Information Avoidance¶

Actively decline information that is freely available because the anticipated content carries disutility — affective pain, identity threat, or an unwanted obligation — so the resulting non-knowledge is a chosen decision, not an absence.

Core Idea¶

Information avoidance is the active decision by an agent not to acquire information that is freely available, when the anticipated content of that information carries disutility — not because the cost of acquisition is prohibitive, but because the agent expects that knowing will be worse than not knowing.

The pattern is structurally distinct from ordinary ignorance and from rational inattention. Ignorance is the absence of information the agent would want if they could have it. Rational inattention, in the technical sense associated with Christopher Sims, is the allocation of a limited information-processing budget across possible signals — a cost-benefit calculation in which acquisition cost is positive and must be weighed against informational value. Information avoidance is neither: the information is freely available, the acquisition cost is negligible, and the agent declines anyway because the disutility attaches to the state of knowing itself, not to the effort of finding out.

Three types of anticipated disutility drive the pattern and can combine. Affective cost is the most immediate: the agent predicts that learning the result will be emotionally painful. An investor who expects their portfolio has fallen avoids checking their account; a patient with a worrying family history declines a genetic test. Identity threat is the second type: learning would challenge a self-image the agent is invested in preserving. Dana, Weber, and Kuang showed in 2007 that subjects in modified dictator-game settings will pay to avoid learning whether their selfish allocation harms another participant, because knowing would force a revision of their self-assessment as a fair person. Obligation imposition is the third: the agent anticipates that learning will create a duty to act that they do not want to assume — a consumer who suspects a product's supply chain involves harm may avoid finding out because knowing would impose a costly response obligation.

Across these three drivers the structural signature is the same: a freely available signal with anticipated negative utility to knowing it, a low-cost avoidance act — not booking the appointment, not opening the email, not asking — and a consequent state indistinguishable from ignorance to outside observers and, often, to the agent themselves. The agent can sincerely report "I don't know," and the report is true; the structural origin of the non-knowledge is avoidance, not absence.

The pattern appears in a family of named instances, each a specialization to a particular domain. The ostrich effect names the financial-portfolio variant: investors check accounts less frequently after losses and in down markets. Willful ignorance names the moral-choice variant: not learning whether one's action harms another preserves a clean conscience and the ability to claim good faith. Plausible deniability names the institutional and legal variant: principals structure information flow so they cannot be charged with knowing what they have arranged not to learn, even when the information is freely available and the arrangement is deliberate. Strategic non-knowing names the negotiation variant: a negotiator who avoids learning the counterpart's reservation price maintains credible commitment to a take-it-or-leave-it position. All four share the three structural elements; they differ in what the preserved benefit is.

Structural Signature¶

Sig role-phrases:

the agent with preferences over information states — a substrate to whom utility attaches to knowing or not-knowing, not only to the underlying state
the freely available signal — information whose acquisition cost is negligible, so declining is not a cost-benefit calculation about paying for it
the anticipated disutility of knowing — the agent's preview-belief that the content will be bad, with the cost attaching to the knowing rather than to the underlying state
the affective-cost driver — disutility as emotional pain (the unchecked portfolio, the declined genetic test)
the identity-threat driver — disutility as a forced revision of a self-image the agent is invested in preserving (the dictator-game subject keeping "I am fair")
the obligation-imposition driver — disutility as a duty to act that knowing would create (the consumer who would have to respond to a harmful supply chain)
the low-cost avoidance act — the recognizable not-receiving move: not booking, not opening, not asking, delegating, or designing the signal away
the indistinguishability from ignorance — the resulting non-knowledge opaque to outside observers and to the agent's own sincere "I don't know," its origin being decision, not absence
the matched-lever family — the architecture-level remedies the structure prescribes, each acting pre-receipt: default-toward-disclosure, pre-commit-to-receive, reduce-anticipated-disutility, externalize-to-a-trustee, audit-for-avoidance
the agentive boundary — the precondition that walls the pattern in: it requires an agent with utility over information states, so non-agentive substrates cannot instantiate it

