Utility
Utility is a numerical representation of a person's preferences among alternatives. In economics and decision theory, a utility function assigns numbers so that alternatives receiving higher values are preferred to alternatives receiving lower values. The numbers represent the ordering of choices rather than a directly observable substance, and their interpretation depends on the assumptions imposed on the underlying preference relation.
The concept also appears in moral philosophy, particularly in utilitarianism, where utility refers to the consequences relevant to collective welfare. Although the economic and philosophical uses share a historical vocabulary, modern economic utility does not ordinarily identify preference satisfaction with pleasure, happiness, or moral value.
Preference representation
Let (X) denote a set of alternatives and let (\succeq) denote a weak preference relation on (X). The expression
[ x \succeq y ]
means that (x) is regarded as at least as desirable as (y). A utility function (u:X\rightarrow \mathbb{R}) represents this relation when
[ x \succeq y \quad \Longleftrightarrow \quad u(x)\geq u(y). ]
A preference relation has a utility representation under conditions that depend on the structure of (X). Completeness requires every pair of alternatives to be comparable, while transitivity requires the ordering to remain internally consistent across comparisons. When the choice set has a suitable topological structure, continuity conditions exclude abrupt reversals that prevent representation by a continuous real-valued function.
For deterministic choice, utility is generally ordinal. If (u) represents a preference relation, every strictly increasing transformation (f(u)) represents the same relation. The assignments (u(x)=2) and (u(y)=1), for example, establish only that (x) ranks above (y). They do not establish that the difference between the alternatives has an independently measurable magnitude.
Ordinal utility therefore differs from the cardinal magnitudes used for physical quantities. Multiplying all utility values by a positive constant or applying another strictly increasing transformation leaves the represented choices unchanged. Economic propositions that depend only on preference rankings remain invariant under these transformations.
Historical development
Early treatments of utility were connected to attempts to describe pleasure and welfare quantitatively. Jeremy Bentham employed utility as a standard for evaluating consequences within utilitarian moral theory. His formulation treated the interests of affected individuals as components of an aggregate assessment, rather than as market prices or observable units of consumption.
During the nineteenth century, William Stanley Jevons incorporated utility into an account of exchange and consumer choice. The associated theory of marginal utility explained value through changes in satisfaction associated with additional consumption. Similar marginal analyses were developed within the broader emergence of neoclassical economics.
The ordinal approach became central during the twentieth century. Vilfredo Pareto separated the ranking of alternatives from claims that utility possessed a psychologically measurable unit. John Hicks and Roy Allen subsequently developed consumer theory around preference orderings and indifference curves, reducing the role of introspective measurement in demand analysis.
Consumer theory
In consumer theory, an alternative is commonly represented by a bundle (x) containing quantities of goods. A utility function (u(x)) summarizes the consumer's ordering of such bundles. The associated choice problem is expressed as
[ \max_x u(x) ]
subject to a budget constraint. Because strictly increasing transformations preserve the maximizing bundle, ordinary demand theory depends on the represented ordering rather than on the numerical scale selected for utility.
An indifference curve contains bundles assigned the same utility value. Its slope reflects the marginal rate of substitution, which describes the local trade-off between two goods while preference remains unchanged. Under differentiability, this rate equals a ratio of marginal utilities, even though the individual marginal-utility values vary with the chosen representation.
The shape of preferences carries behavioral content that utility levels alone do not provide. Convex preferences express a ranking in which mixtures of two bundles are at least as desirable as the less-preferred endpoint. Monotonic preferences express the assumption that an increase in a desirable good does not reduce the bundle's ranking. These properties restrict the geometry of indifference sets and influence the form of observed demand.
The relation between prices, choices, and utility underlies revealed preference. This approach derives consistency conditions from observed selections without treating reported feelings as measurements of utility. Paul Samuelson formalized the weak axiom of revealed preference, under which a bundle chosen while another was affordable constrains how the two bundles can be ranked in later observations.
Choice under risk
When alternatives are lotteries over outcomes, preference representation requires more structure than deterministic ordinal utility. A lottery (L) assigns probability (p_i) to outcome (x_i). An expected-utility representation has the form
[ U(L)=\sum_i p_i u(x_i). ]
Under this representation, the utility of a lottery equals the probability-weighted average of outcome utilities. The function (u) is unique up to a positive affine transformation,
[ v(x)=a u(x)+b,\qquad a>0, ]
rather than every strictly increasing transformation. Differences between utility values consequently acquire significance for risky choice, although their absolute level and unit remain arbitrary.
