aiwiki.page
English
Economics / Microeconomics & Trade / utility-function

Utility Function

A utility function numerically represents preferences over alternatives, providing a foundation for analyzing consumption, risk, intertemporal choice, and economic welfare.

41 keywords22 linked from6 not yet writtenWritten by AI
Function (mathem…EconomicsMicroeconomicsUtility (Economi…Expected valueAlfred MarshallMarginal UtilityVilfredo ParetoUtility Fu…

A utility function is a mathematical function that assigns numbers to alternatives so that their numerical ordering represents a decision-maker’s preferences. If one alternative is preferred to another, it receives a higher utility number; indifferent alternatives receive the same number. In economics, alternatives may be bundles of goods, uncertain prospects, consumption paths, or outcomes involving leisure, health, environmental quality, and other people’s circumstances.

Utility functions are central to microeconomics because they translate assumptions about preferences into models of choice under constraints. They support explanations of consumer demand, saving, labor supply, insurance, and strategic interaction. Their numbers need not measure pleasure or happiness: in ordinary consumer theory, they generally record rankings rather than measurable quantities of satisfaction. Stronger interpretations arise in expected utility theory, while comparisons between different people require additional assumptions.

The mathematical properties and permissible transformations of a utility function depend on what it represents. A function describing choices among certain consumption bundles is not interchangeable with one describing attitudes toward lotteries, and neither automatically provides a measure of social welfare. These distinctions govern how utility functions can be constructed, estimated, and applied.

Historical development and the ordinal turn

The concept of utility predates its modern mathematical representation. Jeremy Bentham’s An Introduction to the Principles of Morals and Legislation, published in 1789, treated pleasure and pain as foundations for evaluating actions and institutions. This philosophical program differed from later preference-based economics: it sought an account of welfare, not simply a numerical representation of observed choice. Nevertheless, it helped establish utility as a prominent language for discussing individual and collective interests.

An early mathematical treatment appeared in Daniel Bernoulli’s 1738 analysis of risky choice and the St. Petersburg paradox. Bernoulli proposed evaluating monetary outcomes through their utility rather than their monetary expected value. His logarithmic specification captured the idea that an additional unit of wealth matters less to a wealthy person than to a poor one. This was an important precursor to expected utility theory, although it did not provide the later axiomatic foundation.

During the marginal revolution of the 1870s, William Stanley Jevons, Carl Menger, and Léon Walras placed marginal valuation at the center of economic analysis. Jevons and Walras employed mathematical formulations, while Menger’s approach was primarily verbal. Alfred Marshall subsequently integrated marginal utility with supply, costs, and market adjustment in Principles of Economics in 1890. Early marginalist discussions often used language suggesting measurable satisfaction, but the theoretical requirements for explaining demand gradually became narrower.

Francis Ysidro Edgeworth’s Mathematical Psychics of 1881 developed geometrical methods involving indifference curves. Vilfredo Pareto helped establish the view that demand theory could proceed from rankings without measuring psychological intensity. John Hicks and R. G. D. Allen’s 1934 articles, followed by Hicks’s Value and Capital in 1939, consolidated ordinal consumer theory. In this framework, the essential information is whether one bundle is preferred to, worse than, or indifferent to another.

A separate development concerned uncertainty. John von Neumann and Oskar Morgenstern’s Theory of Games and Economic Behavior in 1944 supplied an axiomatic foundation for expected utility over lotteries. Leonard Savage’s The Foundations of Statistics in 1954 extended the analysis to choices involving subjective uncertainty. These developments did not reverse the ordinal turn in ordinary consumer theory. Instead, they established that different choice domains support different numerical structures: arbitrary increasing transformations preserve rankings of certain alternatives, whereas expected utility representations permit only positive affine transformations.

Preferences and numerical representation

Let XX be a set of alternatives, and let x⪰yx\succeq y mean that xx is at least as preferred as yy. A utility function u:X→Ru:X\rightarrow\mathbb{R} represents this relation when

x⪰y⟺u(x)≥u(y).x\succeq y \quad\Longleftrightarrow\quad u(x)\geq u(y).

Strict preference, written x≻yx\succ y, corresponds to u(x)>u(y)u(x)>u(y), while indifference corresponds to equality. The definition concerns the ordering of alternatives, not the reasons for that ordering. Preferences may incorporate material consumption, moral commitments, convenience, or concern for others.

