbearance because commission is more culpable — a harm one caused by acting is judged and blamed more harshly than the same harm one merely allowed. Two reinforcing sources produce it. First, an asymmetry in causal attribution: an act is a salient intervention readily traced to the actor, whereas an omission blends into the background flow of what would have happened anyway. Second, an asymmetry in counterfactual and emotional regret: harm caused by acting invites a vivid 'if only I had not done that,' while harm from omission invites only the weaker 'if only I had done something.' The essential commitment is that the bias lives in the evaluation layer, not the decision substrate: the consequences of acting and not-acting may be equivalent, but they are evaluated differently — the act-caused harm carries an extra culpability penalty that the omission-allowed harm does not — and that differential evaluation, not any difference in expected value, drives the choice. This locates a correctable layer: making the agent equally accountable for the consequences of inaction — treating a foreseeable unprevented harm as one's responsibility on par with a caused one — removes the attributional and culpability asymmetry, and with it the bias.

Structural Signature¶

a decision between acting and not acting with similar expected outcomes — an evaluation environment that holds the agent responsible — a culpability and causal-attribution asymmetry making commission more blameworthy than omission — a tilt toward inaction exceeding what the outcome calculus justifies — a separation of the decision substrate from the evaluation layer — a correctability invariant: making the agent accountable for omissions removes the bias

The pattern is present when each of the following holds:

A symmetric-outcome decision. A choice between acting and forbearing whose expected consequences are similar, or where action is at least competitive — sometimes better.
An accountability environment. A setting in which the agent is held responsible for outcomes, whether by external judgment (others, the law, the market) or internal judgment (conscience, anticipated regret).
A commission–omission asymmetry. A bad outcome following an act is more readily attributed to the agent as its cause, and judged more culpable, than the same bad outcome following an omission, which blends into the default course of events.
A biased tilt. A systematic preference for not acting that exceeds what the expected-outcome comparison would justify, including a willingness to accept a worse outcome from inaction over a better one from action when the action would make the agent the salient cause of any harm.
A two-layer separation. The bias lives in the evaluation layer (how options are judged and blamed), not the substrate layer (what the options actually produce); the consequences may be equivalent while the culpability evaluations differ.
A correctability invariant. Because the bias is produced by the attributional and culpability asymmetry, making the agent equally accountable for the consequences of inaction — counting an unprevented harm as the agent's responsibility on a par with a caused harm — reduces or removes it; the single correctable lever is the symmetry of accountability across action and inaction.

The components compose so that the distortion is located in the evaluation layer, not the decision substrate: the structure predicts the tilt from the culpability-attribution structure rather than the substance, predicts it is strongest exactly where the act would make the agent the salient cause of a harm, and identifies the correctable lever — make omissions as accountable as commissions.

What It Is Not¶

Not action bias. action_bias (the nearest embedding neighbor) is the mirror tendency — a tilt toward acting because action is more visible and creditable. Omission bias is the tilt toward not acting because commission is more culpable. The two are a context-dependent pair of opposing pulls in the same agent; they are not the same bias, and the corrective levers flip between them.
Not status-quo bias. status_quo_bias is a preference for the current state — keeping things as they are across any choice, including choices between two actions. Omission bias is specifically about the act-versus-refrain asymmetry: it penalizes the option that constitutes an intervention. The two often coincide (inaction usually preserves the status quo) but diverge when acting would preserve the status quo and not acting would change it — there omission bias still favors not acting while status-quo bias favors the action that holds the state.
Not loss aversion. loss_aversion weights losses more heavily than equal gains, independent of agency. Omission bias persists even between two equally-framed losses: a loss one caused by acting is judged worse than an equal loss one allowed by not acting. The extra weight comes from causal-and-moral attribution, not from the loss/gain frame.
Not bias in general. bias is the umbrella structure of a systematic deviation; omission bias is the specific attributional-and-culpability asymmetry between commission and omission in an accountable setting.
Not justified caution. A reasoned judgment that action carries genuinely worse expected outcomes (irreversibility, real downside risk) is not omission bias — it is correct decision-making. Omission bias is the tilt toward inaction that exceeds what the outcomes justify, operative precisely where the expected outcomes are similar or favor acting.
Common misclassification. Reading a refusal to intervene as a reasoned judgment that forbearance was superior. Catch it by asking whether the same choice would be made if a harm from inaction were attributed to the agent as squarely as a harm from action; if the tilt toward not acting vanishes once omissions are made equally accountable, it lived in the evaluation layer.