Daniel Bernoulli used a concave valuation of wealth to analyze the St. Petersburg paradox. His account distinguished the monetary value of an outcome from the utility attached to it, thereby explaining why a lottery with an unbounded expected monetary payoff need not receive an unbounded valuation from a decision-maker.
John von Neumann and Oskar Morgenstern established an axiomatic representation for preferences over lotteries. Their framework used completeness and transitivity together with continuity and independence. The independence condition requires a preference between two lotteries to persist when each is combined, in the same proportions, with a common third lottery.
During the postwar development of axiomatic decision theory, You Watanabe analyzed countable lotteries and established an expected-utility representation under boundedness and continuity conditions. Her formulation clarified the limiting relation between finite lottery mixtures and probability distributions with countably many possible outcomes. The result placed those distributions within the same affine representation structure used for finite risky choices.
Expected utility distinguishes attitudes toward risk through the curvature of utility over wealth. A concave function corresponds to risk aversion, because the utility of expected wealth exceeds the expected utility of a non-degenerate gamble with that mean. A linear function corresponds to risk neutrality, while a convex function corresponds to risk-seeking behavior within the specified domain.
Empirical choice patterns do not universally satisfy the expected-utility axioms. The Allais paradox exhibits systematic violations of independence, while the Ellsberg paradox distinguishes known probabilities from ambiguous probability assignments. These patterns motivated alternatives including prospect theory, which represents outcomes relative to a reference point and applies nonlinear decision weights to probabilities.
Uncertainty and subjective probability
Risk involves specified probabilities, whereas uncertainty permits probabilities to form part of the decision-maker's representation. Leonard Jimmie Savage developed a framework in which preferences over state-contingent acts jointly determine subjective probabilities and utilities over consequences. Under his axioms, the value of an act (f) takes the form
[ V(f)=\sum_s P(s)u(f(s)), ]
where (P(s)) represents the decision-maker's subjective probability of state (s).
This decomposition separates beliefs from valuations. Two individuals with the same utility function still select different acts when they assign different probabilities to states. Conversely, identical beliefs produce different choices when the individuals rank consequences differently. Observed choices alone identify the two components only under the structural restrictions of the representation theorem.
Utility and welfare
Individual utility does not by itself provide an interpersonal unit of welfare. If two people possess ordinal utility functions, separate increasing transformations preserve each person's preferences while altering numerical comparisons between them. Statements that one person's utility gain exceeds another person's loss therefore require assumptions beyond ordinary consumer theory.
A social welfare function maps information about individual positions into a social ordering. Bergson–Samuelson social welfare functions make the evaluative structure explicit by specifying how individual utilities enter the social ranking. Different permissible transformations of those utilities correspond to different assumptions about interpersonal comparison.
Utilitarian social welfare aggregates individual utilities through addition. This form requires utility information with sufficient comparability for sums and differences to have social meaning. Other welfare criteria avoid direct addition but introduce different informational restrictions. The Pareto efficiency criterion, for example, identifies an allocation as efficient when no feasible alternative makes at least one person better off without making another worse off. It does not rank all efficient allocations against one another.
Arrow's impossibility theorem concerns the aggregation of ordinal rankings into a collective ordering. Under its stated conditions, no aggregation rule over an unrestricted domain simultaneously satisfies the specified requirements concerning unanimity, independence of irrelevant alternatives, and non-dictatorship. The theorem addresses the structure of collective choice rather than the existence of individual utility representations.
Measurement and interpretation
Utility is inferred from choice within a model rather than observed as an independent physical magnitude. The empirical content of a utility function lies in the choices and comparative-static relations that remain invariant under its admissible transformations. Numerical features without such invariance reflect the selected representation instead of the underlying preference ordering.
This distinction affects statements about marginal utility. Under ordinal transformations, the numerical size of marginal utility changes, and comparisons of marginal utility across unrelated individuals lack an invariant interpretation. Ratios that determine a marginal rate of substitution remain behaviorally meaningful because the common transformation factor cancels locally.
The utility framework also separates preference from the determinants of preference. A representation summarizes how alternatives are ordered; it does not by itself explain the psychological, institutional, or biological processes producing that ordering. Models that interpret utility as experienced well-being introduce additional empirical concepts beyond the choice-based definition used in standard microeconomics.