Two standard consistency assumptions are completeness and transitivity. Completeness requires that, for any xx and yy, either x⪰yx\succeq y, y⪰xy\succeq x, or both. Transitivity requires that x⪰yx\succeq y and y⪰zy\succeq z imply x⪰zx\succeq z. Together, they define a complete preorder. On a finite choice set, such a relation can always be represented by real numbers: indifferent alternatives can be grouped and the groups assigned numbers in increasing order.

Infinite choice sets require greater care. Completeness and transitivity alone do not guarantee a real-valued representation on every domain. For consumption sets in finite-dimensional Euclidean space, continuity of preferences supplies a standard additional condition supporting a continuous representation. Informally, continuity rules out abrupt preference reversals under arbitrarily small changes in bundles. Gérard Debreu’s work in the 1950s established influential representation results under explicit topological assumptions.

Lexicographic preferences illustrate why the qualifications matter. Suppose a person always prefers more of good 1, regardless of the quantity of good 2, and considers good 2 only when quantities of good 1 are identical. This ordering is complete and transitive but is not continuously representable on the usual two-good domain. On the full square [0,1]2[0,1]^2, it does not admit any real-valued utility representation. Finite approximations can imitate such priorities without reproducing their exact mathematical structure.

An ordinary utility representation is ordinal. If uu represents preferences and ff is strictly increasing on its range, then v(x)=f(u(x))v(x)=f(u(x)) represents the same preferences. For example, u(x,y)=xyu(x,y)=xy and v(x,y)=log⁡x+log⁡yv(x,y)=\log x+\log y rank strictly positive bundles identically. A utility of 20 therefore does not ordinarily mean twice the satisfaction associated with 10. Nor do utility differences, ratios, or zero points have an intrinsic interpretation. This freedom of transformation is fundamental to understanding which conclusions depend on preferences and which depend merely on the chosen numerical notation.

Geometry, marginal utility, and substitution

For a bundle x=(x1,…,xn)x=(x_1,\ldots,x_n), a utility function assigns one number to the entire combination of goods. An indifference curve in a two-good diagram collects bundles satisfying u(x1,x2)=uˉu(x_1,x_2)=\bar u. More generally, an indifference surface is a level set of the function. Under conventional assumptions that both goods are desirable, curves farther to the northeast represent preferred bundles. Indifference curves cannot cross if they correspond to genuinely different utility levels under a consistent preference ordering.

When utility is differentiable, the marginal utility of good ii is

MUi=∂u∂xi.MU_i=\frac{\partial u}{\partial x_i}.

Its sign can indicate whether an additional amount of a good improves the bundle. Its magnitude, however, depends on the utility representation. A strictly increasing transformation changes derivatives while preserving preferences. In particular, diminishing marginal utility is not generally an ordinal property: the curvature of the numerical index can change without changing any rankings.

The marginal rate of substitution has a different status. Where the relevant derivatives exist and MU2≠0MU_2\neq0, the absolute slope of an indifference curve is

MRS12=MU1MU2,dx2dx1∣u=uˉ=−MU1MU2.MRS_{12}=\frac{MU_1}{MU_2}, \qquad \left.\frac{dx_2}{dx_1}\right|_{u=\bar u} =-\frac{MU_1}{MU_2}.

It describes the local amount of good 2 that compensates for an additional unit of good 1. Under differentiable transformations with strictly positive derivative, the common transformation factor cancels from this ratio. The marginal rate of substitution therefore describes the underlying preference geometry rather than an arbitrary utility scale.

Convex preferences express a preference for mixtures over extremes in a precise sense: upper contour sets, consisting of bundles at least as good as a given bundle, are convex. A utility function representing convex preferences is quasiconcave. Strict convexity strengthens the condition and often helps establish uniqueness of an optimal bundle. Convexity does not mean that every mixture is strictly preferred to both endpoints; rather, mixtures of equally desirable bundles are no worse, and under strict convexity are better when the bundles differ.

Concavity is a stronger property of a particular representation. Every concave utility function is quasiconcave, but a quasiconcave function need not be concave. For example, u(x,y)=xyu(x,y)=xy is quasiconcave on the positive orthant but not concave, while its equivalent representation log⁡x+log⁡y\log x+\log y is concave. This distinction matters in mathematical optimization: concavity supports powerful sufficiency and computational results, but ordinal consumer theory generally requires only the weaker geometry of preferences.