Broad Use¶

The pattern recurs across human decision-makers in accountable settings. In clinical medicine and public health it is the most-studied case: vaccination hesitancy in which a parent prefers to risk a worse outcome from a disease (a harm of omission, not their doing) over a far smaller risk from a vaccine (a harm of commission, their doing), and the broader clinical pull toward not intervening where a bad outcome after treating would invite blame that the same bad outcome after watchful inaction would not. In ethics and moral philosophy it is the structure behind the acts/omissions doctrine and the intuitions probed by trolley problems — the widespread sense that actively diverting harm onto one person is worse than passively allowing a greater harm to befall others, even when the outcomes favor acting. In law it is the deep and near-universal distinction between acts and omissions: legal systems punish killing far more readily than letting die, recognize a general duty not to harm but only narrow duties to rescue, and treat commission and omission as categorically different even at equal outcomes — the doctrine that encodes the bias into institutions. In policy and regulation it is the asymmetric scrutiny of errors of commission (a drug approved that causes harm) over errors of omission (a beneficial drug delayed or denied, whose victims are statistical and unattributed), biasing regulators toward inaction. In investing and finance it is the status-quo holding in which an investor prefers the regret of a loss they allowed by not rebalancing over the sharper anticipated regret of a loss they caused by an active trade. In management and parenting it is the reluctance to make a hard active call — a firing, a discontinuation, an intervention in a struggling report or child — when allowing the situation to deteriorate is less attributable than acting and getting it wrong. Across all of them the substrate is a human decision-maker in an environment of moral or reputational accountability, and the cross-domain reach is among such agents rather than to any non-human system.

Clarity¶

The concept makes visible the asymmetry in how commission and omission are evaluated even when their consequences are equivalent. Ordinary reasoning about a choice attends to the expected outcomes of the options; omission bias points instead at the evaluation layer, where a harm one caused by acting is scored as worse — more attributable, more culpable, more regretted — than an equal harm one allowed by failing to act. Naming the bias separates the substrate (what each option will actually produce) from the evaluation (how each option will be judged and blamed), which is the move that exposes the distortion: the agent is not choosing the better outcome, but the less-culpable role.

The clarifying force is also to distinguish omission bias from neighbors that share its surface. It is not loss aversion, which weights losses over gains regardless of agency; omission bias adds weight specifically to agent-caused harms, persisting even between two equal losses. It is not status-quo bias, which prefers the current state across any choice; omission bias is keyed to the act/refrain distinction and can diverge from the status-quo pull when acting is what preserves the state. And it is the mirror image of action bias — the preference for action even when inaction has equal or better expected value, typically because action is more visible and creditable — with the two applying in different contexts and together forming a context-dependent pair of opposing pulls. Holding these distinct lets an analyst attribute a refusal to intervene to the culpability asymmetry rather than to a genuine judgment that forbearance was superior, and lets them see that the vaccine-refusing parent, the trolley-problem intuition, and the regulator who delays a beneficial drug are running the same evaluation-layer asymmetry beneath very different surfaces.

Manages Complexity¶

The concept lets a decision-maker reduce a confused "should I really intervene?" question to a structured pair. First, what is the expected outcome of acting versus refraining — the substrate question. Second, and separately, am I being pulled toward not acting by the lesser culpability of an allowed harm rather than by its expected value — the evaluation question. Splitting the single muddled impulse into these two tractable questions is the compression: it isolates the part of the decision that is about the world from the part that is about how the decision-maker's role in any harm will be judged.

