Motivated Reasoning

Version

v4 · 2026-08-30 · History

Prime #

1403

Origin domain

Psychology Cognitive Science

Subdomain

reasoning and motivation → Psychology Cognitive Science

Related primes

Bias, Preference

Core Idea

Motivated reasoning occurs when a non-epistemic goal or preferred conclusion steers which evidence is sought, admitted, interpreted, weighted, remembered, or allowed to update belief. ^[1]

What makes the pattern distinctive is not that the reasoner has a stake — almost anyone reasoning about anything consequential has one — but that the stake is directional and accuracy-independent. It picks out a conclusion in advance, and it would go on picking out that conclusion if the evidence were reversed. Reputation, professional identity, a budget already spent, a position stated in public, the survival of a research programme, the wish to keep thinking well of a friend: any of these can occupy the role, and none of them is a reason to believe the favored answer true.

The canonical identity is narrower than the phrase's everyday use. Four commitments must hold at once. There must be a truth-directed task, one where the reasoner is at least ostensibly trying to get an answer right. There must be a directional motive indifferent to whether the answer is right. Evidence-handling must contain discretionary stages — places where a defensible-looking judgment call could go either way: which sources to consult, when to stop searching, how to code an ambiguous datum, which methodological objection to press, how far a disconfirming result should move the estimate. And the motive must predictably shift one or more of those stages, so that the resulting judgment departs from the verdict the same record supports under a rule applied evenhandedly to friendly and unfriendly evidence. ^[1]

Drop any one of the four and the classification fails: without discretion there is nothing for the motive to act on, and without the departure there is a stake but no distortion. The agent need not consciously intend the outcome, and characteristically does not. From the inside the process feels like ordinary care — the extra check, the reasonable doubt, the fair demand for better data — applied unevenly without the unevenness being noticed. ^[2]

Structural Signature

Directional stake → asymmetric gating at one or more discretionary stages of an evidence pipeline → belief displaced from the symmetric-rule verdict.

The formula is staged rather than atomic on purpose. The prime does not name a single act of believing what one wants; it names a stake propagating through a pipeline that has slack in it, with the slack doing the work. Where a procedure leaves no room for judgment, a motive has no purchase, however strong it is. ^[1]

Recurring features:

• An epistemic task with a determinate verdict-space. An agent or agent-like collective is trying to settle a question that admits of better and worse answers.
• An accuracy-independent stake carrying a sign. A goal, identity, commitment, or preferred conclusion that ranks the candidate verdicts on grounds unrelated to their truth, and ranks them before the record is complete.
• Discretionary stages in the pipeline. Search and stopping; admission and exclusion; interpretation of ambiguous material; credibility assignment and weighting; retrieval from memory; the threshold at which belief is permitted to move.
• Sign-dependent handling across those stages. One methodological standard, two stringencies, selected by which way the material points. The asymmetry, not the severity, is the marker: uniformly harsh scrutiny is conservatism, not motivation.
• A symmetric-handling counterfactual. The benchmark verdict — what this record yields once the stake is removed or its sign flipped. Without that comparison there is a complaint about someone's conclusion but no diagnosis of their process. ^[3]
• Exclusion: strategic persuasion with intact private handling. Reporting otherwise than one has read the evidence is something adjacent and different.
• Exclusion: cold effects with no favored conclusion. Accessibility, fluency, anchoring, and computational shortcuts displace judgment without any preferred answer, and belong to the parent rather than here.

What It Is Not

Identifying motivated reasoning is not the same as showing the conclusion is wrong. A person with every reason to want a drug to work may reach, by a thoroughly corrupted route, the verdict that it works — and be right. The prime indicts a process, not a proposition, and the two come apart in both directions: sound process can yield error, and gated process can land on truth by luck. Treating a motive attribution as a refutation converts a diagnostic tool into an ad hominem, which is a common way it is misused. ^[4]

It is not an accusation of lying, nor even of bad faith. The characteristic case is sincere: the reasoner would pass a lie-detector on every individual step, because every individual step was defensible on its own. What was not defensible was the pattern of which steps got taken.

