Decision Theory

framework · mathematics · formal-scientific

Normative analysis of choice under uncertainty using utility and probability.

Decision Theory is the mathematical framework for analyzing rational choice under uncertainty, integrating probability theory (representing beliefs about possible states of the world) and utility theory (representing preferences over outcomes) to identify choices that maximize expected utility. The framework's foundational result is the von Neumann-Morgenstern utility theorem (1944): any agent whose preferences satisfy four axioms (completeness, transitivity, continuity, independence) can be represented as maximizing the expected value of some utility function. Subsequent foundational work includes Leonard J. Savage's Foundations of Statistics (1954, the personalist or subjectivist Bayesian foundation grounding probabilities in choice behavior), Howard Raiffa's Decision Analysis (1968, applied decision-theoretic methodology), and the substantial body of decision theory under various conditions (decision under risk, decision under uncertainty, decision under ignorance, decision under conflict). Decision theory is foundational to microeconomics, finance (portfolio theory, real options), operations research, statistics (Bayesian decision theory combines posterior with loss function), AI (planning, reinforcement learning), and applied decision analysis in medicine, engineering, and policy. Behavioral economics (Kahneman and Tversky's Prospect Theory, 1979 — separately enriched in psychology) substantially documented systematic deviations from expected utility maximization that descriptive decision theory must address; this complicates without replacing the normative framework.

Originators

Daniel Bernoulli (St. Petersburg paradox, expected utility precursor 1738); John von Neumann and Oskar Morgenstern (1944 utility theorem); Leonard J. Savage (1954 Foundations of Statistics); Howard Raiffa (applied decision analysis); Frank P. Ramsey (foundational antecedent, 1926) high

Year / Decade

1738 (Bernoulli precursor); 1926 (Ramsey antecedent); 1944 (von Neumann-Morgenstern); 1954 (Savage); ongoing development high

Primary sources

Bernoulli, D. (1738). 'Specimen Theoriae Novae de Mensura Sortis', Ramsey, F.P. (1926). 'Truth and Probability', von Neumann, J. & Morgenstern, O. (1944). Theory of Games and Economic Behavior, Savage, L.J. (1954). The Foundations of Statistics, Raiffa, H. (1968). Decision Analysis high

Core components

Expected utility maximization as decision rule
Von Neumann-Morgenstern axioms (completeness, transitivity, continuity, independence)
Probability of states of the world
Utility function over outcomes
Decision under risk vs uncertainty (Knightian distinction)
Bayesian decision theory (posterior + loss function)
Subjective expected utility (Savage)
Application across economics, finance, operations research, statistics
Distinction from descriptive decision theory (Prospect Theory and behavioral alternatives)

Primary use case

Foundation of microeconomic theory of choice; finance (portfolio theory, options pricing, risk management); statistics (Bayesian decision theory); operations research (decision analysis); medical decision-making (cost-effectiveness analysis, clinical decision analysis); AI planning and reinforcement learning; foundation for substantial applied decision analysis across engineering, policy, and management.

Common criticisms

Empirical research substantially documents that humans systematically violate expected utility axioms (Allais paradox, Ellsberg paradox, framing effects, loss aversion) — Prospect Theory (Kahneman-Tversky 1979) and subsequent behavioral economics provide descriptive alternatives that the normative framework doesn't fully accommodate
specification of utility functions and probability distributions is contested in real applications
ambiguity (Knightian uncertainty, where probabilities themselves are unknown) raises challenges that classical decision theory addresses incompletely
multi-attribute decision-making with incommensurable values (cost vs lives saved, present vs future generations) strains the framework
commercial decision-analysis industry has produced compliance-style applications with varying analytical fidelity
integration with decision-making in groups and organizations involves substantial complexity beyond individual decision theory
foundational debates between subjectivist and objectivist probability, and between expected utility and alternative decision rules (maximin, regret minimization), remain unresolved.

Lineage

Parent of: Markov Decision Processes
Siblings: Game Theory, Bayesian Inference