The compression also sorts the interventions, each targeting the culpability-attribution asymmetry. Counting omissions explicitly — making the foreseeable harms of inaction visible and tallied alongside the harms of action — strips the asymmetry by which omission-harms blend into the background. Reframing the default as a choice — presenting "do not vaccinate," "do not rebalance," "do not approve" as an active decision with its own attributable consequences rather than as a neutral non-event — re-credits the agent with responsibility for inaction. Symmetric-accountability rules — holding regulators answerable for the statistical victims of delay as squarely as for the victims of approval, or holding a clinician responsible for the harms of under-treatment as for over-treatment — remove the institutional asymmetry at its source. Pre-committing to outcome-based decision rules — "treat the expected outcomes as decisive regardless of which is the act and which the omission" — removes the choice from the moment of culpability-driven regret. Anticipated-regret recalibration — deliberately weighing the regret of a harm allowed as heavily as the regret of a harm caused — addresses the emotional source. Each lever makes inaction as accountable as action, which is the structural complement of the bias, and having the structure in hand is what makes the choice among them deliberate.

Abstract Reasoning¶

Holding omission bias as a unit permits the question: is my reluctance to act here a property of the expected outcomes or a property of how my role in any resulting harm will be judged? Would I make the same choice if a harm from my inaction were attributed to me as squarely as a harm from my action? These questions operationalize the core structural claim — that the bias sits in the evaluation layer — by giving the decision-maker a test that isolates the culpability-attribution pull from the genuine expected-value comparison.

The decisive inference the abstraction licenses is that the bias is predictable from the attribution structure rather than from the substance of the decision. Where a bad outcome following an act is more readily attributed to the agent than the same outcome following an omission, and the agent is held responsible, the abstraction predicts a systematic tilt toward inaction regardless of the domain, and predicts that the tilt will be strongest exactly where acting would make the agent the salient cause of a harm — because that is where the culpability asymmetry, rather than the outcome difference, decides. Reasoning from the pattern, an analyst can therefore anticipate the bias in advance from the accountability environment, can predict that it will not appear where omissions are attributed and judged as squarely as commissions, and can identify the single correctable lever: make the agent equally accountable for the consequences of inaction. This converts a vague injunction to "don't be too timid" into a structural prediction about which decisions, in which environments, will be over-tilted toward not acting — and explains the otherwise puzzling phenomenon of agents knowingly accepting a worse expected outcome to avoid being the cause of a smaller harm.

Knowledge Transfer¶

The structural roles map across human decision-making settings, and with them the interventions transfer intact. The symmetric-outcome decision corresponds to the vaccinate-or-not clinical choice, the divert-or-allow trolley case, the approve-or-delay regulatory decision, the rebalance-or-hold portfolio choice, the intervene-or-let-deteriorate management choice; the accountability environment to the setting in which outcomes are pinned to the agent by others or conscience; the commission–omission asymmetry to the greater attributability and culpability of a caused harm over an allowed one; the biased tilt to the over-preference for inaction; the correctable layer to the symmetry of accountability across action and inaction. Because the roles correspond, a physician who recognizes the bias in vaccine hesitancy sees it in a regulator's delay or an investor's failure to rebalance without retranslation.

The interventions inherit that portability. Counting omissions explicitly is one move whether realized as a clinical tally of the harms of under-treatment, a regulatory accounting of the statistical victims of delay, or an investment review that scores the cost of not rebalancing — each makes the harms of inaction visible and attributable. Reframing the default as a choice recurs across medicine, policy, and finance, presenting forbearance as an active, owned decision rather than a neutral non-event. Symmetric-accountability rules and outcome-based pre-commitments transfer with the same rationale wherever an accountable agent faces a symmetric act/refrain choice. The transfer is real but bounded: because omission bias is a cognitive-bias category inherently bound to human decision-makers and their causal-attribution and culpability machinery, what travels is a repertoire for equalizing the accountability of action and inaction, and the reach is among human-agent settings rather than to non-human substrates — there is no commission/omission asymmetry for a thermostat or an optimizer, because neither can be held morally responsible for what it caused versus allowed. Within that bound the structure is portable — a culpability-and-attribution asymmetry between commission and omission that biases choice toward inaction — and a reader presented with that bare description recognizes their own reluctance to be the cause of harm without the vaccine example, paired in their mind with its mirror, action bias, as the opposite pull in a different context.