It does not claim that the favored direction is always the comfortable one. Motives can point toward unpleasant conclusions as readily as pleasant ones. A person who wants a failure to be their own fault, because that version is one they can fix, is gating evidence toward self-blame.

It does not claim that having a stake is sufficient. Interested parties frequently reason well, and some of the best evidence in any field is produced by people who badly wanted a particular result and disciplined themselves against it. The prime is silent until the stake demonstrably enters the pipeline.

Nor is it a claim about rarity or pathology. Nothing in the structure requires an unusual person, a defect, or a strong emotion; ordinary competent reasoners with ordinary commitments satisfy the conditions routinely. And it is not a general account of why people disagree. Disagreement also arises from different evidence, different priors, different definitions, and plain error, and reaching for a motive first is itself a shortcut worth resisting.

Broad Use

The pattern recurs wherever a decision procedure hands discretion to someone with something riding on the outcome. Institutional settings with explicit epistemic duties are its most legible hosts, precisely because they write down the standard the asymmetry then departs from. ^[5]

Science and peer review: analytic flexibility in a study with a preferred hypothesis — which outliers are excluded, which covariates enter the model, which of several defensible specifications gets reported; and on the reviewing side, a referee who reads a paper threatening their own programme with a rigor never applied to one supporting it.

Clinical medicine: diagnostic momentum, where a clinician who has committed to an initial reading treats confirming findings as decisive and disconfirming ones as artifact; and the reverse case, where a feared diagnosis raises the evidentiary bar for admitting it.

Law and forensic practice: an investigator whose case has already been staked on a suspect deciding which leads are worth the overtime, and a forensic examiner exposed to the domain-irrelevant context of what the prosecution expects to find.

Intelligence and threat assessment: analysts working under a policy commitment, where the discretionary stages are source-reliability grading and the interpretation of ambiguous collection.

Organizational audit and internal investigation: the scope of an inquiry, the choice of which documents count as responsive, and the point at which the investigation is declared complete — all discretionary, all under a stake belonging to whoever commissioned it.

Finance and valuation: assumption-setting in a model whose output determines whether a deal proceeds, and the asymmetric treatment of upside and downside scenarios once a position is held. ^[6]

Politics and civic belief: identity-protective evaluation of arguments, where the same statistical reasoning is executed competently on a congenial dataset and abandoned on an uncongenial one.

Memory and autobiography: retrieval is itself a discretionary stage, and self-narrative supplies the stake — recollection of one's own past positions drifts toward consistency with present ones.

Moral judgment: intuitive condemnation or exculpation arrives first, and the search for principles is conducted under a stake already assigned.

Clarity

The confusion motivated reasoning dissolves is the false dichotomy between honest error and deliberate deception. Faced with a conclusion that looks obviously wrong to everyone outside the reasoner's position, observers reach for one of two explanations: they were mistaken, or they were lying. Both are usually wrong, and both misdirect the response, the first prescribing better information and the second sanctions. The prime supplies the missing third category: a sincere reasoner, correct at every step, whose selection among steps was steered. ^[1]

That distinction explains an otherwise puzzling fact about remedies. Giving disputants more evidence, better evidence, or the same evidence together often fails to close the gap between them, and can in some settings widen it, because a richer record offers more discretionary material to each side's gating. Once the mechanism is located in the handling rather than in the supply, the failure of information-provision stops being a mystery and starts being a prediction.

It also clarifies why expertise offers so little protection. Skill in a domain is largely skill at constructing and evaluating arguments in it, which is exactly the capacity a directional stake conscripts. The expert has more admissible objections available, more defensible specifications to choose among, and more credibility to lend the result. Sophistication changes the quality of the reasons produced without changing the direction they were produced in. ^[7]

Finally, it separates two questions that are constantly run together: whether a judgment is wrong, and whether the route to it was sound. Keeping them apart is what allows a reviewer to say "your conclusion may well be correct and your process is not one I can rely on," which is both the honest verdict in most disputes and the one the honest-error/lying dichotomy cannot express.

