Cue Validity¶
A cue's validity for a category is the chance of category membership among objects bearing that cue in a stated comparison population.
Core Idea¶
Cue validity asks how strongly observing a feature predicts membership in a category. For a category c, cue f and declared comparison population, it is the conditional probability P(c|f). Rosch and colleagues' original account says a cue becomes more valid for c when it appears with c and less valid when it also appears with other categories. That direction is essential: P(f|c), the fraction of category members with the cue, answers a different question. A feature can be common among chairs and also common among many non-chairs, so it may describe chairs well but identify them poorly.[1][2]
The original basic-object work also considers a category-level sum of the cue validities of its attributes. That sum is an aggregate score, not itself a probability—it can exceed one. A proposed P(c|f)−P© adjustment can serve as a diagnostic; such a baseline subtraction can be useful, but it is not interchangeable with Rosch's defining conditional probability. Moreover, later original analysis identifies a serious nested-category limit: a cue-bearing bird is also an animal, so P(animal|wings) cannot be smaller than P(bird|wings). Raw single-cue validity alone does not prove that an intermediate “basic level” such as bird always maximizes the metric.[1][2]
Structural Signature¶
Sig role-phrases: comparison population; target category; observable cue; conditional direction; competing categories; optional aggregate convention.
- Reference population: fix what objects are counted. Conditional probability changes with the mix of categories in the sample; the same cue is not intrinsically diagnostic without that base.
- Feature f: define an observable property and whether it is present. Vague terms such as “has a typical shape” need an operational coding rule before counting.
- Category c: identify the target membership and any nesting. Bird and animal are different targets for the same wings cue.[2]
- Direction: count category members among cue-bearing objects. Do not substitute P(f|c), the prevalence of f among category members.
- Summary: if comparing whole categories, declare the attributes selected and how their validities are summed or weighted; a total score is no longer a conditional probability.[1]
What It Is Not¶
Cue validity is not category validity P(f|c), nor raw prevalence P(f). It is not a causal claim that f makes an object a c; correlation in a chosen population is enough for the measure. It is not a universal proof that one basic category level is optimal. Corter and Gluck show that nested superordinates can dominate a raw conditional criterion even when people most naturally name an intermediate category. Their proposed category-utility framework is related but distinct; this entry does not rebrand it as “cue validity.”[2]
Scope of Application¶
The measure is literal when people or models use features to classify objects within a specified set of alternatives. Rosch et al. studied natural object categories, including basic-level names contrasted with superordinates such as furniture and vehicle; the use of cue validity was one proposed account of their differentiation. The original paper's abstract reports basic categories as highly informative/differentiated, but this should be read as the paper's research claim rather than a universal mathematical corollary of P(c|f). Later critique matters because the same hierarchy can make the unadjusted probability increase toward a parent category.[1][3][2]
Clarity¶
A clear report gives counts or an explicit model. If 90 of 100 objects have legs, and only 18 of those 90 are chairs, P(chair|legs)=18/90=0.20. If there are 20 chairs and 18 have legs, P(legs|chair)=18/20=0.90. The first says legs weakly identify chairs in this constructed population; the second says legs are common on chairs. The example is invented to expose the conditional reversal, not presented as Rosch's experimental data. Changing the sampling frame changes the first result. Summing the validities for several features may help describe category differentiation, but the sum must not be interpreted as “110% probability.”[1][2]
Manages Complexity¶
A cue-to-category conditional compresses many category–feature co-occurrences into a number useful for comparing candidate cues. It directs attention to outside-category competition instead of only asking whether a feature is frequent inside the target. That compression discards how a cue interacts with other cues, how features were elicited or coded, and which categories are nested. The later original critique shows why a single scalar cannot by itself settle cognitive “basicness”: the superordinate inclusion relation mechanically biases P(c|f). The measure manages feature selection best when the population, target contrast and inference task are retained beside the number.[2]
Abstract Reasoning¶
Fix a universe of objects and define c and f. Form a two-by-two count of category membership versus feature presence, calculate P(c|f)=count(c and f)/count(f), then compare with competing categories in the same universe. Separately calculate P(f|c) if the question is how well the category predicts the feature. If the task is basic-level selection across a hierarchy, test whether the metric is biased toward inclusive or specific nodes; do not declare an optimal level from one raw conditional. A baseline-adjusted score P(c|f)−P© asks about improvement over prior prevalence and must be labeled as a variant.
Knowledge Transfer¶
The probability direction transfers from natural categories to designed feature classifiers and linguistic category learning, provided the observer, population and target label are specified. What does not transfer automatically is Rosch's empirical basic-level interpretation; new domains can have different hierarchies and frequency distributions. Wings-for-bird illustrates the same mathematics as legs-for-chair, but the latter numerical data here are constructed. “Cue” can also mean a causal prompt or experimental instruction; the present identity is specifically predictive feature-to-category evidence.