What It Is Not¶

Not ignorance. Ignorance is the absence of information the agent would want if they could have it; information avoidance is the active refusal of information that is freely available and whose content the agent previews as bad. The two look identical from outside — the agent can sincerely report "I don't know" — but the origin differs: in ignorance the information never came within reach, in avoidance the agent declined it. The structural source is a decision, not an absence.
Not rational inattention. In the Sims sense, declining a signal is a cost-benefit calculation in which acquisition cost is positive and weighed against informational value. Information avoidance has the opposite cost structure: acquisition is essentially free, and the disutility attaches to the state of knowing itself, not to the effort of finding out. Reading avoidance as ordinary budget-limited inattention misses that the agent declines despite negligible cost, precisely because knowing would be worse.
Not denial or motivated reasoning. Those are post-receipt failures — they distort how evidence already in hand is weighed. Information avoidance operates upstream, preventing the evidence from arriving so that no reasoning about it is ever required. The distinction is consequential: belief-correction remedies (better arguments, debiasing, confrontation with facts) target the wrong stage and do not bite, because the agent never let the facts in.
Not irrationality. The avoidance is purposive: the not-knowing buys a preserved benefit — affect spared, identity protected, an obligation evaded, or a bargaining position held — and the agent has a genuine preference over information states. It is a choice that pays, not a lapse of reasoning; treating it as mere foolishness misses that the agent is rationally protecting something the information would cost them.
Not an exculpatory account. "I didn't know" remains true as a report but stops being a defense as an account, because the non-knowledge was deliberately arranged. The whole diagnostic point is that sincere non-knowledge can have a chosen origin; reading the true self-report as automatically blameless is exactly the inference the concept blocks, and what lets an investigator separate genuine lack of data from engineered deniability.
Not merely passive looking-away. The avoidance act ranges from not opening an email to designing the system so the disutility-bearing signal never arrives — compartmentalization, firewalls, "we don't ask" sourcing. Plausible deniability is deliberately built architecture, not inattention. Reading information avoidance as only a passive failure to notice misses the engineered, institutional variant in which the non-knowledge is constructed on purpose.

Scope of Application¶

Information avoidance is substrate-bound to agentive systems with utility over information states; its reach spans the agent-bearing fields below — psychology, behavioral economics, ethics, law, negotiation, institutional design — but stops at the agent/non-agent line, since a substrate with no one for whom knowing is worse than not knowing cannot instantiate it. The four named effects below are its specializations, not cross-domain analogues; the unifying pattern is this entry itself, while non-agentive look-alikes belong to other primes entirely.

Financial behavior — the ostrich effect: investors check portfolios less often after losses and in down markets (Karlsson, Loewenstein, and Seppi's brokerage data), sparing the affective cost of a known loss.
Medical testing — declining freely available genetic tests (Huntington's, BRCA), HIV-status, or cancer screening after a worrying family event, where the disutility attaches to knowing rather than to the disease state.
Performance feedback — employees and students leaving negative reviews unread, and authors delaying opening reviewer reports, to defer identity-threatening information.
Moral choice — willful ignorance: Dana, Weber, and Kuang's dictator-game subjects paying not to learn whether their selfish choice harms a partner, preserving a self-image as fair.
Institutional and legal design — plausible deniability: compartmentalized intelligence operations and campaign-finance firewalls engineered so principals cannot be charged with knowing what they arranged not to learn.
Negotiation — strategic non-knowing: declining to learn a counterpart's reservation price to hold a credible take-it-or-leave-it commitment.
Ethical consumption — consumers and "we don't ask" sourcing arrangements avoiding knowledge of labour or environmental conditions that would impose a costly response obligation.
Existential-risk attention — individuals and institutions declining to attend to large-scale risks whose acknowledgment would demand costly action.

Clarity¶

Naming information avoidance makes inspectable a structural choice that the agent can otherwise characterize to themselves, sincerely, as a different kind of state. Three routes to not-knowing collapse to the same outward appearance: not knowing because the information was unavailable (ignorance), not knowing because acquisition was judged not worth its cost (rational inattention), and not knowing because the agent declined freely available information whose content they previewed as bad (avoidance). The first two are blameless and familiar; the third is the one the concept claims, and the analyst's sharper question becomes diagnostic rather than charitable — was the signal actually free to obtain, and did the agent take a recognizable act to not-receive it? When the answer is yes, "I didn't know" remains true as a report while ceasing to be exculpatory as an account, because the origin of the non-knowledge is a decision, not an absence. This is exactly what lets a designer or investigator distinguish a manager who genuinely lacked the data from one who arranged not to have it.