Manages Complexity

The abstraction lets an analyst stop tracking the object-level dispute. To ask whether a directional stake is operating, one does not need to adjudicate whether the drug works, the deal is sound, or the defendant is guilty; one needs the stake, the discretionary stages, and the counterfactual. This is a large saving, because the object-level question is typically the hard one and is often unresolvable within the time available, while the structural question is answerable from the procedural record alone.

It also lets one stop modelling the reasoner's interior. Sincerity, intent, self-awareness, and emotional intensity are all unnecessary inputs. Nothing in the diagnosis turns on what it was like to be the person, which removes the least tractable part of the inquiry and the part most likely to make the conversation adversarial.

Most usefully, it collapses a sprawling catalogue of separately named effects into one generator with one intervention family. Selective exposure, biased assimilation, disconfirmation bias, asymmetric standards for evidence quality, self-serving attribution, and convenient forgetting are not six phenomena to be learned and countered individually; they are six discretionary stages of a single pipeline under the same directional pressure. One diagnosis reaches all of them, and the same countermeasures — remove the discretion, remove the stake, or make the asymmetry visible — apply at every stage. ^[1]

What it does not let you stop tracking is the procedural detail. The compression is on the psychological side, not the documentary one: without the stopping points, exclusions, and specification choices on record, the analysis degrades to guesswork.

Abstract Reasoning

The prime licenses a four-step diagnostic that can be run on a record rather than on a person.

Step one: state the epistemic task and its verdict-space. What question was the reasoner formally answering, and what answers were available?

Step two: name the accuracy-independent stake and its sign. Which verdict does the reasoner's position favor, on grounds unconnected to the evidence? If the stake cannot be named in advance of looking at the reasoning — if it is inferred backwards from the conclusion reached — the diagnosis is circular and must be abandoned.

Step three: enumerate the discretionary stages actually present. Mark each point where the record shows a judgment call: sources consulted and not consulted, the stopping rule, inclusion and exclusion decisions, the coding of ambiguous items, the threshold set for updating.

Step four: run the symmetric counterfactual. This is the operative test, and it has three practical forms. The sign-flip test asks whether the same standard would have been applied to a mirror-image finding: would this objection have been raised against a study reaching the opposite result, would this witness have been discounted for the same inconsistency had they said the other thing? The asymmetric-scrutiny test compares effort rather than outcome — time spent, checks performed, and objections generated on congenial versus uncongenial material, on the principle that motivated handling shows up as unequal work before it shows up as unequal verdicts. The stopping-rule test asks when the search ended and whether that moment was fixed in advance or arrived when the answer became satisfactory. ^[2]

Then rule out the cold explanations. Not every asymmetry is motivated: a well-calibrated prior legitimately makes surprising evidence harder to accept. The sharpest boundary case is source discounting. Lowering one's trust in a source after learning it fabricated data is evidence-sensitive reliability assessment, not motivated reasoning — the discount is licensed by a fact about accuracy, and the test that confirms this is that the same discount would be applied to a source whose fabrications happened to favor the reasoner. When it would not be, the discount was the gate. ^[8]

Knowledge Transfer

What travels is the role structure and everything built on it. A stake, a pipeline with discretion in it, sign-dependent handling, and a symmetric benchmark can be identified in an individual clinician, a review committee, a firm's disclosure process, a regulator, an academic discipline, or a training-and-selection pipeline that promotes analysts whose conclusions please. The bearer changes and the diagnostic does not: for a committee the discretionary stages are agenda-setting, who is in the room, and what gets minuted; for a discipline, which findings are publishable and which questions fundable. The countermeasure family transfers with the diagnosis, since blinding, pre-commitment, and adversarial scrutiny are defined over stages of a pipeline rather than over minds. ^[9]

What does not travel is the phenomenology. The felt sense of conviction, the discomfort of a threatened belief, and the experience of an argument seeming compelling are real for individual human reasoners and have no counterpart in a firm or a funding body; importing them produces an explanation that sounds mechanistic and is in fact metaphor. Specific psychological mechanisms — affect-laden retrieval, threat response, the timing of intuition and justification — are likewise local, and a collective can display the identical input-output asymmetry through entirely procedural means.

