Wason Selection Task¶

A four-case conditional-rule probe asks which partially observed cases must be inspected to find the uniquely falsifying conjunction, exposing how logic, interpretation, and context shape human selections.

Version: v2 · 2026-09-06 · History
Domain-specific #: 3092
Origin domain: cognitive psychology
Subdomain: psychology of reasoning
Aliases: Wason four-card task, Four-card problem, Four-card selection task

Core Idea¶

The Wason selection task is a controlled experimental paradigm for studying how people interpret and test conditional rules. A participant sees a rule of the form if P, then Q and four partially observed cases displaying P, not-P, Q, and not-Q. Each case has a hidden feature from the other category. The participant chooses which cases must be inspected to determine whether the rule is violated. Under the canonical universal material-conditional interpretation, the only falsifying configuration is P and not-Q; therefore the P and not-Q cases must be inspected, while not-P and Q cannot reveal a violation merely by being turned.^[1]

The paradigm is more than the conditional truth table rendered on cards. Its scientific identity joins a normative target with an observation protocol: a stated conditional, a deliberately incomplete display, a choice of information to reveal, a scoring rule, a participant response, and experimental manipulations of wording, negation, content, perspective, or goal. The resulting selection pattern is then evidence to be explained. Wason's foundational work treated difficulty with the contrapositive-relevant case as a problem in reasoning about a rule; later work showed that the same surface error cannot safely be assigned to one mechanism. Lexical matching, semantic interpretation, pragmatic schemas, expected information gain, utilities, and social-rule framing can each change which selections are made or what task the participant is effectively solving.^[2]^[3]^[4]^[5]

This stable role package makes the task an autonomous domain-specific abstraction. It recurs across abstract letter-number versions, concrete descriptive rules, permissions, obligations, social contracts, perspective manipulations, negated rules, and computerized variants. Yet it does not clear the prime bar. Remove the experimental participant, card-like partial observations, response coding, and psychology-of-reasoning question, and the remaining skeleton is conditional rule checking through possible counterexamples. That portable structure is already carried by Deductive Reasoning and Verification. “Wason selection task” properly names the cognitive-science instrument and research tradition built around that skeleton.

Structural Signature¶

A canonical Wason selection task contains these roles:

the conditional rule — normally expressible as P -> Q, with its intended reading declared or experimentally manipulated;
the case universe — objects that each bear one value from an antecedent partition (P or not-P) and one from a consequent partition (Q or not-Q);
the partial display — four visible representatives, one for each of P, not-P, Q, and not-Q, while the paired attribute of each remains hidden;
the inspection action — turning, querying, or otherwise revealing a selected case's hidden attribute;
the falsifier — for a universal material conditional, the conjunction P and not-Q, the only truth-table row in which P -> Q is false;
the normatively necessary selection — inspect P for hidden not-Q and inspect not-Q for hidden P;
the irrelevant selections — not-P cannot violate a rule that says nothing about cases lacking P, and Q does not imply P;
the response record — which cases the participant selects, often with order, latency, confidence, or explanation as additional measurements;
the scoring and interpretation policy — a declared criterion for correctness plus a theory of what the observed choice does and does not diagnose; and
the experimental framing — abstract, descriptive, deontic, permission, social-exchange, negated, or perspective-specific content that may alter comprehension and action goals.

The locked canonical chain is:

conditional P -> Q + four one-sided cases {P, not-P, Q, not-Q} -> choose information to reveal -> inspect P and not-Q for P and not-Q counterexamples -> score the choice -> compare selection patterns across framings.

The formal core can be shown compactly:

Visible case	Hidden value that matters	Can it falsify `P -> Q`?	Inspect?
`P`	`not-Q`	yes	yes
`not-P`	either `Q` or `not-Q`	no	no
`Q`	either `P` or `not-P`	no	no
`not-Q`	`P`	yes	yes

This table is the canonical material-conditional scoring rule, not a claim that every natural-language conditional is understood or should be modeled that way. In deontic versions, participants may be checking violations of an obligation from a particular perspective rather than evaluating a descriptive law. The four-case architecture can remain recognizable while the operative semantics and utilities change.^[4]^[5]