The concept also fixes a second line the literature needs kept sharp: the difference between pre-receipt avoidance and the post-receipt failures it is constantly confused with. Denial, motivated reasoning, and dissonance reduction all operate on information already in hand, distorting how evidence is weighed; information avoidance operates upstream, preventing the evidence from arriving so that no reasoning about it is ever required. Locating the move at this earlier stage tells the practitioner that interventions aimed at correcting belief — better arguments, debiasing, confrontation with the facts — are aimed at the wrong stage entirely, since the agent never let the facts in. The lever has to act on the architecture of information flow itself, and seeing that the problem lives before receipt rather than after it is what the label supplies.

Manages Complexity¶

Instances of someone not knowing something they could have known are scattered across finance, medicine, performance feedback, moral choice, negotiation, and institutional design, and an unaided observer would have to reason about each from scratch, weighing whether a given non-knowledge is innocent or strategic in the particulars of that setting. The concept compresses that field to a three-part diagnostic asked of any case: was the signal freely available, did the agent anticipate disutility from knowing, and did they take a recognizable act to not-receive it — a yes on all three being the signature, and the same three questions serving the investor, the patient, the dictator-game subject, the negotiator, and the institutional designer alike. The internal variety reduces further to a typology of three disutility drivers — affective cost, identity threat, obligation imposition — so the four named instances (ostrich effect, willful ignorance, plausible deniability, strategic non-knowing) are read as one structure differing only in which driver supplies the preserved benefit. And because the move is fixed at the pre-receipt stage, the matched intervention family follows directly from the driver in play rather than being reinvented per domain: default-toward-disclosure, pre-commitment to receive, reducing the anticipated disutility, or externalizing the decision to a trustee. A high-dimensional "is this non-knowledge blameless or arranged, and what would change it?" problem thereby collapses to three diagnostic checks, a three-driver typology, and a short roster of architecture-level levers.

Abstract Reasoning¶

Information avoidance licenses reasoning moves that all turn on the three-part signature — freely available signal, anticipated disutility, recognizable not-receiving act — and the three-driver typology beneath it.

Diagnostic (resolve which kind of not-knowing this is): confronting an agent who does not know something, infer the structural origin by running the three checks — was the signal actually free to obtain, did the agent anticipate disutility from knowing, did they take a recognizable act to not-receive it. A yes on all three discriminates avoidance from its blameless look-alikes: ignorance (the information was unavailable) and rational inattention (acquisition cost was positive and judged not worth paying). The move runs from the cost structure and the avoidance act to the type of non-knowledge: when acquisition cost is negligible and a deliberate not-receiving act is present, the non-knowledge is a decision, not an absence — so "I didn't know" stays true as a report while ceasing to be exculpatory as an account. This is what lets an investigator separate a manager who genuinely lacked data from one who arranged not to have it.

Driver-attribution (read the preserved benefit): having identified avoidance, infer which of the three disutilities is doing the work by asking what benefit the not-knowing preserves. Affective cost predicts the agent is protecting against emotional pain (the unchecked portfolio, the declined genetic test); identity threat predicts they are protecting a self-image that the information would force them to revise (the dictator-game subject preserving "I am fair"); obligation imposition predicts they are avoiding a duty to act that knowing would create (the consumer who would have to respond to a harmful supply chain). The move runs from the avoided content to the driver: the same outward act of not-receiving resolves into different mechanisms depending on what knowing would have cost, and the four named instances — ostrich effect, willful ignorance, plausible deniability, strategic non-knowing — are read as one structure differing only in which driver supplies the preserved benefit.

Stage-locating (boundary-drawing against post-receipt failures, with an intervention prediction): locate the move at the pre-receipt stage and infer that it is upstream of denial, motivated reasoning, and dissonance reduction, which all operate on information already in hand. The consequence is predictive about interventions: because the agent never let the facts in, remedies aimed at correcting belief — better arguments, debiasing, confrontation with the evidence — are aimed at the wrong stage and will not bite. The reasoning is from the stage at which non-knowledge is produced to the class of levers that can change it: the lever must act on the architecture of information flow itself, before receipt, not on reasoning that never occurs.

