Probability is a fundamental concept that transcends disciplinary boundaries, including mathematics, statistics, physics, philosophy, and finance. Over time, different disciplines have developed distinct interpretations, each with specific strengths and limitations.
The table below offers a unified view of probability, presenting the main interpretations, associated formulas, contexts where they are appropriate, as well as the advantages and limitations of each. The aim is not merely to list the concepts, but also to highlight how some frameworks integrate or generalize others, thereby enabling a coherent perspective that is applicable across domains.
This table is optimized for viewing on a MacBook and is therefore wider than a phone screen. Swipe right to view all columns.
| Sphere | Interpretation | Short definition | Typical formula/object | When appropriate | Strength | Limitation |
|---|---|---|---|---|---|---|
| Mathematics | Axiomatic (Kolmogorov) | Probability is a measure on a sigma-algebra satisfying Kolmogorov’s axioms | \(P:\mathcal{F}\to[0,1]\), \(P(\Omega)=1\), countable additivity | Theoretical foundations, general modeling | Maximum rigor, unifying framework | Doesn’t say what probability is in the world |
| Statistics | Frequentist | Limit of the relative frequency in a hypothetical sequence of identical repetitions | \(P(A)=\lim_{n\to\infty} N_A(n)/n\) (if it exists) | Repeatable processes, quality control, asymptotics; entails rejection or failure to reject the null hypothesis based on observed data. | Directly tied to repeated data; powerful classical tools | Challenging for one-off events; limits may fail to exist |
| Statistics/Decision | Bayesian (subjective) | An agent’s coherent degree of belief, updated via Bayes’ rule | \(P(\theta\mid D)\propto P(D\mid \theta)P(\theta)\) | Sequential learning, decisions under uncertainty, one-off events; combines prior knowledge with observed data to produce the posterior distribution | Integrates prior knowledge; decision-theoretic coherence | Depends on the prior; unavoidable subjectivity |
| Engineering/ML Actuarial | Maximum Entropy (Jaynes) | Assigns distributions by maximizing entropy subject to known constraints | Maximizing \(H(P)\) under moment/constraint information | Partial information (aggregates, moments) and neutrality | The least-committed rule, transparent and principled; recovers many of the above frameworks without extra assumptions; alleviates many paradoxes | Assignment rule, not an ontological definition |
| Philosophy of Science | Propensity (Popper) | Objective physical tendency of a setup to produce outcomes | Propensity of the device (e.g., slightly unbalanced coin) | Causal random processes in physics/biology | Links probability to physical mechanisms | Hard to measure/define operationally |
| Philosophy | Logical (Carnap) | Logical degree of confirmation given a language and background knowledge | \(P\) as a logical relation between propositions in a formal language | Formal deductive–inductive inference | An attempt at epistemic objectivity | Sensitive to language choice; little practical use |
| Decision/Betting | Coherence (Dutch book, de Finetti) | Probabilities are betting prices that avoid arbitrage | Coherence implies the axioms; with exchangeability ⇒ de Finetti’s theorem | Modeling beliefs and bets, decision-making | Clear connection to decision and risk | Normative; agent-dependent, not physical |
| Physics | Quantum (Born) | Probability is the squared modulus of the wave amplitude | \(P(a)=\lVert P_a\,\psi\rVert^2\) or \(\mathrm{Tr}(\rho P_a)\) | Quantum mechanics, measurement | Experimentally confirmed; necessary for quantum phenomena | Non-classical (noncommutative); subtle interpretation |
| Information Theory | Coding/Shannon | Links surprise and optimal code length with probability | Ideal length \(\approx -\log_2 p(x)\) | Compression, communications, source modeling | Operational interpretation of probability | Requires a given/estimated distribution |
| Finance | Risk-neutral (martingale) | A pricing probability under which discounted prices are martingales | Measure \(Q\) under which \(S_t/B_t\) is a martingale | Derivative pricing, absence of arbitrage | Powerful for pricing and hedging | Not a physical probability; may be non-unique |
| Generalized Uncertainty | Dempster–Shafer | Degrees of belief and plausibility, non-additive | Belief and plausibility functions (non-additive) | Evidence fusion, sensors, epistemic uncertainty | Expresses ignorance separately from risk | Harder to integrate with classical statistics |