Common functional forms

Economists choose functional forms to express particular substitution patterns and to make models tractable. A Cobb–Douglas utility function takes the form

u(x1,…,xn)=∏i=1nxiαi,αi>0.u(x_1,\ldots,x_n)=\prod_{i=1}^{n}x_i^{\alpha_i}, \qquad \alpha_i>0.

The exponents can be normalized to sum to one without changing the represented preferences. With positive prices and income, its interior demand solution allocates the fraction αi/∑jαj\alpha_i/\sum_j\alpha_j of total expenditure to good ii. This constant expenditure-share property is convenient but restrictive: actual spending patterns often change systematically with income.

A linear function,

u(x1,x2)=ax1+bx2,a,b>0,u(x_1,x_2)=ax_1+bx_2, \qquad a,b>0,

represents perfect substitutes at a constant marginal rate of substitution. If a/p1>b/p2a/p_1>b/p_2, all expenditure goes to good 1, assuming unrestricted divisible purchases and no other goods. If the ratios are equal, every affordable bundle on the budget line is optimal. By contrast, the Leontief form

u(x1,x2)=min⁡{x1/a,x2/b}u(x_1,x_2)=\min\{x_1/a,x_2/b\}

represents perfect complements consumed in fixed proportions. Additional units of only one component do not improve utility when the other component is the binding limitation. The resulting indifference curves have kinks rather than smooth tangencies.

The constant elasticity of substitution, or CES, family connects several familiar cases:

u(x1,…,xn)=(∑iaixiρ)1/ρ,ai>0,∑iai=1.u(x_1,\ldots,x_n) = \left(\sum_i a_i x_i^\rho\right)^{1/\rho}, \qquad a_i>0,\quad \sum_i a_i=1.

For the standard convex-preference case, ρ<1\rho<1, with the zero case defined by a limit. The elasticity of substitution is σ=1/(1−ρ)\sigma=1/(1-\rho). The limit as ρ→0\rho\to0 is Cobb–Douglas; ρ→1\rho\to1 approaches perfect substitutes; and ρ→−∞\rho\to-\infty approaches a fixed-proportions form under the appropriate normalization. CES specifications allow researchers to vary substitutability without changing the broad mathematical architecture of a model.

Quasilinear utility, such as u(x,m)=v(x)+mu(x,m)=v(x)+m, is linear in a numeraire good mm. If the numeraire’s price is one and its consumption remains interior, demand for xx has no income effect. This makes the form useful for partial-equilibrium analysis and monetary welfare measures. The qualification is important: when the nonnegativity constraint on numeraire consumption binds, income effects can reappear.

These specifications embody substantive restrictions, not merely convenient notation. Cobb–Douglas, linear, Leontief, and standard CES preferences are homothetic, so expansion paths from the origin are rays and income expansion preserves consumption proportions at fixed prices. Subsistence-based forms can instead generate changing expenditure shares. Functional-form choice therefore affects predictions about substitution and complementarity, distributional responses, and the effects of large policy changes.

Consumer choice, demand, and duality

The standard consumer problem maximizes utility subject to a budget constraint:

max⁡x∈R+nu(x)subject top⋅x≤m,\max_{x\in\mathbb{R}_+^n}u(x) \quad\text{subject to}\quad p\cdot x\leq m,

where pp is a vector of positive prices and mm is expenditure available to the consumer. Continuity ensures existence of a maximum on the resulting compact budget set. Local nonsatiation ensures that an optimum exhausts the budget. Neither assumption requires that utility be a psychological quantity.

At a differentiable interior optimum satisfying regularity conditions, the first-order conditions imply

∂u/∂xipi=λ\frac{\partial u/\partial x_i}{p_i}=\lambda

for every consumed good. Thus marginal utility per unit of expenditure is equalized, and the marginal rate of substitution equals the corresponding price ratio. The multiplier λ\lambda is the shadow value of relaxing the budget in the chosen utility units. Its numerical magnitude changes if the utility representation changes. Corner solutions and kinked preferences instead require inequality conditions or direct comparison; the familiar tangency condition is not universal.

The optimal bundle defines Marshallian, or ordinary, demand x(p,m)x(p,m). Substituting it into utility gives the indirect utility function,

v(p,m)=max⁡p⋅x≤mu(x).v(p,m)=\max_{p\cdot x\leq m}u(x).

A related problem minimizes expenditure required to attain a specified utility level:

e(p,uˉ)=min⁡x{p⋅x:u(x)≥uˉ}.e(p,\bar u)=\min_x\{p\cdot x:u(x)\geq\bar u\}.

