Imprecise probability¶

Represent incomplete probabilistic commitment by a coherent set of admissible probability measures or equivalent lower and upper expectations, so conclusions expose a range and distinguish robust agreement from decisions that depend on an unresolved model choice.

Version: v2 · 2026-08-30 · History
Domain-specific #: 2047
Origin domain: probability and decision theory
Subdomain: imprecise uncertainty models

Core Idea¶

Imprecise probability is a family of uncertainty models that permits a set of probability measures or noncoincident lower and upper probabilities or expectations instead of forcing one fully specified additive distribution; envelope models use \(\underline P(A)=\inf_{P\in\mathcal K}P(A)\) and \(\overline P(A)=\sup_{P\in\mathcal K}P(A)\) for a credal set \(\mathcal K\).^[1]^[1] partial commitments define a feasible family of precise laws, and lower and upper envelopes propagate the least and greatest compatible expectation through queries; a narrow interval records strong consensus, a wide interval records unresolved alternatives, and robust conclusions are those shared across the admissible family.

Its autonomous residual is deliberate coherent partial specification of probability through a family or lower-upper envelope, including explicit width and update semantics, rather than ordinary sampling uncertainty, one approximate probability, a fuzzy membership grade, or an unstructured admission of ignorance. The identity fails when lower bounds exceed upper bounds, conjugacy or normalization fails under the chosen theory, endpoints are asserted without warrant, dependence assumptions change silently, interval width is confused with confidence coverage, one distribution is arbitrarily selected and the unresolved set forgotten, or a decision rule is inferred from bounds alone.

Recognition requires an analyst to state the sample space and represented uncertainty, identify whether the primitive object is a credal set, lower prevision, capacity, p-box, belief function, or another typed model, verify coherence and conjugacy assumptions, declare conditioning and independence, and keep epistemic imprecision distinct from numerical error. Once established, it supports representing scarce, conflicting, interval, set-valued, or partially elicited knowledge; sensitivity and robust Bayesian analysis; propagating probability bounds; separating agreement across models from model-sensitive conclusions; and making forced precision visible without turning those uses into the definition.

Structural Signature¶

Carrier: a measurable possibility space, a nonempty class of admissible probability measures or assessments, and events, gambles, or random quantities evaluated relative to that class
Inputs or antecedent state: partial probability constraints, lower and upper probabilities or previsions, a credal set representation, coherence conditions, conditioning rule, dependence assumptions, elicitation or data uncertainty, and a decision criterion if action is considered
Constitutive operation: partial commitments define a feasible family of precise laws, and lower and upper envelopes propagate the least and greatest compatible expectation through queries; a narrow interval records strong consensus, a wide interval records unresolved alternatives, and robust conclusions are those shared across the admissible family
Invariant: the model intentionally leaves at least some probabilistic quantities non-singleton while satisfying its declared consistency or coherence conditions and preserving the set, envelope, or behavioral semantics used for updating and decision
Recognition test: state the sample space and represented uncertainty, identify whether the primitive object is a credal set, lower prevision, capacity, p-box, belief function, or another typed model, verify coherence and conjugacy assumptions, declare conditioning and independence, and keep epistemic imprecision distinct from numerical error
Output or consequence: representing scarce, conflicting, interval, set-valued, or partially elicited knowledge; sensitivity and robust Bayesian analysis; propagating probability bounds; separating agreement across models from model-sensitive conclusions; and making forced precision visible
Failure boundary: lower bounds exceed upper bounds, conjugacy or normalization fails under the chosen theory, endpoints are asserted without warrant, dependence assumptions change silently, interval width is confused with confidence coverage, one distribution is arbitrarily selected and the unresolved set forgotten, or a decision rule is inferred from bounds alone