What It Is Not¶

Not the 2-4-6 rule-discovery task. Both were devised by Wason and both concern hypothesis testing, but 2-4-6 asks participants to infer an unknown rule by proposing sequences. The selection task supplies a conditional rule and asks which existing cases to inspect.
Not a material conditional itself. P -> Q is a logical relation. The Wason task is an experimental arrangement that presents a conditional, limits visible information, records selections, and supports contrasts among psychological explanations.
Not generic conditional reasoning. Conditional reasoning also includes modus ponens, modus tollens, conditional proof, probabilistic conditionals, causal conditionals, and everyday inference without a four-case selection protocol.
Not confirmation bias by definition. Selecting P or P and Q was historically described as verification seeking, but matching-bias experiments showed that selection can follow lexical values named in the rule even when negation reverses their logical role.^[2] The response pattern alone does not uniquely identify confirmation bias.
Not congruence bias. Congruence bias concerns designing probes that fit a favored hypothesis while failing to distinguish rivals, canonically illustrated by 2-4-6. The Wason task gives a fixed array and measures which cases a participant requests.
Not the social-contract account. Social exchange is one influential explanation and content family. Permission-schema, relevance, semantic, decision-theoretic, and other accounts also predict content effects, and the task exists independently of any one theory.^[3]^[6]^[7]
Not every four-card puzzle. Four cards are insufficient. The cards must instantiate complementary antecedent and consequent values under a conditional-testing instruction, with hidden attributes and a determinate inspection criterion.
Not the Monty Hall problem. Monty Hall studies conditional probability under informed host behavior and switching. Wason studies selection of partially hidden cases under a conditional rule.

Scope of Application¶

The home domain is experimental cognitive psychology, especially the psychology of reasoning. Researchers use the task to study how people select evidence for conditional claims, how linguistic form guides attention, how descriptive and deontic readings differ, and how knowledge, perspective, goals, and content change performance. Wason's original experiments centered on the difficulty of making the transformation needed to inspect the contrapositive-relevant case. Evans and Lynch later varied affirmative and negative components, finding strong matching tendencies and weakening the simple claim that ordinary responses reveal a single verification bias.^[1]^[2]

The task also supports research programs that disagree about what counts as the right cognitive model. Cheng and Holyoak used permission framings to argue for pragmatic reasoning schemas. Cosmides used social-exchange variants to test hypotheses about specialized reasoning for detecting violations of social contracts. Oaksford and Chater modeled card choice as Bayesian optimal data selection rather than as failed Popperian falsification, distinguishing the logical rule-checking benchmark from an information-gain account of behavior. Stenning and van Lambalgen emphasized interpretation and the difference between descriptive and deontic logical forms. Sperber and Girotto argued that relevance-guided comprehension and variant-task confounds limit the paradigm's inferential reach.^[3]^[6]^[4]^[5]^[7]

The abstraction therefore covers both the canonical task and controlled variants that preserve its recognition conditions: a conditional rule, complementary visible cases, selective access to hidden values, a participant choice, and a scoring or model-comparison policy. It does not cover ordinary survey questions about conditionals, unconstrained proof problems, free-form hypothesis generation, or card puzzles lacking the conditional-falsifier architecture.

Clarity¶

The Wason selection task clarifies reasoning research by separating four questions that conversational descriptions often collapse.

First is the normative question: which observations could reveal the counterexample under the stipulated interpretation? For a universal material conditional, the answer is fixed by the P and not-Q falsifier. Second is the behavioral question: which cards did participants actually choose? Third is the interpretive question: what did participants take “if” and the task instruction to mean? Fourth is the explanatory question: which cognitive process best accounts for their selections?

This decomposition prevents a common invalid inference: “the participant did not choose P and not-Q, therefore the participant cannot reason deductively.” A noncanonical response may arise from lexical matching, a biconditional reading, a pragmatic interpretation, a different goal, an information-selection policy, an attention failure, or a derivational failure. The task makes those possibilities experimentally tractable because wording, negation, context, perspective, and response can be manipulated independently. It is a diagnostic platform, not a one-response mind-reading device.

Manages Complexity¶

The paradigm compresses an otherwise sprawling inquiry into a small factorial design. A full conditional truth table, partial observability, information choice, and falsification criterion are represented by four cases. Researchers can then change one dimension while holding the others approximately fixed: affirmative versus negative wording, abstract versus thematic content, descriptive versus deontic interpretation, familiar versus unfamiliar rules, or rule-enforcer versus rule-subject perspective.