Interventionist (match an architecture-level lever to the driver in play): to change the behavior, the move is to select from a fixed roster of pre-receipt levers according to the diagnosed driver — default-toward-disclosure or opt-out delivery (remove the avoidance act from the agent's hand), pre-commitment to receive at a calmer moment (bind the future self before the disutility is salient), reducing the anticipated disutility (pair the information with support, frame it as actionable), or externalizing the decision to a trustee who receives and acts in the agent's place. Each pairing is a prediction that the lever closes the gap by neutralizing the specific driver: default-toward-disclosure defeats any driver by removing the act, while reducing anticipated disutility targets the affective driver directly. The move runs from the driver and the pre-receipt stage to the one family of levers that can reach the behavior — and predicts the inertness of belief-correction remedies that target the wrong stage.

Design-screening (predictive, on systems): examining a choice architecture before any failure has occurred, infer whether it creates avoidance incentives — whether the flow of information is arranged so that agents who would prefer not to know can readily arrange not to know. The move runs from the structure of information flow to the predicted presence of arranged non-knowledge: a system with opt-in delivery of disutility-bearing signals, a low-cost not-receiving act, and a benefit to remaining uninformed is one in which avoidance is predictable, so the designer or auditor can flag deliberately-built deniability (compartmentalization, firewalls) as engineered avoidance rather than incidental ignorance.

Knowledge Transfer¶

This entry is unusual in that it is the general pattern its better-known relatives specialize, so the within-domain transfer is broad and runs as mechanism — but the domain is agentive systems, and that boundary is the whole story. Across finance (the ostrich effect — investors checking accounts less after losses, documented by Karlsson, Loewenstein, and Seppi in brokerage data), medicine (genetic-test, HIV-status, and screening avoidance), performance feedback (delayed reviewer reports, unopened evaluations), moral choice (willful ignorance — Dana, Weber, and Kuang's dictator-game subjects paying not to learn whether they harmed a partner), institutional and legal design (plausible deniability — compartmentalized intelligence operations, campaign-finance firewalls), negotiation (strategic non-knowing — declining to learn a reservation price to hold a credible take-it-or-leave-it line), ethical consumption, and existential-risk attention, the same three-part signature applies (freely available signal, anticipated disutility, low-cost not-receiving act), the same three-driver typology sorts the cases (affective cost, identity threat, obligation imposition), and the same matched intervention family transfers intact (default-toward-disclosure, pre-commit-to-receive, reduce-anticipated-disutility, externalize-to-a-trustee, audit-for-avoidance). The four named effects are not cross-domain analogies of information avoidance; they are its specializations — it is their parent in the structure, differing only in which preserved benefit pays for the foregone information. So the diagnostic and the levers move from an investor to a patient to a negotiator to an institutional designer without translation. This is genuine mechanism transfer, and it is wide.

The honest boundary is not metaphor-versus-mechanism within the domain but the agent/non-agent line at the domain's edge. The pattern is substrate-bound to agentive systems with utility over information states: it presupposes an agent to whom the disutility of knowing attaches, an anticipation of that disutility, and a benefit to not-knowing. Non-agentive substrates — physical processes, algorithmic systems without preferences, biological systems with no represented self — cannot instantiate it in the strict sense, because there is no one for whom knowing is worse than not knowing. This is why the pattern, broad as it is across psychology, economics, ethics, law, and design, does not cross the agent/non-agent boundary the way primes like feedback, accumulation, or coupling do — and that confinement is exactly what keeps it a domain-specific abstraction rather than a prime. (Should a clear non-agentive instance ever surface — say, an information-theoretic account of selection against costly receiver states — the case for promotion would open; absent that, the agentive boundary holds.)