Transfer also fails where there is no discretion to gate. A fixed estimator with a skewed prior produces systematically displaced output, but nothing in it would have gone the other way had the sign of some stake been reversed, so it is the parent pattern and not this one. Applying the prime to an automated pipeline is warranted only when something plays the motive role: an objective, a selection pressure, or a designer's stake that would have licensed different handling had the outputs pointed elsewhere. Where that role is genuinely occupied, the analysis carries; where "the system prefers" is a manner of speaking, it does not.

Examples

Formal/abstract

An analyst is testing whether a proposed change improves a metric, and the analyst's team proposed the change. Data arrive in weekly batches. After batch three the effect is positive but short of the threshold; the analyst collects another batch. After batch four it clears the threshold and the analysis is written up. Nothing here involves a fabricated number or a false statement; every batch was real and every calculation correct.

Now flip the sign. Suppose batch four had reversed the effect. Would the analyst have stopped there and written up a null result? On the record, no: the stopping rule in force was "continue until the result is satisfactory," which is not a rule about the world but a rule about the analyst's stake. A threshold calibrated for a fixed sample size does not hold under a stopping rule that samples until the threshold is crossed, so the reported strength of evidence is not the strength of evidence the record actually contains. A pre-registered stopping point, or a sequential rule with its thresholds set in advance, would have produced a different verdict from identical data. ^[10]

Mapped back: The epistemic task is estimating an effect; the accuracy-independent stake is authorship of the proposal; the discretionary stage is the stopping rule. The handling is sign-dependent — a congenial interim result terminates the search, an uncongenial one extends it — and the symmetric counterfactual, a stopping point fixed before the data were seen, yields the benchmark verdict the reported one departs from. No step is individually indefensible, which is exactly the signature.

Applied/industry

A manufacturer receives field reports of a component failing early in a small fraction of units. A recall would be expensive and would land three weeks before a launch the division has publicly committed to. An engineering review is convened and concludes the failures are not attributable to the component.

Trace the discretion. First, scope: the review considers reports filed under one warranty code, and reports of the same physical symptom filed under an adjacent code are not pulled. Second, admission: several reports are set aside as "user damage" on the basis of photographs that are ambiguous. Third, threshold: the test protocol requires failure within a stress window chosen to match nominal duty cycles, and the field failures cluster in units run above nominal. Fourth, stopping: when two test batches pass, testing ends; the standing practice for an anomaly under investigation had been three. Each choice was argued for on the day, and each argument was reasonable.

The sign-flip test is decisive here. Had the ambiguous photographs pointed toward a manufacturing defect rather than user damage, would they have been resolved the same way? Had two test batches failed rather than passed, would testing have stopped at two? The asymmetric-scrutiny form makes the same point from effort: the review generated four separate objections to the field data and none to its own protocol.

Mapped back: The task is a causal attribution with a determinate answer; the stake is a launch date and a liability exposure, both indifferent to what actually caused the failures; four discretionary stages — scope, admission, threshold, stopping — are each nudged toward the same sign; and the symmetric counterfactual is a review chartered by a party with no position in the launch. That the conclusion might still turn out to be correct is beside the point structurally: the process gives no warrant for it, which is precisely the distinction between the process claim and the verdict claim.

Structural Tensions

T1 — Unfalsifiability of motive attribution. Because the mechanism operates without the reasoner's awareness, denial is not evidence against the charge, and any argument offered in defense can be re-described as further motivated production. This makes the attribution rhetorically unbeatable and therefore epistemically suspect. The prime earns its keep only by paying the cost of falsifiability: the stake must be nameable before the reasoning is examined, and the symmetric counterfactual must be specifiable in advance. An attribution that cannot say what evidence would retract it is a weapon, not a diagnosis.

T2 — Rational source-discounting versus directional gating. Legitimate Bayesian practice makes surprising evidence harder to accept and unreliable sources easier to dismiss, so the very behavior that marks motivated reasoning is also the behavior of a well-calibrated agent. The two are separable only by the counterfactual, which is often unavailable: one rarely observes the same reasoner handling the mirror-image evidence. In practice the distinction rests on whether the discount tracks accuracy-relevant properties that were specified beforehand, and reasoners can supply such properties after the fact with disconcerting ease.