Examples¶

Formal/abstract¶

Model the choice with an expected-utility decision augmented by an evaluation layer, exactly mirroring the action-bias model with the sign of the asymmetry reversed. An agent faces two options — act (\(A\)) or forbear (\(F\)) — with substrate payoffs \(U(A)\) and \(U(F)\). In the bias-relevant regime these are similar, \(U(A) \approx U(F)\), or \(A\) is even slightly better. A rational agent maximizing substrate utility would be indifferent or lean to \(A\). But the agent is held responsible, and the evaluation layer adds a culpability penalty \(\gamma\) to the option that makes the agent the salient cause of any resulting harm — the act: effective \(V(A) = U(A) - \gamma\), effective \(V(F) = U(F)\), with \(\gamma > 0\) because a harm caused by acting is attributed to the agent more readily and judged more culpable than the same harm allowed by forbearing (and invites sharper anticipated regret). The agent chooses \(F\) whenever \(U(F) > U(A) - \gamma\), i.e. whenever \(U(A) - U(F) < \gamma\) — so \(F\) is chosen even across a band of outcome differences where \(A\) is substantively superior, the band's width set by \(\gamma\). Two structural predictions fall out exactly: the tilt toward inaction is strongest where \(|U(A) - U(F)|\) is smallest (the culpability term, not the outcome, decides), and where acting would make the agent the salient cause of a harm; and the bias is correctable by acting on \(\gamma\) rather than the substrate — making omissions as attributable and accountable as commissions drives \(\gamma \to 0\) and restores substrate-optimal choice. The bias lives in the \(\gamma\) term (the evaluation layer), not in \(U\) (the decision substrate) — and the contrast with action bias is now precise: action bias adds a credit term \(+\beta\) to the act, omission bias subtracts a culpability term \(-\gamma\) from it, and which dominates in a given context fixes the direction of the agent's tilt.

Mapped back: The augmented decision model instantiates every role — symmetric-outcome choice, accountability environment (\(\gamma\)), commission–omission asymmetry, a tilt toward inaction exceeding the outcome calculus, the substrate/evaluation separation, and the correctability invariant (\(\gamma \to 0\)) — and locates the distortion precisely in the evaluation layer, as the sign-flipped twin of the action-bias model.

Applied/industry¶

In vaccination decisions, a parent facing a choice whose expected outcomes favor vaccinating — a small vaccine risk against a larger disease risk — systematically tilts toward not vaccinating. The substrate payoffs favor action, but the evaluation layer penalizes it: a rare adverse event caused by the vaccine they chose to give is felt and judged as the parent's doing, while the larger expected harm from a disease they merely failed to prevent blends into the background of "what happened," not "what I did." The prime's interventions apply directly: counting the omission explicitly — presenting the expected harms of non-vaccination as a consequence the parent is choosing and accountable for, not a neutral default — and reframing the default as an active choice strip the asymmetry by which the disease risk reads as nobody's doing. The identical structure governs drug regulation: a regulator faces sharper accountability for an error of commission (approving a drug that later harms identifiable victims) than for an error of omission (delaying or denying a beneficial drug whose victims are statistical and unattributed), biasing the agency toward inaction even when faster approval has better expected outcomes; symmetric-accountability rules that hold the agency answerable for the victims of delay as squarely as for the victims of approval re-credit the omission and curb the tilt. And in clinical care, a physician deciding between an active intervention and watchful inaction at equal expected outcomes tilts toward not treating, because a bad outcome after treatment invites blame for "what the doctor did" while the same outcome after waiting reads as "the disease ran its course"; an outcome-based decision rule that treats expected outcomes as decisive regardless of which option is the act removes the choice from the moment of culpability-driven regret.

Mapped back: Across vaccination, drug regulation, and clinical care the same roles recur — a symmetric-outcome choice, an accountability environment, a commission–omission culpability asymmetry, and a tilt toward inaction beyond what the outcomes justify — and the same intervention family transports: count and reframe the omission, impose symmetric accountability, and pre-commit to outcome-based rules so that equalizing the attribution of action and inaction, the single correctable lever, removes the bias at its source.