What It Is Not¶

It is not the whole field of probability and decision theory; many objects in that field do not satisfy its constitutive rule.
It is not its canonical example. A closed convex credal set K contains every probability distribution satisfying declared linear constraints, and event A receives the interval formed by the minimum and maximum of P(A) over K. That is an instance, not a definition.
It is not Probability. Probability is the strict parent supplying normalized uncertainty measures; imprecise probability retains a non-singleton family or lower-upper envelope and the robust-versus-sensitive reasoning that follows from it.
It is not an unrestricted metaphor. some named frameworks use nonadditive set functions directly while others derive them as envelopes of additive measures, so equivalence claims require regularity, convexity, closure, and interpretation hypotheses

Scope of Application¶

Imprecise probability applies when the analyst can specify a measurable possibility space, a nonempty class of admissible probability measures or assessments, and events, gambles, or random quantities evaluated relative to that class and establish that the model intentionally leaves at least some probabilistic quantities non-singleton while satisfying its declared consistency or coherence conditions and preserving the set, envelope, or behavioral semantics used for updating and decision. The entry maps a family of formal models rather than declaring one universally correct interpretation or decision criterion; each application must state its consistency, updating, dependence, and action conventions.^[2]

Recognition. state the sample space and represented uncertainty, identify whether the primitive object is a credal set, lower prevision, capacity, p-box, belief function, or another typed model, verify coherence and conjugacy assumptions, declare conditioning and independence, and keep epistemic imprecision distinct from numerical error
Comparison. Compare legitimate instances through primitive representation, credal-set shape, lower and upper width, coherence, convexity, closure, conditioning, independence, dependence ignorance, elicitation, robustness criterion, and decision rule.
Boundary. some named frameworks use nonadditive set functions directly while others derive them as envelopes of additive measures, so equivalence claims require regularity, convexity, closure, and interpretation hypotheses
Use. Preserve every assumption when using the identity for representing scarce, conflicting, interval, set-valued, or partially elicited knowledge; sensitivity and robust Bayesian analysis; propagating probability bounds; separating agreement across models from model-sensitive conclusions; and making forced precision visible.

Clarity¶

A clear claim names the carrier, governing rule, assumptions, and recognition test. This matters because imprecise can be mistaken for inaccurate, numerically rounded, statistically estimated, or poorly measured, whereas the technical identity is a deliberate set-valued or lower-upper representation of probabilistic commitment. The disciplined statement is that the object counts as Imprecise probability exactly when the model intentionally leaves at least some probabilistic quantities non-singleton while satisfying its declared consistency or coherence conditions and preserving the set, envelope, or behavioral semantics used for updating and decision

Identity and measurement remain separate. Narrow endpoints can be spuriously precise even when the interval is wide; elicitation, finite data, optimization error, and model misspecification each add distinct uncertainty that must not be hidden by the envelope calculation. Approximation or noisy evidence may weaken a classification without changing its definition.

Manages Complexity¶

The abstraction compresses credal sets, lower and upper probabilities, lower previsions, interval probabilities, probability boxes, belief functions, random sets, possibility measures, robust Bayesian classes, and sets of desirable gambles into a stable carrier, rule, invariant, and failure boundary. It makes comparison tractable while retaining the variables that control validity.

Compression can hide assumptions. A responsible use therefore declares primitive representation, credal-set shape, lower and upper width, coherence, convexity, closure, conditioning, independence, dependence ignorance, elicitation, robustness criterion, and decision rule and returns to the full diagnostic whenever a convention or boundary case changes.

Abstract Reasoning¶

Type the carrier. Establish a measurable possibility space, a nonempty class of admissible probability measures or assessments, and events, gambles, or random quantities evaluated relative to that class and reject examples from a different problem.
Lock the rule. Express that the model intentionally leaves at least some probabilistic quantities non-singleton while satisfying its declared consistency or coherence conditions and preserving the set, envelope, or behavioral semantics used for updating and decision independently of one notation or implementation.
Derive carefully. Infer representing scarce, conflicting, interval, set-valued, or partially elicited knowledge; sensitivity and robust Bayesian analysis; propagating probability bounds; separating agreement across models from model-sensitive conclusions; and making forced precision visible only under the stated assumptions.
Stress-test. Contrast the legitimate boundary case—some named frameworks use nonadditive set functions directly while others derive them as envelopes of additive measures, so equivalence claims require regularity, convexity, closure, and interpretation hypotheses—with this counterexample: reporting 0.40 plus or minus 0.03 as a sampling standard error does not by itself define an imprecise probability model, because repeated-sampling uncertainty about an estimator differs from a set of admissible probability laws.