T3 — Accuracy motives that are themselves directional. The standard remedy for a directional stake is to install a competing accuracy motive — reputation for calibration, a scoring rule, an accountability audience. But wanting to be seen as accurate is itself an accuracy-independent stake, and it gates evidence in its own way: toward safe hedges, toward whatever the evaluator will score as correct, and away from positions that are right but unpopular with the grader. The cure inherits the disease's structure. What changes is the sign and the salience of the pressure, not its presence.

T4 — Collective versus individual bearers. A committee can display the full input-output signature with no member individually motivated: each contributes a defensible judgment, and the selection of whose judgment governs each stage does the gating. Conversely, a body of thoroughly motivated individuals with opposed stakes can aggregate to an unbiased verdict. So the prime's roles are satisfiable at either level, and reading a collective's asymmetry as evidence about its members — or a member's stake as evidence about the collective's output — is invalid in both directions.

T5 — The accuser is also motivated. Diagnosing motivated reasoning is itself an epistemic task performed by someone with a stake, usually a stake in the opposite conclusion. The diagnosis has abundant discretionary stages: which stopping points count as suspicious, which alternatives count as ruled out, whose stake is named. This generates a regress with no neutral terminus, and it is asymmetrically exploitable, since the party with more analytic resources can always produce the more detailed indictment. Reciprocal application — running the test on one's own record first — is the only available discipline.

T6 — Harnessed asymmetry versus corrupted process. Several institutions deliberately install directional stakes and rely on them: adversarial litigation assigns each side a conclusion to defend, competing laboratories are motivated to overturn each other, and a red team is paid to find fault. In these designs the individually gated process is a component of a collectively unbiased one, and removing the motive would degrade the output. The prime therefore cannot be applied as a blanket fault at the component level; whether gating is pathology or engineering depends on the aggregating structure it sits inside.

Structural–Framed Character

Motivated Reasoning sits on the framed side of the structural–framed spectrum, graded mixed-framed with an aggregate of 0.50 — every criterion reads exactly half, so the grade rests on a consistent tilt rather than one heavy reading. Underneath is a relation that travels: an agent or agent-like collective performing a truth-directed task, an accuracy-independent goal or stake, a directional difference in how evidence is accessed, generated, interpreted, weighted, or updated, and a counterfactual benchmark in which the same record is handled without that stake.

The criterion doing the most work is human-practice-bound, at half. An agent is written into the structural signature: something must perform an epistemic task and hold a directional preference, and preference alone is not motivated reasoning until it enters the epistemic pipeline. It stops at half because the bearer may be an agent-like collective, conscious intent is unnecessary, and the motive–evidence-gate–directional-belief roles hold across politics, organizational audits, science, risk assessment, memory, and moral judgment.

Vocabulary travels at half: motive, evidence-handling, and belief update carry a cognitive tint into whatever field borrows them. Evaluative weight is half because the prime is defined by departure from the verdict a symmetric rule would support: stated structurally, but naming a fault. Institutional origin is half, the paradigm cases sitting inside settings with explicit epistemic duties such as audits and science. Import-vs-recognize is half: politics imports the framing, an audit recognizes a directional gate already operating.

Mixed-framed means carrying the prime with its agentive frame acknowledged rather than filed off; cold fluency or efficiency effects with no preferred conclusion are excluded.

Substrate Independence

Motivated Reasoning is a highly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. The structure that travels is staged: a directional stake indifferent to accuracy, an evidence pipeline containing discretionary junctures where a defensible call could go either way, asymmetric gating at those junctures, and a verdict displaced from what an evenhanded rule applied to the same record would return. Those roles are recoverable without metaphor in identity-protective belief, partisan judgment, internal audit, defence of a research programme, risk assessment, selective recall, and moral evaluation. What holds it below the ceiling is a commitment the signature keeps: it presupposes an agent with a stake and slack in its own procedure, so no formal or physical system can instantiate it.

