1 Foundations
Decision theory provides a structured way to analyze choice. It represents alternatives, possible states of the world, and the consequences that follow from each action-state combination. By combining preferences with information about uncertainty, the framework helps compare options in a disciplined manner. It is used both as a normative standard for rational choice and as a basis for models of real decision-making.
1.1 Basic concepts
At its core, decision theory begins with a set of available actions, the possible conditions in which those actions may be taken, and the outcomes produced by each combination. These elements form the decision problem. The theory then asks how a choice should be evaluated, either by direct comparison of consequences or through a more formal rule.
1.1.1 Decision-makers and actions
A decision-maker is the agent whose choices are being analyzed. This may be an individual, a firm, a public institution, or an algorithm. Actions are the alternatives available to the agent, such as selecting a plan, placing a bet, or adopting a policy. A decision problem is often represented by listing the available actions and the circumstances under which each action is performed.
1.1.2 States of the world
States of the world are the relevant conditions that the decision-maker does not fully control. They may describe natural events, market conditions, or any other factors that influence outcomes. A key feature of the framework is that the decision-maker usually does not know in advance which state will obtain. Actions are therefore assessed relative to multiple possible states.
1.1.3 Outcomes and consequences
Outcomes are the results associated with each action-state pair. Consequences may be monetary, physical, social, or psychological, depending on the problem. Decision theory often treats outcomes as objects of preference, allowing them to be ranked or assigned utility values. In many applications, the same action can lead to very different consequences depending on the state of the world.
1.2 Preferences and utility
Preferences describe how a decision-maker ranks alternatives or outcomes. Utility provides a numerical representation of these preferences, making them easier to compare and analyze. The idea is not that utility must be directly measurable in a physical sense, but that it can serve as a consistent scale for choice.
1.2.1 Preference ordering
A preference ordering arranges outcomes from more preferred to less preferred. It may be complete, meaning any two options can be compared, or partial, meaning some pairs remain incomparable. In formal models, preference orderings often serve as the starting point for deriving utility functions and choice rules. They also help distinguish between stable tastes and temporary judgments.
1.2.2 Utility functions
A utility function assigns a real number to each outcome or lottery, preserving the decision-maker’s ordering of preferences. Higher utility corresponds to more preferred outcomes. Utility functions are especially useful because they permit algebraic analysis of choice under uncertainty. Different utility representations can encode the same ordering while differing in scale or transformation.
1.2.3 Expected utility
Expected utility is the average utility of outcomes weighted by their probabilities. It is one of the most influential concepts in decision theory. The idea is that a rational agent should choose the option with the highest expected utility when probabilities are known or can be estimated. This principle forms the basis of many normative and applied models.
1.3 Information and uncertainty
Decision-making often occurs when the outcome is not certain. The quality and amount of information available to the decision-maker affect how choices are evaluated. Decision theory distinguishes between predictable variation, incomplete knowledge, and structured uncertainty.
1.3.1 Risk
Risk refers to situations in which the possible outcomes are known and probabilities can be assigned to them. Insurance, gambling, and many financial choices are classic examples. Under risk, decision theory can compare actions by combining likelihoods with utilities. This setting is especially amenable to quantitative analysis.
1.3.2 Uncertainty
Uncertainty arises when probabilities are not fully known or cannot be assigned with confidence. In such cases, the decision-maker may know the set of possible states but lack reliable frequency information. Decision theory then relies on alternative criteria, subjective beliefs, or robust rules. Uncertainty is a central topic because many real decisions do not fit tidy probabilistic assumptions.
1.3.3 Information sets
An information set is the collection of facts available to the decision-maker at the time of choice. It shapes both the feasible actions and the expected consequences. Better information may reduce uncertainty, though it can also increase complexity. In formal analysis, information sets are used to describe what the agent knows before acting and how that knowledge changes with new evidence.
2 Normative decision theory
Normative decision theory specifies how choices ought to be made if the decision-maker is to act rationally. It provides standards for consistency, coherence, and optimality. Many of its principles are expressed as axioms from which choice rules can be derived.