Knowledge Transfer¶

Transfer within probability and decision theory is strong when new cases preserve the same carrier, mechanism, and diagnostic. The move from A closed convex credal set K contains every probability distribution satisfying declared linear constraints, and event A receives the interval formed by the minimum and maximum of P(A) over K. to A probability-box model bounds an uncertain cumulative distribution between lower and upper cumulative functions and propagates those bounds through a system model. demonstrates that continuity.^[3]

Outside the domain, only the skeleton—retain a feasible family rather than collapse uncertainty to one point, then answer queries by the range across members and distinguish invariant conclusions from choice-sensitive ones—travels automatically. The terms lower probability, upper probability, prevision, credal set, envelope, coherence, capacity, belief function, p-box, conditioning, independence, robustness, and dilation retain domain-specific meanings, so every role and inference must be revalidated.

Examples¶

Canonical¶

A closed convex credal set K contains every probability distribution satisfying declared linear constraints, and event A receives the interval formed by the minimum and maximum of P(A) over K. When all members agree, the interval collapses to a precise value for that event even if other events remain imprecise; when they disagree, the width records sensitivity to unresolved probabilistic commitments.^[2] It is canonical because the carrier, rule, invariant, and consequence are all inspectable.^[1]

Mapped back: a measurable possibility space, a nonempty class of admissible probability measures or assessments, and events, gambles, or random quantities evaluated relative to that class → partial commitments define a feasible family of precise laws, and lower and upper envelopes propagate the least and greatest compatible expectation through queries; a narrow interval records strong consensus, a wide interval records unresolved alternatives, and robust conclusions are those shared across the admissible family → the model intentionally leaves at least some probabilistic quantities non-singleton while satisfying its declared consistency or coherence conditions and preserving the set, envelope, or behavioral semantics used for updating and decision → representing scarce, conflicting, interval, set-valued, or partially elicited knowledge; sensitivity and robust Bayesian analysis; propagating probability bounds; separating agreement across models from model-sensitive conclusions; and making forced precision visible

Applied / In Practice¶

A probability-box model bounds an uncertain cumulative distribution between lower and upper cumulative functions and propagates those bounds through a system model. The output enclosure is meaningful only relative to declared dependence and input assumptions; it is not a frequentist confidence interval and may widen substantially when dependence is left unspecified.^[3] It qualifies only after the same diagnostic and failure boundary are checked.^[2]

Mapped back: declared instance → recognition test → boundary check → qualified use

Structural Tensions¶

T1: Exact identity vs. practical recognition. The constitutive condition may be exact while evidence is indirect. Diagnostic: Can the reviewer state both the condition and the warrant?
T2: Canonical form vs. variants. credal sets, lower and upper probabilities, lower previsions, interval probabilities, probability boxes, belief functions, random sets, possibility measures, robust Bayesian classes, and sets of desirable gambles can preserve or change the identity. Diagnostic: Which named role is invariant across the variants?
T3: Compression vs. hidden assumptions. The label is useful only while prerequisites remain visible. Diagnostic: Can each downstream inference be traced to a declared assumption?
T4: Autonomy vs. reduction. The candidate uses broader structures but claims deliberate coherent partial specification of probability through a family or lower-upper envelope, including explicit width and update semantics, rather than ordinary sampling uncertainty, one approximate probability, a fuzzy membership grade, or an unstructured admission of ignorance. Diagnostic: Does that residual still support independent recognition after the parent and neighbors are subtracted?