That compression makes competing accounts comparable. A verification-bias account predicts selection oriented toward apparent confirmation. A matching-bias account predicts choices aligned with values explicitly named in the rule, including predictable reversals under negation. A pragmatic-schema account predicts facilitation when a permission structure is evoked. A social-exchange account predicts sensitivity to benefit-and-requirement relations and cheater detection. A Bayesian account ranks cards by expected information gain under assumptions about environmental probabilities. A semantic account asks whether the participant assigned the experimenter's intended logical form at all. The task does not settle among these accounts by itself, but it creates a common apparatus in which their divergent predictions can be tested.

Abstract Reasoning¶

The task licenses several disciplined inferences.

Diagnostic inference: identify the truth-defeating configuration before selecting evidence. For P -> Q, write the sole falsifier P and not-Q; then choose visible cases whose hidden side could complete that conjunction. This derives P and not-Q without relying on a memorized card answer.

Boundary inference: if a purported Wason variant lacks complementary antecedent/consequent cases, selective hidden information, or a conditional-testing instruction, it is merely a related reasoning problem. If the rule is deontic, causal, probabilistic, or biconditional, the analyst must state the intended semantics before importing the material-conditional score.

Experimental inference: a content effect is evidence that framing matters, but not by itself evidence for a specific mechanism. Competing explanations must predict contrasts that separate semantic interpretation, lexical matching, pragmatic schema activation, expected information gain, and specialized inference.

Intervention inference: make the counterexample criterion explicit when the goal is instruction; vary negation, content, and perspective when the goal is mechanism identification. Better performance after a thematic rewrite may mean that the intended relation became easier to interpret, that a familiar action schema was evoked, or that the utilities of violations became salient. The intervention must be tied to the hypothesized locus.

Predictive inference: under a fixed material-conditional reading, not-P remains irrelevant even when it is perceptually salient; under lexical matching, changing positive to negative wording can change selections without changing logical equivalence; under deontic framing, the violation-search goal can make the relevant cases easier to identify. These are testable pattern predictions, not claims that every participant follows one model.

Knowledge Transfer¶

Exact transfer occurs within reasoning research when the complete apparatus is retained. A letter-number task, an age-and-beverage permission task, and a computerized rule-enforcement task can all be genuine Wason variants if they preserve two complementary dimensions, one-sided observations, selective inspection, a conditional rule, a response record, and an explicit scoring interpretation. This lets results be compared across content while keeping the task family recognizable.

Outside that domain, only the skeletal lesson transfers safely: when testing a universal conditional, inspect cases capable of exposing the antecedent together with failure of the consequent. A quality auditor checking “if a record is privileged, it must be encrypted” should inspect privileged records for non-encryption and non-encrypted records for privilege. That is conditional verification by counterexample. It becomes a Wason task only if arranged as a selective-information reasoning probe with responses being measured.

This boundary matters because loose borrowing can confuse an application of Deductive Reasoning or Verification with the psychological paradigm. The Wason name should travel when the experimental role graph travels, not whenever someone mentions falsification, cards, or a conditional rule.

Examples¶

Canonical abstract case¶

Four cards show A, D, 4, and 7. Each card has a letter on one side and a number on the other. The rule is: “If a card has a vowel on one side, then it has an even number on the other.” Let P mean vowel and Q mean even.

Turn A (P) to see whether the hidden number is odd (not-Q).
Do not turn D (not-P); the rule makes no claim about consonants.
Do not turn 4 (Q); an even number may have either a vowel or consonant on the reverse.
Turn 7 (not-Q) to see whether the hidden letter is a vowel (P).

The required selection is therefore A and 7. Choosing A and 4 feels confirmatory but leaves the only falsifying configuration insufficiently checked.

Deontic permission case¶

Four records show “drinking beer,” “drinking soda,” “age 16,” and “age 25.” The rule is: “If a person is drinking beer, then the person must be at least 18.” To police violations, inspect the beer drinker for an under-18 age and the 16-year-old for beer. The soda drinker and 25-year-old cannot reveal a violation simply from those visible faces.

