Introduction to the Theory of Imprecise Probability

Erik QuaeghebeurSpringerBriefs 2021 · DOI · OpenAlex

Abstract

The theory of imprecise probability is a generalization of classical ‘precise’ probability theory that allows modeling imprecision and indecision. This is a practical advantage in situations where a unique precise uncertainty model cannot be justified. This arises, for example, when there is a relatively small amount of data available to learn the uncertainty model or when the model’s structure cannot be defined uniquely. The tools the theory provides make it possible to draw conclusions and make decisions that correctly reflect the limited information or knowledge available for the uncertainty modeling task. This extra expressivity however often implies a higher computational burden. This chapter presents the essential concepts and techniques from the theory, as well as an overview of specific uncertainty models used.

Introduction: Beyond Precise Probabilities

Classical probability theory requires the assignment of precise numerical values to uncertainty, even when data is insufficient to justify such precision. In many real-world scenarios, particularly those involving "small data" or high ambiguity, choosing a single probability (or a single prior in Bayesian analysis) can be arbitrary, leading to overconfident and potentially unreliable decisions.

Imprecise Probability (IP) addresses these limitations by providing a mathematical framework to model imprecision (lack of knowledge about the exact probability) and indecision (insufficient evidence to choose between options). By allowing for "sets" of probabilities, IP enables models to honestly reflect ignorance and sparse data.

Theoretical Framework

The theory is built upon three primary mathematical representations:

  1. Sets of Acceptable Gambles: This is the most fundamental layer. An agent expresses uncertainty by defining which "gambles" (real-valued functions over outcomes) they are willing to accept.
  2. Lower and Upper Previsions: These generalize the concept of expectation. Instead of a single expected value, IP utilizes a lower prevision (the maximum price an agent would pay to buy a gamble) and an upper prevision (the minimum price they would accept to sell it).
  3. Credal Sets: A credal set is a closed, convex set of multiple "precise" probability distributions. A larger set indicates greater vagueness in the data; as more information is acquired, the set shrinks toward a precise model.

The core mechanism for reasoning within this framework is Natural Extension. This process takes initial assessments and logically expands them into a full, coherent uncertainty model while ensuring the avoidance of "sure loss"—situations where an agent is guaranteed to lose money regardless of the outcome.

The Pedagogical Example: Belgium vs. Japan

To illustrate these concepts, the author employs a running example involving two characters, Wiske and Yoko Tsuno, betting on a sports match between Belgium (BE) and Japan (JP). This example demonstrates how conflicting individual assessments are combined and how mathematical "absurdity" or "sure loss" is detected when pooling assessments.

Table 1: Lower and upper probability assessments for the Belgium vs. Japan match example.

As shown in Table 1, different agents provide different bounds. For instance, Wiske's refusal to bet against Belgium results in a lower probability $\underline{P}(\text{BE})$ of 5/6 and an upper probability $\overline{P}(\text{BE})$ of 1. Conversely, the "Empty pool" represents the vacuous model, where the state of total ignorance is reflected by bounds of 0 and 1. The "Simple pool" illustrates a case of sure loss, where the combined assessments lead to logically impossible bounds.

Key Insights and Results

  • Indecision as a Feature: Unlike classical theory, which mandates comparability between all options, IP allows for "incomparability" when evidence is insufficient. This is particularly valuable in safety-critical engineering.
  • The Vacuous Model: Total ignorance is represented by the interval [0, 1]. This is a more honest starting point than a uniform distribution, which makes an unjustified claim of symmetry.
  • Coherence and Rationality: A model is considered rational only if it is "coherent," meaning the set of acceptable gambles does not allow for a "Dutch book" (guaranteed loss).
  • Lower Envelope Theorem: This theorem bridges previsions and credal sets, proving that the lower prevision of a gamble is the minimum expectation taken over all distributions within the corresponding credal set.

Limitations and Future Directions

The primary hurdle for the widespread adoption of IP is computational complexity. While IP is more expressive than classical probability, calculating lower and upper bounds often requires solving linear programming problems or complex optimizations. This is significantly more demanding than the simple arithmetic of precise probability.

Future research aims to improve the efficiency of these calculations for continuous spaces and high-dimensional problems, utilizing tools such as P-boxes (probability boxes) or imprecise Dirichlet models to make the framework more accessible for large-scale engineering applications.

Summary updated

Abstract

Abstract The theory of imprecise probability is a generalization of classical ‘precise’ probability theory that allows modeling imprecision and indecision. This is a practical advantage in situations where a unique precise uncertainty model cannot be justified. This arises, for example, when there is a relatively small amount of data available to learn the uncertainty model or when the model’s structure cannot be defined uniquely. The tools the theory provides make it possible to draw conclusions and make decisions that correctly reflect the limited information or knowledge available for the uncertainty modeling task. This extra expressivity however often implies a higher computational burden. The goal of this chapter is to primarily give you the necessary knowledge to be able to read literature that makes use of the theory of imprecise probability. A secondary goal is to provide the insight needed to use imprecise probabilities in your own research. To achieve the goals, we present the essential concepts and techniques from the theory, as well as give a less in-depth overview of the various specific uncertainty models used. Throughout, examples are used to make things concrete. We build on the assumed basic knowledge of classical probability theory.
BibTeX
@inproceedings{quaeghebeur2021introduction,
  author = {Erik Quaeghebeur},
  title = {{Introduction to the Theory of Imprecise Probability}},
  booktitle = {SpringerBriefs},
  year = {2021},
  doi = {10.1007/978-3-030-83640-5\_3},
}
Figure