2.1 Rational choice
Rational choice theory treats preferences and decisions as governed by principles that avoid contradiction and arbitrariness. It does not claim that people always follow these principles, but it offers a benchmark for evaluating choices. The resulting framework is central to economics and formal philosophy.
2.1.1 Completeness and transitivity
Completeness means that any two alternatives can be compared, while transitivity means that if A is preferred to B and B to C, then A should be preferred to C. Together, these conditions ensure that preferences can be ordered consistently. They are foundational assumptions in many models of rational choice. Without them, choice rankings may become unstable or cyclical.
2.1.2 Dominance
Dominance holds that if one option is at least as good as another in every relevant respect and better in at least one respect, it should be chosen. This principle is often regarded as a minimal requirement of rationality. It helps eliminate options that are plainly inferior. Dominance is especially important in decision trees and strategic comparisons.
2.1.3 Independence
The independence principle states that preference between two options should remain unchanged when they are mixed with the same third option. It is a key assumption in many expected utility models. Independence supports the idea that choices depend on relative differences rather than irrelevant background alternatives. It has also been a focal point in discussions of preference anomalies.
2.2 Expected utility theory
Expected utility theory formalizes rational choice under risk by assigning utilities to outcomes and weighting them by probabilities. It became a major framework for analyzing uncertainty in economics, statistics, and philosophy. The theory offers both predictive power and normative guidance.
2.2.1 von Neumann–Morgenstern utility
The von Neumann–Morgenstern utility theorem shows how preferences over lotteries can be represented by a utility function that is linear in probabilities. This utility is unique up to positive affine transformation. The result is a cornerstone of modern decision theory. It demonstrates that coherent preferences over risky options can be summarized numerically.
2.2.2 Subjective expected utility
Subjective expected utility extends the framework by allowing probabilities to reflect the decision-maker’s beliefs rather than only objective frequencies. It combines personal credences with utility to evaluate options. This approach is especially useful when data are incomplete or when judgments must be made about unique events. It links rational choice to Bayesian reasoning.
2.2.3 Linearity in probabilities
Linearity in probabilities means that the value of a lottery is the weighted sum of the utilities of its outcomes. This feature makes expected utility mathematically tractable and conceptually clear. It also implies that changing probabilities has a proportional effect on evaluation. Many alternative models relax this assumption to better capture observed behavior.
2.3 Axiomatic approaches
Axiomatic decision theory begins with general principles and derives representation theorems from them. This method clarifies which assumptions are doing the theoretical work. It also helps reveal the limits of each model by showing exactly where it applies.
2.3.1 Utility axioms
Utility axioms specify the conditions under which preferences can be represented numerically. Common axioms include completeness, transitivity, continuity, and independence. These requirements allow the construction of a utility scale that respects observed rankings. Different axiomatic systems yield different forms of utility representation.
2.3.2 Probability axioms
Probability axioms define how uncertainty should be quantified. The standard axioms require nonnegativity, normalization, and additivity for disjoint events. In decision theory, these rules support coherent belief formation and consistent updating. They are often treated as the mathematical basis for rational uncertainty.
2.3.3 Representation theorems
Representation theorems prove that if certain axioms hold, then preferences or beliefs can be represented in a specific formal way. In decision theory, they connect intuitive principles to utility and probability models. Such theorems give the subject much of its rigor. They also explain why apparently abstract axioms have practical significance.
3 Decision making under uncertainty
When probabilities are incomplete or unavailable, decision-making requires special treatment. Decision theory offers ways to combine partial beliefs, cautious criteria, and information updates. The result is a set of tools for choices made without full statistical certainty.
3.1 Utility and probability in uncertainty
Uncertainty-based models often rely on beliefs that are not objectively known. These beliefs may be subjective, Bayesian, or expressed as degrees of confidence. The challenge is to use them in a way that still yields coherent action.
3.1.1 Bayesian beliefs
Bayesian beliefs represent uncertainty as degrees of belief that can be updated when new evidence appears. They treat probability as a rational measure of confidence rather than only a long-run frequency. In decision theory, this view allows choices to change systematically as information accumulates. It is widely used in statistics and artificial intelligence.