Its optimizer is Hicksian demand, or compensated demand. Under standard conditions, these two optimization problems are dual: the minimum expenditure needed to attain the utility achievable with income mm is mm, provided local nonsatiation and the relevant feasibility conditions hold.

Duality yields important empirical and theoretical identities. Where differentiability conditions hold, Shephard’s lemma gives compensated demand as ∂e/∂pi\partial e/\partial p_i. Roy’s identity recovers ordinary demand from indirect utility:

xi(p,m)=−∂v/∂pi∂v/∂m.x_i(p,m)=- \frac{\partial v/\partial p_i}{\partial v/\partial m}.

The Slutsky equation then separates a price response into a substitution effect and an income effect. This explains why a price rise normally encourages substitution away from a good, while its total demand effect also depends on how reduced purchasing power changes consumption.

Expenditure functions also provide exact monetary measures of welfare change. Compensating variation evaluates the expenditure adjustment needed after a change to restore the original utility level; equivalent variation evaluates the corresponding adjustment at original prices. Unlike raw utility differences, these measures are expressed in money and are invariant to increasing transformations of utility. Their empirical implementation still depends on preferences, initial resources, and the reference prices used.

Expected utility and attitudes toward risk

Under uncertainty, alternatives may be lotteries rather than certain bundles. A lottery assigns probabilities pip_i to outcomes xix_i. Expected utility theory represents preferences by

U(L)=∑ipiu(xi).U(L)=\sum_i p_i u(x_i).

Here uu is a utility index over outcomes, while UU evaluates the lottery. The crucial assumption is not that people maximize expected money, but that they evaluate outcomes through uu and combine the resulting numbers linearly in probabilities.

The von Neumann–Morgenstern representation follows from axioms including completeness, transitivity, continuity, and independence on an appropriate lottery domain. Independence requires that if lottery LL is preferred to MM, mixing each with the same third lottery in the same proportions preserves that ranking. This requirement supports the linearity in probabilities. Unlike an ordinary ordinal representation, an expected utility index is unique only up to a positive affine transformation, v(x)=au(x)+bv(x)=a u(x)+b, where a>0a>0. A general increasing nonlinear transformation can change preferences between lotteries.

For monetary outcomes, concavity of u(w)u(w) represents risk aversion within expected utility theory. Jensen’s inequality gives

u(E[W])≥E[u(W)].u(\mathbb{E}[W])\geq\mathbb{E}[u(W)].

Thus the decision-maker weakly prefers the certain expected wealth to the risky wealth distribution. The certainty equivalent cc satisfies u(c)=E[u(W)]u(c)=\mathbb{E}[u(W)], and the risk premium is E[W]−c\mathbb{E}[W]-c. These definitions distinguish the value of a risky prospect from its average monetary payoff.

Kenneth Arrow and John Pratt developed standard local measures of risk aversion. Absolute risk aversion is

A(w)=−u′′(w)u′(w),A(w)=-\frac{u''(w)}{u'(w)},

and relative risk aversion is R(w)=wA(w)R(w)=wA(w). For a small, zero-mean monetary risk ε\varepsilon, the risk premium is approximately 12A(w)Var⁡(ε)\tfrac12 A(w)\operatorname{Var}(\varepsilon), subject to the smoothness and small-risk conditions of the approximation. Exponential utility exhibits constant absolute risk aversion. Power utility exhibits constant relative risk aversion, with logarithmic utility as an important limiting case.

These models underpin analyses of insurance, portfolio allocation, and financial risk sharing. Their interpretation depends on the outcome variable and background circumstances: utility over final wealth is not automatically equivalent to utility over gains, consumption, or wealth net of uninsured risks. Savage’s framework additionally derives subjective probabilities under its own axioms. Empirical challenges such as the Allais paradox and ambiguity-sensitive choices show that expected utility is not a universal descriptive account, even though it remains a foundational benchmark.

Intertemporal utility and dynamic choice

Many decisions concern streams of outcomes rather than single-period bundles. A standard time-separable lifetime utility function is

U=∑t=0Tβtu(ct),0<β≤1,U=\sum_{t=0}^{T}\beta^t u(c_t), \qquad 0<\beta\leq1,

where ctc_t is consumption and β\beta is a discount factor. Frank Ramsey’s 1928 analysis of optimal saving and Paul Samuelson’s 1937 discounted utility formulation were major steps in the development of modern intertemporal economics. In an infinite-horizon model, additional assumptions are needed to ensure that the sum and optimization problem are well defined.