• Composite substrate independence — 4 / 5
• Domain breadth — 4 / 5
• Structural abstraction — 4 / 5
• Transfer evidence — 4 / 5

Relationships to Other Abstractions

Current abstraction Motivated Reasoning Prime

Parents (2) — more general patterns this builds on

• Motivated Reasoning is a kind of Bias Prime

  Motivated Reasoning is a strict specialization of Bias.

• Motivated Reasoning presupposes Preference Prime

  The child presupposes Preference.

Children (10) — more specific cases that build on this

• Linguistic intergroup bias Domain-specific is a kind of Motivated Reasoning

  The proposed strict upward parent is prime:motivated_reasoning.

• Wishful thinking Prime is a kind of Motivated Reasoning

  The accepted reference-grade review places Wishful thinking under Motivated Reasoning because the child instantiates or depends on the parent's broader structure while retaining its own constitutive identity.

• Black Elephant Domain-specific is part of Motivated Reasoning

  Motivated Reasoning is a constitutive part of Black Elephant.

▸ Show 7 more

Hierarchy paths (2) — routes to 2 parentless roots

• Motivated Reasoning → Bias

Show alternative path (1)

Neighborhood in Abstraction Space

Motivated Reasoning sits in a moderately populated region (57^th percentile for distinctiveness): it has near-neighbors but no dense thicket of synonyms.

Family — Argumentative Traps & Framing Fallacies (15 primes)

Nearest neighbors

• Confirmation Bias — 0.74
• Belief Formation — 0.71
• Inductive Reasoning — 0.71
• Deductive Reasoning — 0.70
• Assumption — 0.70

Computed from structural-signature embeddings · 2026-09-10

Not to Be Confused With

Motivated Reasoning must first be distinguished from Confirmation Bias, with which it is most often merged. Confirmation bias is anchored to what the reasoner already believes: hypotheses in hand attract supporting search and lenient evaluation, whatever the believer's stake in them. Its reference point is a prior; motivated reasoning's reference point is a preference. The two come apart cleanly in both directions. A researcher testing a hypothesis they hold but do not care about can show the full confirmatory pattern with no directional stake anywhere in the picture. Conversely, a person who wants a conclusion they do not currently believe will gate evidence toward it, which is motivated reasoning running against the prior rather than with it: the executive who hopes the disappointing quarter was an anomaly is not confirming a belief, they are recruiting one. They also coincide constantly, since people tend to want their existing beliefs vindicated, and that overlap is why the terms blur. The test that separates them is whether removing the stake, holding the prior fixed, changes the handling.

Nor is it Wishful Thinking, which names the direct route: belief adopted because the state of affairs is desired, with no evidential detour. Wishful thinking can operate on an empty record — a gambler certain the next hand will come good has processed nothing selectively, because there was nothing to process. Motivated reasoning is constitutively pipeline-mediated; it requires evidence to work on and discretionary stages to work through. The practical difference shows up in what shakes them. Wishful thinking is vulnerable to plain disconfirmation. Motivated reasoning is not, because the machinery that produced the belief also produces reasons to dismiss whatever arrives against it.

Strategic Deception shares motivated reasoning's directional stake and differs on where the distortion lands. The deceiver's own evidence-handling is intact — indeed accurate private belief is usually a prerequisite for effective deception, since one must know the truth to steer around it — and what is bent is the public representation. Motivated reasoning bends the private handling and may leave the public report a faithful statement of what the reasoner actually concluded. The two mix in practice: an advocate who began by arguing a case they did not believe can end up believing it, at which point sanctioning the deception no longer describes what is happening. The diagnostic is whether the actor's own decisions, made where no audience can see them, track the public claim or the private one.

Rationalization is close enough to be treated as a phase rather than a rival. It names the construction of acceptable reasons for a conclusion or action already fixed, and it is one stage of the pipeline — the justification stage — rather than the whole of it. Motivated reasoning includes upstream stages that rationalization does not touch: what gets searched for, what is admitted, what is remembered. A reasoner whose search and admission were gated may never need to rationalize, because the record they assembled genuinely supports the verdict. The relation is part-to-whole, not synonym-to-synonym.