Structural Tensions¶

T1 — Evaluation Layer versus Decision Substrate (scopal). The prime locates the bias in the evaluation layer (how commission and omission are attributed and blamed), not the substrate (what they produce) — the consequences may be equal while the culpability evaluations differ. The failure mode is substrate misattribution: trying to fix the bias by changing the actual outcomes when the distortion was in how they were judged. Diagnostic: would the same choice be made if a harm from inaction were attributed to the agent as squarely as a harm from action? If the tilt toward not acting vanishes, it lived in the evaluation layer, and substrate-level fixes miss it.

T2 — Omission Bias versus Action Bias (sign/direction). Omission bias is the mirror of action bias — the same agent over-prefers inaction in some contexts and action in others, and the two form a context-dependent pair. The failure mode is single-pull assumption: installing accountability-for-omissions rules to curb omission bias in a context where action bias was actually the operative pull, worsening a different distortion. Diagnostic: is commission more culpable than omission here (omission bias), or is action more visible and creditable than inaction (action bias)? The corrective lever flips between the two; misreading the context applies the wrong fix.

T3 — Symmetric Outcomes versus Genuine Inaction Superiority (scalar). The bias operates where outcomes are similar, and is strongest where acting would make the agent the salient cause of harm — but sometimes inaction genuinely is superior (irreversibility, real downside risk), and over-applying the frame forces unwarranted action. The failure mode is action overcorrection: an outcome-based rule that compels a genuinely risky or irreversible intervention because the program defaults to symmetric treatment of act and omission. Diagnostic: how large is the actual outcome gap, and is the action irreversible? The bias correction is calibrated to near-symmetric choices; where forbearance has a real edge, forcing action harms.

T4 — Accountability for Omissions versus Surveillance and Blame Cost (coupling). The single correctable lever is making omissions as attributable and accountable as commissions, but instrumenting and blaming omissions (tallying every unprevented harm, holding agents answerable for all foregone benefits) imposes its own overhead and can over-correct into an action bias or a culture of blame. The failure mode is omission-accounting overreach: holding agents responsible for every conceivable foregone benefit, paralyzing them with limitless liability for inaction. Diagnostic: does the decision's stakes justify per-decision omission-accounting, or only the consequential subset? Re-crediting omissions costs attention and can tip into over-blame, so it must be sized to the stakes.

T5 — Culpability Asymmetry versus Justified Acts/Omissions Distinction (scopal). The prime treats the commission/omission gap as a bias to be corrected, but some of that distinction is normatively defensible — many ethical and legal systems hold, on reflection, that a duty not to harm is genuinely stronger than a duty to rescue. The failure mode is flattening a real moral distinction: treating every preference for omission as irrational bias and overriding considered ethical or legal commitments to the acts/omissions doctrine. Diagnostic: is the asymmetry here an unreflective attribution artifact operating where outcomes are equivalent, or a considered normative position about the differential stringency of duties? The bias frame applies to the former; applying it to the latter smuggles a contested ethical claim in as if it were a debiasing.

T6 — Individual Bias versus Institutional Encoding (scalar). The prime treats the bias as a property of the individual decision-maker, but the commission/omission asymmetry is often built into institutions — legal codes that punish acts over omissions, regulatory regimes that scrutinize errors of commission over omission, liability rules that attach to what one did but not what one allowed. The failure mode is individual-level fix for a structural cause: training individuals to weigh omissions equally while the legal and regulatory structure keeps attaching liability asymmetrically. Boundary with agency_problem. Diagnostic: is the asymmetry in the person's attribution psychology or in how the institution assigns liability and blame to acts versus omissions? A structural asymmetry requires changing the institutional accountability design, not individual debiasing.

Structural–Framed Character¶

Omission bias sits on the framed side of the structural–framed spectrum, matching its aggregate of 0.5 — and its placement is the mirror of its sibling action_bias (also 0.5), as one would expect of two members of the same cognitive-bias family. There is a clean relational structure underneath — a symmetric-outcome choice plus a culpability penalty \(\gamma\) subtracted from the more-attributable (commission) option, biasing choice toward inaction beyond what the outcome calculus justifies, and even formalizable as an augmented expected-utility model — but the prime is a cognitive-bias category bound to human decision-makers, and that binding is decisive for the grade.