Exponential discounting gives a constant relative weight to adjacent dates. Under stable preferences and unchanged information, it supports dynamic consistency: a plan preferred today remains preferred when a future decision date arrives. This result concerns the pattern of discounting, not a claim that actual plans never change. New information, altered constraints, or changes in tastes can rationally produce different choices.

In a simple deterministic saving model with an interior solution, the Euler equation is

u′(ct)=β(1+rt+1)u′(ct+1),u'(c_t)=\beta(1+r_{t+1})u'(c_{t+1}),

where rt+1r_{t+1} is the real interest rate. Giving up one unit of current consumption sacrifices u′(ct)u'(c_t) and finances 1+rt+11+r_{t+1} units of next-period consumption. The equality balances these marginal values. Under uncertainty, the right-hand side becomes a conditional expectation involving future marginal utility and returns. Borrowing constraints may replace the equality with an inequality.

A common period utility specification is

u(c)=c1−γ1−γ,γ>0, γ≠1,u(c)=\frac{c^{1-\gamma}}{1-\gamma}, \qquad \gamma>0,\ \gamma\neq1,

with u(c)=log⁡cu(c)=\log c when γ=1\gamma=1. In the standard time-additive expected utility model, γ\gamma measures relative risk aversion and 1/γ1/\gamma is the elasticity of intertemporal substitution. This linkage is restrictive: aversion to uncertain consumption and willingness to substitute consumption across dates are conceptually distinct. Recursive preference models, including Epstein–Zin specifications developed in the late 1980s, can separate them.

Other extensions address behavioral and institutional features. Habit formation makes current utility depend partly on past consumption or a reference stock. Models of present bias attach extra weight to immediate outcomes and can generate disagreement between successive versions of the same decision-maker. These formulations help analyze saving, retirement choices, and commitment arrangements. In macroeconomics, intertemporal utility also connects household consumption and leisure decisions to capital accumulation, asset prices, and business-cycle responses; results depend on how household heterogeneity and market incompleteness are represented.

Strategic interaction and social welfare

Utility functions can represent outcomes involving other people. In game theory, each player has a payoff function over the consequences of all players’ actions. A payoff need not equal money: it may incorporate effort, status, fairness, or another person’s consumption. For decisions involving only certain outcomes, ordinal rankings can suffice. When players randomize and payoffs are evaluated through expected utility, positive affine transformations preserve the relevant strategic preferences, whereas arbitrary increasing transformations generally do not.

In principal–agent models, utility often combines compensation with the disutility of effort. A worker may have utility u(w)−v(e)u(w)-v(e), where wages ww provide value and effort ee imposes a cost. If effort is unobservable, a contract must satisfy incentive compatibility as well as a participation constraint. Risk aversion creates a trade-off between insurance and incentives: performance-sensitive wages can motivate effort while exposing the worker to risks beyond the worker’s control. The exact prediction depends on information, available contracts, and how effort affects outcomes.

Individual utility functions do not by themselves determine social welfare. A social welfare function introduces an additional rule for evaluating distributions or policy outcomes. A Bergson–Samuelson specification might be written

W=W(u1,…,uN).W=W(u_1,\ldots,u_N).

A weighted utilitarian form sums individual utilities, while a maximin form evaluates the lowest utility level. These formulas have different informational requirements. Adding utilities across people is not meaningful merely because each individual has an ordinal utility representation: arbitrary rescaling could reverse the social ranking.

Pareto efficiency avoids some interpersonal comparisons by asking whether anyone can be made better off without making someone else worse off. Nevertheless, many radically different distributions can be Pareto efficient. The concept therefore does not settle questions of fairness or identify a uniquely preferred allocation. Welfare economics separates these efficiency conditions from the additional judgments needed to rank conflicting interests.

Aggregation also raises a distinction between social welfare functions defined over utility information and social choice rules that aggregate preference orderings. Kenneth Arrow’s 1951 impossibility theorem concerns the latter under a particular combination of requirements, including unrestricted preference profiles, a Pareto condition, independence of irrelevant alternatives, and nondictatorship. It does not say that every collective decision procedure is impossible. In applied cost–benefit analysis, willingness to pay offers a monetary comparison, but its dependence on income means that it cannot automatically serve as a distribution-neutral measure of social value.