Two adjacent confusions must be kept out, because they mark where the mechanism does not reach even within agentive systems. rational_inattention (the Sims tradition) looks identical from outside — an agent not acquiring a signal — but its cost structure is the opposite: acquisition cost is positive and the non-acquisition is a cost-benefit calculation, whereas in information avoidance acquisition is free and the disutility attaches to the knowing. And signaling runs the opposite direction of flow: it produces information to be observed, while information avoidance refuses to receive it (though an avowed "I don't want to know" can itself signal a position). The cross-domain lesson, then, is carried by this very pattern across all agent-bearing substrates — that is its reach — while the named instances (ostrich effect, willful ignorance, plausible deniability, strategic non-knowing) are the domain-specific specializations beneath it, and the agentive precondition is the wall past which neither the pattern nor its levers travel (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

The seminal demonstration is Dana, Weber, and Kuang's "moral wiggle room" experiment (2007). In a modified dictator game, the dictator chose between two allocations. In a baseline condition where the recipient's payoffs were shown, most dictators chose the option that was fair to the recipient. In the key condition the recipient's payoff was hidden but could be revealed at zero cost by clicking a button. A substantial share of subjects chose not to reveal it, and among those who stayed ignorant, most then selected the option that maximized their own payoff — behavior they had largely avoided when the consequences were visible. Because revelation was free, the non-knowing could not be rational inattention; subjects were paying, in foregone certainty, to preserve a self-image as fair while acting self-interestedly.

Mapped back: The costlessly-revealable payoff is the freely available signal; the click-to-remain-uninformed is the low-cost avoidance act; and the preserved "I am fair" self-image is the benefit supplied by the identity-threat driver. That subjects could sincerely disclaim knowing the harm while having arranged not to learn it is the indistinguishability from ignorance — non-knowledge whose origin is decision, not absence.

Applied / In Practice¶

Predictive genetic testing for Huntington's disease is the medical field case. After the causative gene was identified in 1993, an accurate presymptomatic test became freely available to at-risk individuals (children of an affected parent carry a 50 percent risk). Yet uptake has remained persistently low — widely reported at well under a quarter of eligible at-risk people, far below the interest such individuals express in the abstract. Because Huntington's has no cure, learning one carries the mutation delivers certain future disease without treatment recourse: the disutility attaches to knowing itself. Clinicians observe that many at-risk people deliberately decline testing to avoid the affective burden of a foreknown fatal diagnosis and the life-planning obligations it would impose.

Mapped back: The available presymptomatic test is the freely available signal and choosing not to be tested is the low-cost avoidance act. The dread of a foreknown incurable diagnosis is the affective-cost driver, while the planning duties knowing would force are the obligation-imposition driver. That the reach spans medicine here but presupposes a person for whom knowing is worse than not knowing illustrates the agentive boundary.

Structural Tensions¶

T1: Purposive protection versus long-run harm (avoidance pays locally and costs globally). Information avoidance is not irrationality — the not-knowing buys a genuine benefit: affect spared, identity preserved, an obligation evaded, a bargaining line held. But the same move that is locally rational is frequently globally destructive: the investor who avoids the losing portfolio forgoes the rebalancing the loss should trigger; the at-risk patient who declines screening forgoes early treatment; the manager who arranges not to know forgoes the chance to fix the problem. The tension is intrinsic to the concept — the disutility that drives avoidance attaches to knowing, while the value of the information attaches to acting, so protecting the self from the first cost forfeits the second benefit. Avoidance is a real preference being served and a real interest being sacrificed at once, and the concept cannot tell the agent which dominates without knowing the stakes. Diagnostic: Does the benefit this avoidance preserves (affect, identity, deniability) outweigh the value of the action the withheld information would have enabled — and locally or over the long run?

T2: True report versus non-exculpatory account (sincere non-knowledge that is nonetheless chosen). The concept's sharpest move is that "I didn't know" stays true as a report while ceasing to be exculpatory as an account when the non-knowledge was deliberately arranged. This is powerful but treacherous: it drives a wedge between sincerity and blamelessness that is hard to adjudicate, because the diagnostic (was the signal free, was there a recognizable not-receiving act) can misfire in both directions. Genuine ignorance can be wrongly reclassified as arranged avoidance by an investigator who assumes the agent must have previewed the content; and real avoidance can hide behind plausible claims that the signal was not actually free or the act not deliberate. The tension is that the concept licenses attributing culpable choice to a true disclaimer, which is exactly the inference most open to abuse — turning a diagnostic tool into a device for imputing knowledge that cannot be proven. Diagnostic: Is there evidence the signal was genuinely free and a recognizable not-receiving act was taken, or is culpable avoidance being imputed to a non-knowledge that may be ordinary ignorance?