The pinning diagnostic is human-practice-bound, scored at the ceiling. Omission bias is inherently a property of accountable human agents and their causal-attribution and culpability machinery: it requires a decision-maker who can be held responsible, a distinction between a harm one caused and a harm one allowed, and the anticipated regret and blame that attach asymmetrically to the two — none of which exists outside human (or at least agentive, morally-evaluable) cognition. There is no commission/omission asymmetry for a thermostat or an optimizer, because neither can be blamed for what it caused versus what it permitted; the reach is among human-agent settings (medicine, ethics, law, policy, investing) rather than to non-human substrates, and what travels is a repertoire for equalizing the accountability of action and inaction. The remaining diagnostics read mid-scale or low. Vocabulary half-travels: "acts versus omissions," "errors of commission," "culpability," and "duty to rescue" carry a judgment-and-decision-making and moral-philosophy lexicon a new domain must partly adopt. Evaluative weight is moderate — "bias" names a distortion to be corrected, and the prime carries a faint normative charge in calling a culpability-driven preference an error — without full institutional loading; the half-point is honest, since part of the commission/omission distinction is a defensible normative position rather than pure bias. Institutional origin reads zero, fittingly: the bias is rooted in individual causal-attribution and culpability cognition rather than in any formal institution, though the entry notes the asymmetry is often encoded into legal and regulatory institutions. Invoking the prime half-imports a frame (locate the bias in the evaluation layer, make omissions accountable) and half-recognizes a tilt already present.

The prime's substrate reasoning is explicit that the asymmetric preference for inaction over action, though genuinely cross-domain, has only human decision-makers as its instances and is not a substrate-portable structural shape. The augmented decision model shows the structure is real and formalizable, but the \(\gamma\) term is a culpability cost only a morally-accountable agent carries. That is the framed signature — a genuine relational asymmetry that exists only inside human decision-making practice and carries that practice with it.

Substrate Independence¶

Omission bias is a moderately substrate-independent prime — composite 3 / 5 on the substrate-independence scale, the same grade as its mirror sibling action_bias. Its domain breadth is real but confined to a single substrate band: the asymmetric preference for inaction recurs with the same force in clinical medicine and public health (vaccine hesitancy, the pull away from intervention), ethics (the acts/omissions doctrine, trolley intuitions), law (the near-universal distinction between killing and letting die, between a duty not to harm and a narrow duty to rescue), policy and regulation (the asymmetric scrutiny of errors of commission over omission), and investing (status-quo holding to avoid the regret of a caused loss) — but every instance is a human decision-maker in an environment of moral or reputational accountability, and the cross-domain reach is among such agents rather than to any non-human system. What caps the structural-abstraction component at the middle is that the formalizable core — an augmented decision model in which a culpability term \(\gamma\) tilts the choice toward inaction — encodes a cost only a morally-accountable agent bears, so the pattern is constitutively human-cognitive with no physical or biological substrate. Transfer evidence is moderate: the diagnostic (when act and refrain have similar expected outcomes, suspect the culpability asymmetry, and make omissions as accountable as commissions) carries across medicine, ethics, law, policy, and finance, but never leaves the band of human decision-makers. Genuine cross-domain reach among human agents, capped by the absence of any non-cognitive substrate, fixes the composite at 3.

Composite substrate independence — 3 / 5
Domain breadth — 3 / 5
Structural abstraction — 3 / 5
Transfer evidence — 3 / 5

Relationships to Other Primes¶

Parents (1) — more general patterns this builds on

Omission Bias is a kind of Bias

Freshly created (2026-06-13) with an explicit review_flag to "draw the proposed parent/child edges at incorporation." The file's own "What It Is Not" states: "Not bias in general — bias is the umbrella structure of a systematic deviation; omission_bias is the specific attributional-and-culpability asymmetry between commission and omission." That is a textbook is-a / specialization: omission_bias IS a (kind of) bias. bias is canonical. Its true mirror-sibling action_bias is NOT a valid slug (absent from both lists), so the sibling edge the manifest proposed cannot be drawn here; child_of bias is the available high-conviction hierarchy edge. (loss_aversion / status_quo_bias also contrasted but as non-confusions, not parents.)