3.1.2 Prior and posterior probabilities
Prior probabilities represent initial beliefs before new evidence is considered. Posterior probabilities are updated beliefs after observing information. Bayesian decision analysis combines the two through Bayes’ rule. This process gives a formal account of learning under uncertainty.
3.1.3 Credences
Credences are graded beliefs that reflect how strongly a decision-maker accepts a proposition. They are often treated as the subjective counterpart of probability. In decision contexts, credences guide actions when facts are not fully known. They provide a bridge between evidence and preference.
3.2 Decision criteria
Different decision criteria can be used when uncertainty is severe. These rules vary in how optimistic, pessimistic, or cautious they are. They are often applied when probabilities are missing or unreliable.
3.2.1 Maximax rule
The maximax rule selects the option with the best possible outcome. It reflects extreme optimism by focusing only on the most favorable state. This criterion ignores downside risk and is therefore uncommon in serious planning, but it is useful as a benchmark. It highlights the role of aspiration in choice.
3.2.2 Maximin rule
The maximin rule chooses the action whose worst outcome is least bad. It is a cautious, security-oriented approach. By focusing on the minimum payoff, it protects against severe loss. The rule is associated with pessimistic or highly conservative decision attitudes.
3.2.3 Minimax regret
Minimax regret selects the action that minimizes the maximum possible regret. Regret is defined as the loss from not having chosen the best action after the state is known. This criterion shifts attention from outcomes themselves to comparison with hindsight. It is often used when the decision-maker wants to avoid large disappointment.
3.3 Ambiguity and ignorance
Ambiguity refers to uncertainty about probabilities themselves, not just about outcomes. Ignorance is a broader lack of knowledge about what may happen or how likely it is. Decision theory has developed specialized models to address these situations.
3.3.1 Ellsberg paradox
The Ellsberg paradox shows that many people prefer known probabilities over unknown ones, even when expected value would suggest indifference. It reveals that ambiguity affects choice in ways not captured by classical probability models. The paradox became highly influential in challenging standard assumptions. It helped motivate richer theories of uncertainty.
3.3.2 Ambiguity aversion
Ambiguity aversion is the tendency to avoid options with poorly defined probabilities. It differs from ordinary risk aversion, which concerns known probabilistic variation. Ambiguity-averse decision-makers often prefer more informative or better understood alternatives. This pattern has important implications for finance, insurance, and policy design.
3.3.3 Robust decision rules
Robust decision rules aim to perform reasonably well across a wide range of possible assumptions. They are designed for settings where exact probabilities are uncertain or disputed. Instead of relying on a single estimate, they emphasize stability and resilience. Such rules are common in engineering, economics, and model-based planning.
4 Decision making under risk
Risk-based decision theory deals with outcomes whose probabilities are known or can be estimated. This setting allows precise evaluation of choices through expected utility, stochastic dominance, and related tools. It is central to insurance, gambling, and financial analysis.
4.1 Probabilistic outcomes
Probabilistic models describe situations in which an action leads to a distribution of possible results. The structure of the distribution matters, not only the average outcome. Decision theory uses this information to compare alternatives with different chance profiles.
4.1.1 Lotteries
A lottery is a probabilistic mixture of outcomes. In decision theory, lotteries serve as the basic objects of choice under risk. They provide a clear way to represent uncertainty and compare options mathematically. Many foundational models treat complex decisions as preferences over lotteries.
4.1.2 Risk attitudes
Risk attitudes describe how a decision-maker responds to variability in outcomes. A person may be risk-averse, risk-neutral, or risk-seeking depending on whether they prefer certainty, indifference to spread, or exposure to variability. These attitudes are often inferred from choice behavior. They are typically represented by the curvature of a utility function.
4.1.3 Risk aversion
Risk aversion is the preference for a sure outcome over a gamble with the same expected value, when the gamble introduces variability. It is one of the most studied concepts in economics and finance. Risk-averse behavior helps explain insurance purchases and cautious investment strategies. It also shapes how people trade expected gain for security.