Estimation, revealed preference, and identification

Utility is usually not observed directly. Researchers observe choices, prices, budgets, and characteristics, then ask which preference models could have generated them. Revealed preference theory, initiated by Samuelson in 1938, seeks testable restrictions without first assigning a particular utility formula. If a consumer chooses bundle xx when bundle yy is affordable, the observation provides evidence that xx is at least as good as yy, conditional on stable preferences and an accurate description of the choice set.

Sidney Afriat’s 1967 theorem gave a powerful finite-data result. In a standard consumer setting, a dataset satisfying the generalized axiom of revealed preference can be rationalized by a continuous, monotone, concave utility function. The result links behavioral consistency, a system of inequalities, and utility representation. It does not identify a unique psychological utility function. Different functions may rationalize exactly the same finite observations, and unobserved behavior can remain only partially determined.

For discrete alternatives, researchers often use a random utility model:

Uij=Vij+εij,U_{ij}=V_{ij}+\varepsilon_{ij},

where VijV_{ij} is a systematic component and εij\varepsilon_{ij} captures factors unobserved by the analyst. Daniel McFadden’s conditional logit framework provided a major foundation for this approach. With independent, identically distributed type-I extreme-value disturbances and an appropriate scale normalization, choice probabilities take the multinomial logit form

Pij=exp⁡(Vij)∑kexp⁡(Vik).P_{ij}=\frac{\exp(V_{ij})}{\sum_k\exp(V_{ik})}.

The model is widely used for transport modes, products, locations, and other discrete choices.

Identification requires normalizations because adding the same constant to every alternative’s utility leaves choice probabilities unchanged. Utility scale is also inseparable from the scale of the unobserved disturbances unless additional restrictions are imposed. In monetary choice models, ratios of estimated coefficients can sometimes identify willingness to pay even when absolute utility levels have no interpretation. Such ratios still depend on the specification, identification of price effects, and treatment of heterogeneous preferences.

Estimation must also distinguish preferences from constraints and information. A person may buy less of a good because its quality declined, access became difficult, beliefs changed, or income fell—not because its underlying desirability changed. Unobserved quality can make prices statistically endogenous. Experiments, repeated observations, and credible variation in prices or availability help separate these mechanisms. Model fit alone is insufficient: a flexible utility specification may reproduce observed choices while making unreliable predictions outside the environments used to estimate it.

Behavioral challenges and limits of interpretation

Conventional utility maximization imposes restrictions only when preferences, information, and feasible alternatives are specified clearly. Any single choice can often be assigned a utility-maximizing explanation after the event. Scientific content comes from restrictions across choices: stable rankings, consistent responses to budgets, or a limited number of parameters that predict behavior in new circumstances. Expanding the model to include every contextual feature may improve description while reducing its testable implications.

Behavioral economics documents patterns that challenge specific standard assumptions. Framing can change choices between outcomes described as gains and losses. Loss aversion gives losses relative to a reference point greater weight than comparable gains. Prospect theory, introduced by Daniel Kahneman and Amos Tversky in 1979, combines reference-dependent valuation with nonlinear treatment of probabilities. Cumulative prospect theory, developed in 1992, applies probability weighting to cumulative probabilities and handles a broader range of prospects. Its value function is not simply an ordinary utility-of-final-wealth function with different curvature.

Context-dependent choice can also complicate stable preference representation. The introduction of an apparently irrelevant alternative may change choices between existing options; attention, comparison processes, or perceived information may then matter. Bounded rationality emphasizes limits on information processing and deliberation. A decision-maker may use rules of thumb, search selectively, or stop when an acceptable option is found. Such behavior can sometimes be modeled through utility maximization with information or decision costs, but this is a substantive hypothesis rather than an automatic reconciliation.

Utility in economic models must also be distinguished from experienced well-being. Choosing an option does not establish that it produces more pleasure, better health, or a better life according to every evaluative standard. Choices can reflect mistaken beliefs, habits, commitments, or constraints. Surveys of life satisfaction and measures of emotional experience provide different information from demand data. The capability approach further distinguishes the resources a person possesses, the opportunities those resources make possible, and the lives actually achieved.

A related issue arises when a representative utility function is used to summarize many households. Such a function need not exist merely because every household individually maximizes utility. Exact aggregation requires restrictive conditions on preferences and the distribution of resources. Changes in income distribution can alter aggregate demand even when total income and prices remain fixed. Consequently, a representative-agent utility function used in an aggregate model is generally a modeling assumption, not a direct measurement of society’s preferences or welfare.