Self-Deception overlaps heavily and makes a stronger claim. It requires that the agent in some sense possesses the disfavored truth while maintaining the favored belief, which raises the familiar puzzle of how one mind holds both. Motivated reasoning needs nothing of the kind: the gating happens at intake and evaluation, so the disfavored conclusion may never be formed.

Finally, motivated reasoning is a species of Bias, its parent, and must not be used as a synonym for it. Bias is any systematic directional displacement of a process from its target, and most biases have no motive at all: anchoring, availability, and rounding error displace judgment in ways no preference selected. Motivated reasoning is the proper subset in which the direction is set by a non-epistemic stake and the displaced process is evidence handling specifically. It also presupposes Preference, without being it. Preference is an ordering over alternatives, a static relation; motivated reasoning is the process that consumes such an ordering as its directional input. A preference that never enters the epistemic pipeline leaves belief exactly where it found it.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.

References

[1] Kunda, Ziva. "The Case for Motivated Reasoning". Psychological Bulletin, 1990. Establishes that directional goals steer which evidence is accessed and how it is evaluated, within the limits of plausible justification, and unifies the separately named biased-processing effects under one mechanism. registry ↩a ↩b ↩c ↩d ↩e

[2] Ditto, Peter H., and David F. Lopez. "Motivated Skepticism: Use of Differential Decision Criteria for Preferred and Nonpreferred Conclusions". Journal of Personality and Social Psychology, 1992. Shows asymmetric quantity of processing, with more checking and later stopping for unwelcome results, while participants experience the process as ordinary care. registry ↩a ↩b

[3] Lord, Charles G., Lee Ross, and Mark R. Lepper. "Biased Assimilation and Attitude Polarization: The Effects of Prior Theories on Subsequently Considered Evidence". Journal of Personality and Social Psychology, 1979. Demonstrates one methodological standard applied at two stringencies according to which way a study's result points, supplying the symmetric-handling comparison a diagnosis requires. registry ↩

[4] Walton, Douglas. Ad Hominem Arguments. University of Alabama Press, 1998. Analyses the circumstantial and bias forms of ad hominem, in which an arguer's stake is offered in place of an answer to the argument, and sets out when that move is a legitimate credibility challenge rather than a refutation. registry ↩

[5] Moore, Don A., and George Loewenstein. "Self-Interest, Automaticity, and the Psychology of Conflict of Interest". Social Justice Research, 2004. Argues that self-interest bends professional judgement automatically and outside awareness wherever a role carrying an explicit duty also leaves the judgement discretionary. registry ↩

[6] Bazerman, Max H., George Loewenstein, and Don A. Moore. "Why Good Accountants Do Bad Audits". Harvard Business Review, 2002. Documents unconscious self-serving bias in professional judgement where assumption-setting is discretionary and the judge has a stake in the outcome. registry ↩

[7] Taber, Charles S., and Milton Lodge. "Motivated Skepticism in the Evaluation of Political Beliefs". American Journal of Political Science, 2006. Finds that greater sophistication increases rather than reduces directional bias, because expertise supplies more counterarguments to deploy selectively. registry ↩

[8] Druckman, James N., and Mary C. McGrath. "The Evidence for Motivated Reasoning in Climate Change Preference Formation". Nature Climate Change, 2019. Shows that discounting a source on prior credibility beliefs is accuracy-seeking Bayesian updating rather than directional gating, and that the two are separable only against a benchmark stated in advance. registry ↩

[9] Nosek, Brian A., Charles R. Ebersole, Alexander C. DeHaven, and David T. Mellor. "The Preregistration Revolution". Proceedings of the National Academy of Sciences, 2018. Frames pre-commitment as a countermeasure defined over pipeline stages rather than over individual minds, transferable across research settings. registry ↩

[10] Simmons, Joseph P., Leif D. Nelson, and Uri Simonsohn. "False-Positive Psychology: Undisclosed Flexibility in Data Collection and Analysis Allows Presenting Anything as Significant". Psychological Science, 2011. Quantifies how sampling until a threshold is crossed invalidates a fixed-n significance threshold and inflates the apparent strength of evidence. registry ↩