The formal selection positions resemble P and not-Q, but the task now invokes permission and enforcement. Cheng and Holyoak showed that permission schemas can facilitate selection even with abstract content, while later semantic and decision-theoretic work emphasizes that deontic rules may be interpreted as action-guiding constraints rather than descriptive material conditionals.^[3]^[4]^[5] The example demonstrates why identical card choices do not prove identical mental computations.

Negation manipulation¶

Suppose the consequent is phrased negatively: “If P, then not-R.” The falsifier is now P and R, so the cases showing P and R must be inspected. Evans and Lynch varied positive and negative components and found selections tracked matching values, providing evidence that surface lexical form can guide choice independently of a simple confirmation-seeking story.^[2] The example preserves the paradigm while changing which visible token plays the falsifier-completing role.

Structural Tensions¶

Normative logic versus interpreted task. The material-conditional solution is exact once the semantics are fixed, yet natural-language “if” can be read causally, biconditionally, probabilistically, or deontically. Scoring without documenting interpretation mistakes disagreement about the task for failure at the task. Diagnostic: was the participant's derivation invalid under an agreed logical form, or was a different logical form assigned before derivation began?

Stable architecture versus variant drift. The paradigm's productivity comes from modifying content, negation, perspective, and goal, but a sufficiently altered variant may no longer pose the same information-selection problem. Sperber and Girotto's critique turns on the danger of mixing a true selection task with a simpler categorization task.^[7] Diagnostic: does the variant still require choosing which hidden attributes to reveal to test a conditional, or has it made the decisive category directly visible?

Clean benchmark versus theoretical nonuniqueness. A four-card response is easy to score, which invites a strong psychological diagnosis. Yet the same selection may fit multiple mechanisms. Diagnostic: does the study contain a contrast on which the candidate explanations make different predictions, or only a response compatible with all of them?

Falsification versus information gain. Classical scoring asks which cards are logically necessary to detect a counterexample. Bayesian optimal-data-selection models ask which observation is expected to reduce uncertainty most under environmental priors and may assign information value beyond the strict falsifiers.^[4] Diagnostic: is the experiment evaluating truth-functional rule compliance or adaptive information choice under a stated probability model?

Content facilitation versus changed computation. Improved performance with permissions or social exchange can indicate access to useful knowledge, a specialized inference mechanism, a changed semantic representation, or changed utilities. Diagnostic: has content merely made the same derivation easier, or has it changed what relation and goal the participant represents?

Structural–Framed Character¶

The Wason selection task is mixed and framed-leaning. Its logical kernel is strongly structural: once P -> Q is given a material-conditional reading, the unique falsifier and required inspection classes follow independently of culture or researcher preference. The four-role counterexample search can be recognized in any equivalent symbol system.

The named abstraction nevertheless remains framed by experimental practice. Researchers choose the materials, instructions, response format, and correctness policy; participants interpret ordinary language in context; and the task's scientific significance comes from comparing human selections across controlled framings. Its vocabulary—participant, card face, selection, content effect, permission schema, matching bias, response rate—is at home in cognitive psychology. Importing the label into auditing or software testing without a measured reasoner and partial-information selection protocol is analogy, not literal recurrence.

The structural core explains the task's enduring precision, while the framed accent explains its theoretical controversy. A small formal skeleton supports many empirical readings because the act of interpreting the instruction lies between presentation and deduction.

Structural Core vs. Domain Accent¶

The structural core is conditional counterexample search under partial observability: state a universal P -> Q, identify P and not-Q as the defeating configuration, and reveal only cases capable of instantiating it. This skeleton is portable across rule checking, compliance audits, mathematical counterexample search, and property testing. It is already represented more generally by Deductive Reasoning and Verification.

The domain accent is what turns that skeleton into the Wason selection task: a human or reasoning agent receives a four-case display; only one side of each case is visible; selections are recorded rather than merely executed; errors and latencies become dependent variables; wording and content are deliberately varied; and competing cognitive theories explain the resulting pattern. Those roles do not disappear without changing the identity. A database query that searches for P and not-Q implements the logical core perfectly but is not thereby performing the psychological task.