T3: Pre-receipt levers versus autonomy (removing the avoidance act can override a legitimate choice). Locating the move upstream of receipt correctly implies that belief-correction remedies miss, and that effective levers act on information-flow architecture — default-toward-disclosure, opt-out delivery, externalize-to-a-trustee. But these levers work precisely by removing the avoidance act from the agent's hand, and avoidance is sometimes a legitimate, even protective, exercise of preference over one's own information states (declining a Huntington's test one has decided one does not want). The tension is that the most reliable intervention (force receipt by default) is also the most paternalistic, overriding a genuine preference the concept itself insists is real and purposive. A designer who defaults everyone into disclosure to defeat harmful deniability simultaneously defeats the patient's considered choice not to know. The architecture-level lever cannot distinguish protective avoidance from evasive avoidance; it removes both. Diagnostic: Is this avoidance a harmful evasion whose removal serves the agent or others, or a legitimate preference over one's own information states that a default-to-disclosure lever would wrongly override?

T4: Anticipated disutility versus the previewed unknown (avoidance runs on a guess about content not yet seen). The whole structure turns on the agent anticipating that knowing will be bad — a preview-belief about content they have not received. But that anticipation is itself a prediction that can be wrong: the portfolio may have risen, the test may be negative, the supply chain may be clean. The tension is that avoidance is triggered by expected disutility, yet the expectation is formed in the absence of the very information that would confirm it, so the agent can be avoiding good news, or overestimating the pain of bad news, on the strength of a pessimistic prior. This gives avoidance a self-sealing quality — it prevents the feedback that would correct the anticipation — and means the driver (anticipated disutility) can be systematically miscalibrated in a direction the agent never discovers. Diagnostic: Is the anticipated disutility driving this avoidance a well-calibrated forecast of the content, or a pessimistic prior the agent is protecting from the disconfirmation that receiving the signal would provide?

T5: Autonomy versus reduction (a named parent-pattern or the four specializations that actually do the domain work). Information avoidance is unusual: it is not an instance of a broader prime but the parent of four named specializations — the ostrich effect, willful ignorance, plausible deniability, strategic non-knowing — which are its domain instances differing only in which disutility supplies the preserved benefit. So the reduction runs the other way from the typical DS entry: within agentive systems the pattern itself travels as mechanism across finance, medicine, ethics, law, and negotiation, and the named children are what practitioners in each field actually reach for. What keeps it a domain-specific abstraction rather than a prime is the agentive boundary: it presupposes an agent with utility over information states, so non-agentive substrates cannot instantiate it, and it does not cross the agent/non-agent line the way feedback or accumulation do. The tension is whether "information avoidance" earns standing as its own general pattern or is best carried by its four specializations in situ (and would only graduate to a prime if a clean non-agentive instance ever surfaced). Diagnostic: Resolve toward the named specialization (ostrich effect, willful ignorance, plausible deniability, strategic non-knowing) when diagnosing a concrete case in its home field; toward information avoidance as the unifying parent when the point is that all four share one pre-receipt structure — and hold both inside the agentive boundary, past which neither travels.

Structural–Framed Character¶

Information avoidance sits at mixed on the structural–framed spectrum — further from structure than a value-neutral load-balancing mechanism like isostasy, but well short of a practice-constituted verdict-label like ad hominem, because it names a real behavioral regularity that carries a modest culpability charge. On evaluative_weight it leans framed but only partway: the base mechanism — an agent declining a costless signal whose content they preview as bad — is describable with no praise or blame, yet the concept earns much of its bite from stripping "I didn't know" of its exculpatory force (the willful-ignorance and plausible-deniability children exist precisely to impute chosen non-knowledge), so a normative edge rides along that a pure mechanism like feedback does not carry. On human_practice_bound it is firmly framed: the agentive boundary is the entry's whole thesis — the pattern presupposes an agent with utility over information states, and a substrate for whom knowing is not worse than not-knowing simply cannot instantiate it, so remove the valuing agent and there is nothing left to avoid. On institutional_origin it leans structural: the parent pattern is a documented fact about how agents behave (the Dana–Weber–Kuang moral-wiggle-room result, the Karlsson–Loewenstein–Seppi ostrich data), not an artifact minted by one survey or tradition — though its legal child (plausible deniability) is institution-bound, the umbrella is not. On vocab_travels it is mixed: within agentive systems the operative vocabulary (freely available signal, anticipated disutility, avoidance act, driver typology) carries its full content unchanged from an investor to a patient to a negotiator to an institutional designer, but it is pinned to agent-bearing substrates and floats free of nothing beyond them. On import_vs_recognize it is structural within its range — the four named effects are recognized as one pre-receipt structure worn differently, genuine specializations rather than analogies imported cue by cue.