Path to root: Omission Bias → Bias

Neighborhood in Abstraction Space¶

Omission Bias sits among the more crowded primes in the catalog (17^th percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.

Family — Unclustered & Miscellaneous (91 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With¶

The nearest existing prime by embedding is action_bias, and the relationship is the most important one this entry exists to fix: the two are mirror siblings, not the same bias and not unrelated. Action bias is a systematic tilt toward acting when action and inaction have similar expected payoffs, driven by the greater visibility and creditability of action — doing something is rewarded and attributable, so the agent over-acts. Omission bias is a systematic tilt toward not acting under the same symmetric-payoff condition, driven by the greater culpability of commission — a harm one caused by acting is judged more harshly than an equal harm one allowed by forbearing, so the agent over-refrains. They share a structure — a symmetric-outcome choice, an accountable agent, an attribution asymmetry in the evaluation layer, and a single correctable lever — but the sign of the asymmetry is reversed: action bias adds a credit term to the act, omission bias subtracts a culpability term from it. The two are a context-dependent pair: action bias dominates where visible action is rewarded (crisis leadership, trading-as-productivity, "do something" pressure), omission bias dominates where commission is blamed (medical intervention, regulatory approval, irreversible calls). The distinction is load-bearing because the corrective levers are opposite: action bias is curbed by licensing and re-crediting inaction (watchful-waiting protocols, default-no rules), omission bias by making omissions as accountable as commissions (counting the harms of inaction, symmetric-accountability rules). A practitioner who installs a default-no rule to curb over-action in a setting where the operative pull was actually omission bias will deepen the very distortion they meant to fix.

A second genuine confusion is with status_quo_bias. Both can produce a preference for leaving things alone, and the two frequently coincide because inaction usually preserves the current state. But status-quo bias is a preference for the current state as such — it favors whatever option keeps things as they are, across any choice including a choice between two actions — whereas omission bias is keyed specifically to the act-versus-refrain distinction, penalizing the option that constitutes an intervention regardless of whether that intervention changes or preserves the state. The two come apart exactly where acting is what preserves the status quo and not acting would let it change: status-quo bias then favors the action that holds the state, while omission bias still favors not acting and accepts the change. The discriminating question is whether the agent is avoiding change (status-quo bias) or avoiding being the cause (omission bias). Conflating them misroutes the fix: a status-quo bias is countered by neutralizing the default and equalizing the framing of options, while an omission bias is countered by equalizing the accountability of commission and omission — and a harm allowed by inaction can be a departure from the status quo that omission bias nonetheless prefers.

A third confusion worth drawing is with loss_aversion. Both add extra weight to bad outcomes, and both can make an agent reluctant to act. But loss aversion weights losses more heavily than equal gains, a frame-dependent asymmetry indifferent to agency — it would equally deter an action that risked a loss whether the agent or nature was the cause. Omission bias adds weight specifically to agent-caused harms over agent-allowed ones, and it persists even between two outcomes framed identically as losses: a loss one caused by acting is judged worse than an equal loss one allowed by not acting. The discriminating question is whether the extra weight tracks the loss/gain frame (loss aversion) or the commission/omission distinction (omission bias). A loss-averse but attribution-neutral agent would treat a caused and an allowed loss alike; an omission-biased agent distinguishes them sharply. Conflating the two sends a practitioner to reframe losses as gains when the real distortion was about who caused the harm, or vice versa.

For a practitioner, the distinctions sort by what kind of asymmetry is in play. If the agent avoids change across any choice, it is status_quo_bias (neutralize the default); if the agent weights losses over gains regardless of agency, it is loss_aversion (address the frame); if the agent tilts toward acting because action is visible and creditable, it is action_bias (license inaction); and if an accountable agent tilts toward not acting because a caused harm is judged worse than an equal allowed harm, it is omission bias — the only one whose remedy is to make omissions as attributable and accountable as commissions, so that the agent chooses the better outcome rather than the less-culpable role.

Solution Archetypes¶

No catalogued solution archetypes reference this prime yet.