The candidate therefore fails prime qualification. Its generic residue is already covered, while its autonomous contribution is a named, reusable experimental paradigm inside the psychology of reasoning. The same architecture recurs across many studies, but its intact transfer is among studies and reasoning agents, not among arbitrary physical or institutional substrates.

Deductive Reasoning is the minimal prerequisite. The canonical score presupposes truth-preserving reasoning about a conditional and its counterexample; the proposed DAG edge records this without treating the experimental task as a subtype of deduction itself.
Verification captures the portable object-specification-procedure-evidence-verdict pattern. The Wason task specializes the evidence-selection problem but adds participant behavior and cognitive interpretation.
Experimental Design explains how wording, content, perspective, and negation become controlled manipulations. It is a methodological relation rather than the smallest ancestry claim.
Confirmation Bias is historically related but not constitutive. A Wason response cannot be assigned to confirmation bias without contrasts ruling out matching, interpretation, goal, and information-gain accounts.
Deductive Reasoning versus Abductive Reasoning. The frozen semantic rematch ranked Abductive Reasoning highest, but that is a retrieval false neighbor. The task does not generate the best explanation for an observation; it selects cases that can test a given conditional.

Relationships to Other Abstractions¶

Current abstraction Wason Selection Task Domain-specific

Parents (1) — more general patterns this builds on

Wason Selection Task presupposes Deductive Reasoning Prime

Deductive Reasoning is the minimal prerequisite.

Hierarchy path (1) — routes to 1 parentless root

Wason Selection Task → Deductive Reasoning

Neighborhood in Abstraction Space¶

Wason Selection Task sits in a sparse region of the domain-specific corpus (92^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Inference Bias & Multiple Testing (10 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Wason's 2-4-6 task: unknown-rule discovery through self-generated triples, not fixed-rule case selection.
Material conditional: the normative relation often used to score the canonical task, not the experimental apparatus.
Conditional reasoning: the broader family of inferences involving “if”; the selection task is one measurement paradigm within it.
Confirmation bias: biased evidence search or evaluation; one proposed explanation of some responses, not the task's identity.
Matching bias: a lexical selection tendency revealed by negation manipulations; a response mechanism studied with the task, not a synonym for it.
Congruence bias: choosing non-discriminating tests for a favored hypothesis; most closely tied to the separate 2-4-6 paradigm.
Permission schema or social-contract reasoning: content-sensitive accounts and task variants, not definitions of the whole paradigm.
Monty Hall problem: probabilistic updating under an informed host's constrained reveal, not conditional falsifier selection.
A generic logic puzzle: the Wason task is a reproducible research design with hidden paired attributes, selective inspection, response coding, and model-comparison uses.

References¶

[1] P. C. Wason, “Reasoning about a Rule”, Quarterly Journal of Experimental Psychology 20, no. 3 (1968): 273–281. Bibliographic record also available through PubMed. registry ↩a ↩b

[2] J. St. B. T. Evans and J. S. Lynch, “Matching Bias in the Selection Task”, British Journal of Psychology 64, no. 3 (1973): 391–397. registry ↩a ↩b ↩c ↩d

[3] P. W. Cheng and K. J. Holyoak, “Pragmatic Reasoning Schemas”, Cognitive Psychology 17, no. 4 (1985): 391–416. registry ↩a ↩b ↩c ↩d

[4] M. Oaksford and N. Chater, “A Rational Analysis of the Selection Task as Optimal Data Selection”, Psychological Review 101, no. 4 (1994): 608–631. registry ↩a ↩b ↩c ↩d ↩e

[5] K. Stenning and M. van Lambalgen, “A Little Logic Goes a Long Way: Basing Experiment on Semantic Theory in the Cognitive Science of Conditional Reasoning”, Cognitive Science 28, no. 4 (2004): 481–530. registry ↩a ↩b ↩c ↩d

[6] L. Cosmides, “The Logic of Social Exchange: Has Natural Selection Shaped How Humans Reason? Studies with the Wason Selection Task”, Cognition 31, no. 3 (1989): 187–276. registry ↩a ↩b

[7] D. Sperber and V. Girotto, “Use or Misuse of the Selection Task? Rejoinder to Fiddick, Cosmides, and Tooby”, Cognition 85, no. 3 (2002): 277–290. registry ↩a ↩b ↩c