The portable structural skeleton is a purposive gating of a costless-but-available input under anticipated disutility — a decision over information states in which an agent optimizing a utility that ranges over epistemic states forgoes a free signal because the anticipated posterior lowers that utility. That skeleton is what the entry instantiates from the general structure of preference-driven choice, and it is unusually broad because the entry is itself the parent of its four domain children rather than a leaf; but its cross-domain reach still belongs to the agent-general decision structure, not to "information avoidance" as such — the domain-accented specifics (the three-driver typology of affective cost, identity threat, and obligation imposition; the financial, moral, legal, and negotiation children; the whole intervention roster) stay home, and the agent/non-agent line is exactly the wall the skeleton's substrate-portable cousins (feedback, accumulation, coupling) cross and it does not. Its character: a purposive, agent-bound gating of costless information that travels broadly as mechanism across every agentive field yet halts at the agent/non-agent line, carrying just enough culpability charge to settle mixed rather than structural.

Structural Core vs. Domain Accent¶

This section decides why information avoidance is a domain-specific abstraction and not a prime — and it is an unusual case, because the entry is not a leaf under some broader prime but the parent of four named children, so the question is whether its own portable skeleton earns prime standing or belongs, more generally, to the decision structure it specializes.

What is skeletal (could lift toward a cross-domain prime). Strip the affect, the identity, the deniability, and a thin decision-theoretic structure survives: an agent whose utility ranges over its own epistemic states declines a costless, freely available input because it anticipates that the posterior the input would produce lowers that utility. The portable pieces are abstract — a chooser with preferences defined not only over the world but over its own states of knowing; an input available at negligible cost; a preview-belief that receiving it makes things worse; and a gating act that forgoes it. That is purposive choice under preferences over information states, and it is genuinely portable across every agentive field — which is exactly why the four children (the ostrich effect, willful ignorance, plausible deniability, strategic non-knowing) are recognized as one structure worn differently. But this shared core is agent-general decision-theory, not anything proprietary to "information avoidance"; it is the part these cases hold in common, not what makes any of them distinctive.

What is domain-bound. Almost everything with content is agent-and-institution furniture that does not survive extraction. The three-driver typology — affective cost (the unchecked portfolio, the declined genetic test), identity threat (the dictator-game subject keeping "I am fair"), obligation imposition (the consumer who would have to respond to a harmful supply chain) — is a taxonomy of human motives, not substrate-neutral form. The four named children are each pinned to a home field (finance, moral choice, legal/institutional design, negotiation), and the legal child, plausible deniability, is outright institution-constituted. The intervention roster (default-toward-disclosure, pre-commit-to-receive, reduce-anticipated-disutility, externalize-to-a-trustee, audit-for-avoidance) and the empirical cases (Dana–Weber–Kuang's moral wiggle room, the Karlsson–Loewenstein–Seppi ostrich data, Huntington's-test uptake) are the worked substance the fields actually study. The decisive test is the agentive boundary the entry makes its thesis: remove the agent for whom knowing is worse than not-knowing and there is nothing left to avoid — no disutility to anticipate, no benefit to preserve, no "I didn't know" to strip of its exculpatory force. The pattern is constituted by the very valuing agent the prime bar would ask it to shed.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. Information avoidance's transfer is bimodal in an atypical way. Within agentive systems it travels intact and widely — as the parent of its children, its diagnostic and levers move from an investor to a patient to a negotiator to an institutional designer without translation, because every agentive field supplies the one thing it needs, a chooser with utility over information states. Beyond agentive systems it does not travel by analogy — it does not travel at all: a physical process, a preference-free algorithm, or a self-less biological system has no state for which knowing is worse, so the pattern simply cannot be instantiated there. That hard wall is what keeps it domain-specific. And when the bare structural lesson is needed past the wall — for the substrate-general fact that a system can be arranged so a costly input never arrives — it is already carried, in more general form, by the substrate-portable primes the entry names as the ones that do cross the line where it cannot: feedback, accumulation, coupling. The cross-boundary reach belongs to those agent-neutral structures; "information avoidance," as named, carries agent-bound baggage — the disutility, the drivers, the culpability charge — that should stay inside the agentive domain, and would graduate toward a prime only if a clean non-agentive instance ever surfaced.