4.2 Stochastic dominance
Stochastic dominance compares probability distributions without requiring a full utility specification. It gives a partial ordering of risky options. The method is especially useful when one wants broad results that hold across many plausible preferences.
4.2.1 First-order dominance
First-order stochastic dominance occurs when one distribution yields at least as good an outcome as another in every state and better outcomes in some states. Any decision-maker who prefers more to less should choose the dominating option. This criterion is strong and widely applicable. It provides a simple way to eliminate clearly inferior distributions.
4.2.2 Second-order dominance
Second-order stochastic dominance extends the idea to decision-makers who prefer more expected value and dislike risk. It takes account of both mean and dispersion. An option that second-order dominates another is preferable for all risk-averse agents satisfying standard assumptions. This makes the concept useful in finance and welfare analysis.
4.2.3 Comparative risk analysis
Comparative risk analysis studies how different risky options relate to one another in terms of variance, tail behavior, and distribution shape. It is often used to compare investment portfolios or insurance contracts. The analysis may combine dominance tests with utility-based methods. Its aim is to clarify not just expected return, but the pattern of exposure.
4.3 Insurance and gambling
Insurance and gambling are classic examples of decisions under risk. They involve trade-offs between expected value, uncertainty, and willingness to pay for protection or excitement. These cases illustrate how utility and probability interact in everyday and commercial settings.
4.3.1 Fair bets
A fair bet is a gamble with zero expected gain or loss in monetary terms. Rational acceptance of such a bet depends on the decision-maker’s utility, not only on expected value. Risk-averse individuals may reject fair bets because of the disutility of uncertainty. Fairness in expectation therefore does not guarantee attractiveness.
4.3.2 Expected value
Expected value is the probability-weighted average of possible monetary outcomes. It is a basic summary statistic for risky choices. Although useful, it does not by itself determine preference when risk attitudes matter. Decision theory distinguishes expected value from expected utility to capture this difference.
4.3.3 Risk premium
A risk premium is the amount a person would pay to avoid a risky prospect with a given expected value. It measures the subjective cost of uncertainty. Higher risk aversion generally implies a larger premium. This concept is important in insurance pricing, portfolio selection, and the valuation of uncertain gains.
5 Intertemporal and sequential decision theory
Many decisions unfold over time, with later choices depending on earlier ones and new information. Intertemporal decision theory studies how agents evaluate present and future consequences. Sequential decision theory extends this to branching paths, plans, and strategic interactions.
5.1 Dynamic choice
Dynamic choice concerns decisions made across multiple time periods. The agent must consider not only immediate outcomes but also future effects and changing preferences. This makes timing a central part of the analysis.
5.1.1 Intertemporal utility
Intertemporal utility measures how a decision-maker values outcomes occurring at different times. It is used to compare present enjoyment with future benefits or costs. The model may aggregate utility across periods in a structured way. Such formulations are essential in saving, investment, and planning.
5.1.2 Discounting
Discounting reduces the weight assigned to future outcomes relative to present ones. It reflects impatience, opportunity cost, or uncertainty about future realization. Standard models often use exponential discounting, though other forms exist. Discounting is central to economics, environmental analysis, and health planning.
5.1.3 Time inconsistency
Time inconsistency occurs when preferences change in a way that makes earlier plans unattractive later on. A choice judged optimal today may be reversed tomorrow despite no change in the environment. This phenomenon helps explain procrastination, self-control problems, and shifting plans. It has inspired many models of dynamic inconsistency and commitment.
5.2 Decision trees
Decision trees represent sequential choices, chance events, and outcomes in a branching diagram. They make the structure of a decision problem transparent. By laying out alternatives step by step, they support systematic analysis.
5.2.1 Backward induction
Backward induction solves a decision tree by reasoning from the final stage to the initial one. At each node, the decision-maker chooses the branch that maximizes expected utility given future consequences. This method is widely used in game theory and planning. It depends on the assumption that future choices can be anticipated rationally.