Examples¶
Wings, bird and animal: source-attested hierarchy boundary. Corter and Gluck explicitly use wings to explain P(bird|wings), then note that because birds are animals, P(animal|wings) is at least as high under the same sample. Mapped back: cue wings is held fixed; category target changes from bird to its parent; the conditional favors the inclusive category even though bird can be a basic-level name. This is not an empirical count of winged objects; it is a set-inclusion consequence in their original analysis.[2]
Picture versus black ink for tabloid newspapers: source-attested hypothetical worked case. Corter and Gluck explicitly posit a newspaper population in which tabloids are 33%, 90% of tabloids have front-cover pictures, 60% of all newspapers have them, and every newspaper uses black ink. Their paper uses these values for a category-utility illustration, not an empirical survey. Applying the cue-validity conditional to those stated assumptions gives P(tabloid|picture)=0.9×0.33/0.6=0.495, whereas P(tabloid|black ink)=0.33. Mapped back: the target is tabloid in the same newspaper population; the picture and ink are two specified cues; one raises the target probability above its 0.33 base rate, while ubiquitous ink does not. The conditional results are our Bayes derivation from the paper's hypothetical inputs, not numbers printed as cue validity in that paper.[2]
Constructed reversal check. In an invented population of 100 items, 20 are chairs; 18 chairs and 72 non-chairs have legs. Then P(chair|legs)=18/90=0.20 but P(legs|chair)=18/20=0.90. Mapped back: high within-category feature coverage can coexist with low between-category diagnosticity. This calculation is an explicit thought experiment, not a reported study.
Structural Tensions¶
Coverage versus discrimination. A broad cue may be present on most members of c, improving P(f|c) and reducing misses, but if it is also widespread outside c it gives poor P(c|f). A highly distinctive cue can raise P(c|f) but occur on fewer category members. The two conditional directions therefore represent competing desiderata in selecting a feature for recognition or description; neither subsumes the other. Corter and Gluck's original comparison makes this distinction explicit and motivates combined measures for category utility.[2]
Diagnostic: With the same coded population, compare P(f|c) and P(c|f) for the candidate cue; does better coverage sacrifice discrimination against the alternatives?
Structural–Framed Character¶
The probability relation is mathematically structural once events and a probability measure are fixed. The cue validity of a real category, however, is strongly framed by human classification practice: what counts as “chair,” which objects enter the sample, and which feature a participant notices determine the measured conditional. Its evaluative weight is task-relative. A high value can be useful for recognition but says nothing by itself about whether a category is socially desirable or conceptually correct. Human practice is not needed for the conditional-probability theorem, but is needed to select categories and code cues in Rosch's cognitive application.
The vocabulary comes through probabilistic categorization and Rosch's original basic-object research, with later experimental and theoretical debate about what single conditionals can explain.[1][2] It travels to machine classification literally only when the target and reference population are specified; transferring Rosch's basic-level cognitive finding merely because a model computes P(c|f) would import the framing without evidence. Its character: a structural conditional measure embedded in empirical, convention-sensitive category practice.
Structural Core vs. Domain Accent¶
The skeletal relation is conditional prediction from observed event f to target event c. The domain-bound mechanism is a feature of a classified object predicting membership in a cognitive category within a comparison population. The named single-cue entry fails the prime bar because that cue/category orientation, coding and cognitive application are constitutive; abstract Bayes conditioning alone is far broader. Its strict parent is the live Conditional Probability prime. A category-level sum over several cues remains a related score, not a probability-valued instance of this child.
Instantiates / Related Primes¶
This entry is a kind of Conditional Probability.
Conditional Probability is the strict live prime parent of the single-cue P(c|f) measure. Category validity P(f|c) and category utility are neighboring measures, not synonyms or automatic parents.
Relationships to Other Abstractions¶
Current abstraction Cue Validity Domain-specific
Parents (1) — more general patterns this builds on
-
Cue Validity is a kind of Conditional Probability Prime
Single-cue validity is the conditional probability of a target category given an observed feature in a stated population.Each admitted single-cue validity is P(category|feature), a conditional probability with category membership as the target event and feature presence as the conditioning event. Those fixed cognitive-categorization roles distinguish it from arbitrary conditional probabilities. A sum of multiple cue validities is a related score, not an instance of this probability-valued child.
Hierarchy paths (2) — routes to 2 parentless roots
- Cue Validity → Conditional Probability → Probability → Measure → Aggregation → Micro Macro Linkage
- Cue Validity → Conditional Probability → Probability → Measure → Set and Membership
Neighborhood in Abstraction Space¶
Cue Validity sits in a moderately populated region (58th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Empirical Measurement & Statistical Inference Methods (50 abstractions)
Nearest neighbors
- Visual Capture — 0.86
- Join Count Statistic — 0.86
- Scientific Hypothesis — 0.85
- Occupancy–Abundance Relationship — 0.85
- Basic Category — 0.84
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
- Category validity P(f|c): prediction of a feature from the category, the reverse direction.[2]
- Feature frequency P(f): may be high even when cue diagnosticity is low.
- Category-level summed cue score: not a probability and can exceed one.[1]
- Baseline-adjusted P(c|f)−P©: a useful variant, not the original definition.
- Basic-level preference: an empirical/cognitive claim not guaranteed by raw nested probabilities.[2]
References¶
[1] Rosch, Mervis, Gray, Johnson and Boyes-Braem, “Basic Objects in Natural Categories” (1976), original paper, “Cue Validity” discussion; scanned PDF search-indexed but direct web-text access failed. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g
[2] James E. Corter and Mark A. Gluck, “Explaining Basic Categories: Feature Predictability and Information”, original analysis, pp.291–292 on cue/category-validity asymmetry. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m
[3] Original publisher record and abstract, Cognitive Psychology 8 (1976). registry ↩