Relationships to Other Abstractions¶

Current abstraction Information Avoidance Domain-specific

Parents (3) — more general patterns this builds on

Information Avoidance is a kind of Decision Prime

Information Avoidance is a specialization of Decision, retaining the parent's defining structure while adding the child's specific commitments.
Information Avoidance presupposes Agency Prime

Information avoidance presupposes agency because it requires a system to forecast the consequence of knowing, pursue a preserved benefit, and select an action that prevents signal receipt.
Information Avoidance presupposes Preference Prime

Information avoidance presupposes preference because anticipated disutility is an ordering of the not-knowing state above the knowing state in the relevant context.

Children (1) — more specific cases that build on this

Ostrich Effect Domain-specific is a kind of Information Avoidance

Ostrich effect is information avoidance specialized to affectively bad, valence-selected news whose free acquisition is refused before belief revision.

Hierarchy paths (7) — routes to 7 parentless roots

Information Avoidance → Decision → Constraint

Show alternative paths (6)

Not to Be Confused With¶

Ignorance. The plain absence of information the agent would take if it came within reach — the signal never arrived, was never free, or was never noticed. Information avoidance is the reverse: the signal is freely available and the agent declines it because the anticipated content is bad. The two are outwardly identical (both yield a true "I don't know"), which is the whole diagnostic difficulty. Tell: was the signal genuinely free to obtain and did the agent take a recognizable act to not receive it (avoidance), or did the information simply never come within reach (ignorance)?
Rational inattention (Sims). The allocation of a limited processing budget across signals, where acquisition carries a positive cost that is weighed against informational value and some signals rationally go unread. Information avoidance has the opposite cost structure — acquisition is essentially free and the disutility attaches to the state of knowing itself, not to the effort. Tell: is the signal declined because paying for it isn't worth the value (inattention), or declined despite being costless because knowing would be worse than not knowing (avoidance)?
Denial and motivated reasoning. Post-receipt distortions — the evidence is already in hand and the agent misweighs, discounts, or explains it away. Information avoidance operates upstream, preventing the evidence from arriving so no reasoning about it is ever required. The stage difference is consequential: belief-correction remedies bite on denial but miss avoidance entirely, because the agent never let the facts in. Tell: has the disutility-bearing content already been received and is now being distorted (denial), or was it kept from arriving in the first place (avoidance)?
The four named children (ostrich effect, willful ignorance, plausible deniability, strategic non-knowing). These are not confusable peers but specializations of this very pattern — the same three-part structure (free signal, anticipated disutility, low-cost not-receiving act) differing only in which preserved benefit pays for the foregone information: spared affect (ostrich effect), a clean self-image (willful ignorance), legal/institutional cover (plausible deniability), or a bargaining commitment (strategic non-knowing). Information avoidance is their parent, not a sibling. Tell: name the field and the preserved benefit and you have picked a child; hold all four together as one pre-receipt structure and you have the parent.
Signaling. Runs the opposite direction of flow — it produces information for an audience to observe, whereas information avoidance refuses to receive it. They can touch (an avowed "I don't want to know" can itself signal a position), but the core moves are inverse. Tell: is the agent emitting a costly cue for others to read (signaling), or declining to take in a cue that is already on offer (avoidance)?

Neighborhood in Abstraction Space¶

Information Avoidance sits in a crowded region of the domain-specific corpus (9^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Startup Strategy & Adoption Dynamics (16 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12