5.2.2 Sequential planning
Sequential planning organizes decisions into a series of connected steps. Each step may depend on earlier results and newly observed information. This approach is useful when a single one-shot decision is not sufficient. It allows agents to revise plans as the situation evolves.
5.2.3 Dynamic programming
Dynamic programming is a method for solving complex sequential problems by breaking them into smaller subproblems. It exploits the principle that an optimal plan contains optimal subplans. This technique is widely used in optimization, economics, and computer science. It is especially effective when many stages and states are involved.
5.3 Games and strategic interaction
Strategic interaction occurs when each decision-maker’s outcome depends on the actions of others. Decision theory overlaps with game theory in these settings. Choices must then account not only for uncertainty but also for the expected behavior of other agents.
5.3.1 Game-theoretic decision-making
Game-theoretic decision-making analyzes choices made in the presence of strategic opponents, partners, or competitors. Each agent forms expectations about others’ actions and responds accordingly. This framework is useful in bargaining, auctions, coordination, and competition. It extends decision theory beyond isolated choice.
5.3.2 Mixed strategies
A mixed strategy is a probabilistic rule for choosing among actions. It allows a player to randomize rather than selecting a fixed move every time. Mixed strategies can improve predictability, conceal intentions, or achieve equilibrium conditions. They are standard in formal strategic models.
5.3.3 Equilibrium concepts
Equilibrium concepts describe stable outcomes in which no participant has an incentive to deviate given the others’ choices. They provide a way to analyze strategic consistency. Different equilibria may apply depending on information, timing, and the structure of interaction. These concepts are central to game-theoretic decision theory.
6 Descriptive and behavioral decision theory
Descriptive decision theory studies how people actually make choices, rather than how they ideally should. It draws on psychology, behavioral economics, and experimental methods. The field has revealed many systematic departures from classical rational models.
6.1 Human judgment
Human judgment often relies on simplified mental processes. These shortcuts can be efficient, but they may also produce predictable errors. Descriptive models aim to explain such patterns.
6.1.1 Heuristics and biases
Heuristics are mental shortcuts used to make judgments quickly. Biases are systematic errors that can result from these shortcuts. Research on heuristics and biases has shown that people often deviate from normatively optimal reasoning. The topic is influential in psychology and decision research.
6.1.2 Framing effects
Framing effects occur when different descriptions of the same choice lead to different decisions. The content of the problem may be unchanged, yet the presentation alters judgment. This finding shows that preferences can depend on context as well as on outcome structure. Framing is important in consumer behavior, communication, and policy design.
6.1.3 Prospect theory
Prospect theory is a behavioral model that describes how people evaluate gains and losses relative to a reference point. It captures features such as loss aversion, diminishing sensitivity, and nonlinear probability weighting. The theory has become one of the most influential alternatives to expected utility. It is widely used to explain observed deviations from classical choice.
6.2 Bounded rationality
Bounded rationality recognizes that real decision-makers have limited time, knowledge, and computational capacity. Instead of maximizing perfectly, they often seek workable solutions. This perspective emphasizes adaptation to human constraints.
6.2.1 Satisficing
Satisficing means selecting an option that is good enough rather than searching for the best possible one. It reflects practical limits on time and information. The concept helps explain why people stop searching once an acceptable alternative is found. It is associated with Herbert A. Simon’s work on bounded rationality.
6.2.2 Limited information processing
Limited information processing refers to the restricted ability to absorb, store, and evaluate data. Decision-makers may ignore details, compress information, or rely on summaries. This constraint affects both everyday and technical decisions. Models that incorporate processing limits often produce more realistic predictions.
6.2.3 Cognitive constraints
Cognitive constraints include attention limits, memory restrictions, and difficulties in computation. They shape what options are noticed and how choices are compared. Such constraints can lead to simplification, delegation, or rule-based behavior. They are important in understanding why idealized rationality is hard to achieve.
6.3 Empirical models
Empirical models test decision theory against observed behavior. They are built from experiments, surveys, field data, and laboratory tasks. Their goal is to measure how choice actually occurs in practice.