Structural–Framed Character¶

The entry is structurally mixed but domain-framed. Its portable skeleton is retain a feasible family rather than collapse uncertainty to one point, then answer queries by the range across members and distinguish invariant conclusions from choice-sensitive ones; its identity-bearing terms are lower probability, upper probability, prevision, credal set, envelope, coherence, capacity, belief function, p-box, conditioning, independence, robustness, and dilation. Those terms determine admissible objects, evidence, and consequences inside probability and decision theory.

Structural Core vs. Domain Accent¶

The structural core is a carrier governed by partial commitments define a feasible family of precise laws, and lower and upper envelopes propagate the least and greatest compatible expectation through queries; a narrow interval records strong consensus, a wide interval records unresolved alternatives, and robust conclusions are those shared across the admissible family and tested by state the sample space and represented uncertainty, identify whether the primitive object is a credal set, lower prevision, capacity, p-box, belief function, or another typed model, verify coherence and conjugacy assumptions, declare conditioning and independence, and keep epistemic imprecision distinct from numerical error. The domain accent is constitutive rather than decorative, so an analogy that preserves only the skeleton is not another instance of Imprecise probability.

The proposed strict upward parent is prime:probability. Every admissible member and every lower-upper envelope concern probability assignments or expectations; coherent partial commitment, set-valued representation, and robustness semantics supply the autonomous uncertainty-modeling residual. The edge is proposal-only and points to a frozen prior-baseline Prime.

The entry does not collapse into the parent because deliberate coherent partial specification of probability through a family or lower-upper envelope, including explicit width and update semantics, rather than ordinary sampling uncertainty, one approximate probability, a fuzzy membership grade, or an unstructured admission of ignorance A thematic neighbor is declined whenever it does not literally subsume that rule.

The prospective workspace queue contains one strict upward edge to prime:probability. No live DAG mutation is authorized.

Relationships to Other Abstractions¶

Current abstraction Imprecise probability Domain-specific

Parents (1) — more general patterns this builds on

Imprecise probability is a kind of Probability Prime

The proposed strict upward parent is prime:probability.

Hierarchy paths (2) — routes to 2 parentless roots

Imprecise probability → Probability → Measure → Aggregation → Micro Macro Linkage

Show alternative path (1)

Neighborhood in Abstraction Space¶

Imprecise probability sits in a sparse region of the domain-specific corpus (60^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Imprecise Probability & Multiple Testing (5 abstractions)

Nearest neighbors

Continuous-time stochastic process — 0.87
Algebra of random variables — 0.87
Sub-probability measure — 0.86
Probability axioms — 0.86
Cumulative distribution function — 0.86

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

Not to Be Confused With¶

Credal set. One common representation as a set of probability measures; imprecise probability is the broader family and can take other primitives.
Confidence interval. A repeated-sampling procedure for an unknown parameter, not automatically lower and upper probability for an event.
Fuzzy set. Represents graded membership or vagueness; probability bounds represent partial commitment about uncertainty and need separate semantics.
Robust Bayesian analysis. A major use in which priors or likelihoods vary over a class, but not the entirety of imprecise probability.
Dempster–Shafer belief function. A related structured lower-probability model with random-set semantics, constituting one typed part of the wider family.

References¶

[1] Peter Walley, Statistical Reasoning with Imprecise Probabilities, Chapman and Hall, 1991, ISBN 978-0-412-28660-5. registry ↩a ↩b ↩c

[2] Thomas Augustin, Frank P. A. Coolen, Gert de Cooman, and Matthias C. M. Troffaes, eds., Introduction to Imprecise Probabilities, Wiley, 2014, DOI 10.1002/9781118763117. registry ↩a ↩b ↩c

[3] Scott Ferson et al., Constructing Probability Boxes and Dempster–Shafer Structures, Sandia National Laboratories report SAND2002-4015, 2003, DOI 10.2172/809606. registry ↩a ↩b