6.3.1 Choice experiments
Choice experiments present respondents with structured alternatives and record their selections. The design can isolate specific attributes and estimate preference patterns. These experiments are widely used in economics, marketing, and public policy. They provide controlled evidence about trade-offs.
6.3.2 Behavioral data
Behavioral data are observations of real or simulated decisions. They may include purchasing records, reaction times, experimental responses, or choice histories. Such data help identify regularities and departures from theoretical predictions. They are crucial for testing descriptive models.
6.3.3 Decision laboratories
Decision laboratories are controlled settings in which participants make choices under carefully designed conditions. Researchers use them to examine risk preferences, learning, cooperation, and judgment. Laboratory methods allow precise manipulation of variables. They have played a major role in experimental decision theory.
7 Applications
Decision theory has broad practical use across the formal sciences and applied fields. Its models support analysis, prediction, and optimization in settings involving uncertainty, strategic behavior, and constrained resources. Many applications translate abstract principles into operational tools.
7.1 Economics and finance
Economics and finance rely heavily on decision theory to model consumption, saving, investment, and market behavior. The framework helps explain how agents allocate resources under uncertainty. It also supports the analysis of risk and return.
7.1.1 Portfolio choice
Portfolio choice concerns the allocation of wealth across assets with different return patterns. Decision theory evaluates portfolios by balancing expected gains against risk. Preferences, diversification, and horizon all matter. The problem is a classic application of utility under uncertainty.
7.1.2 Asset pricing
Asset pricing studies how uncertain future payoffs are valued in present terms. Decision theory contributes by linking prices to beliefs, risk attitudes, and discounting. It helps explain why investors demand compensation for uncertainty. The framework is central to modern financial economics.
7.1.3 Consumer choice
Consumer choice models how individuals select goods and services subject to preferences and budget limits. Decision theory provides the logic for comparing bundles and evaluating trade-offs. It is used to study demand, substitution, and welfare. The approach underlies much of microeconomic analysis.
7.2 Statistics and machine learning
Statistics and machine learning use decision theory to choose estimators, predictors, and classification rules. The focus is often on minimizing loss rather than maximizing utility in a broad sense. This makes the connection between inference and action explicit.
7.2.1 Bayesian decision theory
Bayesian decision theory combines prior beliefs, evidence, and loss functions to select the best action under uncertainty. It treats inference as a decision problem. The method is highly influential in statistical estimation and prediction. It also connects naturally with subjective probability.
7.2.2 Classification rules
Classification rules assign an observation to one of several categories. Decision theory evaluates these rules by their error rates and associated costs. Different mistakes may carry different penalties, making a purely accuracy-based rule insufficient. This perspective is central to pattern recognition and predictive modeling.
7.2.3 Loss functions
Loss functions quantify the cost of a wrong or suboptimal decision. They are used to compare estimators, forecasts, and classifiers. By formalizing the consequences of error, loss functions make decision criteria precise. They are a core tool in modern statistics and machine learning.
7.3 Artificial intelligence and operations research
Artificial intelligence and operations research apply decision theory to automated reasoning, resource allocation, and planning. These fields often combine probabilistic models with optimization methods. The result is a practical framework for machine-supported choice.
7.3.1 Decision support systems
Decision support systems assist users in evaluating options and selecting actions. They integrate data, models, and preferences into a usable interface. Such systems are common in business, medicine, and logistics. Their aim is to improve judgment without replacing human oversight.
7.3.2 Optimization under uncertainty
Optimization under uncertainty seeks the best action when outcomes are not fully predictable. It may use stochastic models, robust methods, or scenario analysis. The goal is to produce plans that remain effective across plausible conditions. This area is important for scheduling, routing, and resource management.
7.3.3 Automated planning
Automated planning involves generating action sequences for machines or software agents. Decision theory contributes by specifying objectives, costs, and uncertainty handling. Planning systems are used in robotics, logistics, and intelligent agents. They often combine search, inference, and dynamic optimization.