The Utility

Every price is a measurement, and behind every measurement lies a thing measured. Behind price lies utility — the economist's name for the satisfaction, the want-fulfillment, the well-being that a thing affords the person who has it. It is not the price itself, which is only the measure, but the underlying value-to-the-person that the price is meant to register: the worth a thing has in the wanting, not in the market. The marginal revolution of an earlier volume made this subjective value the wellspring of all price, and so made utility the quantity at the center of the subject. Demand is its shadow; price is its trace; choice is its pursuit. Whatever else economics studies, it studies, at bottom, the getting of utility.

The concept was given its grand ambition by Jeremy Bentham, who held that nature had placed mankind under two sovereign masters, pleasure and pain, and who dreamed of a felicific calculus that would weigh the one against the other and add the happiness of a whole society as one adds its wealth. On this vision, utility is the currency of well-being, common to every pleasure and every pain, and the right action — for a person or a government — is the one that yields the greatest sum of it: the greatest happiness of the greatest number. The dream founded a moral philosophy, utilitarianism, and it lent economics its deepest aspiration — to make happiness a quantity that could be measured, compared, and maximized, the foundation alike of private choice and public policy.

But the quantity has a peculiar and unsettling character, and the whole volume turns on it. Utility is the thing everything in economics is built to track, and no instrument can read it. You cannot see satisfaction, weigh it, or measure it directly; there is no thermometer of pleasure, no scale of contentment. It can be approached only obliquely, inferred from the choices it supposedly drives — we say a person wanted the thing they chose, and wanted it more than what they passed over. This is the ghost in the machine: the quantity that the entire apparatus presupposes and that no observation can capture. The plates that follow are the successive attempts to pin it down — to measure it, to demote it to a mere ranking, to read it off from choice, to extend it to risk, to compare it across persons, to ask whether it is even the same as the well-being it was meant to name — and the characteristic way each attempt, however ingenious, falls short.

The first marginalists took Bentham at his word and treated utility as a magnitude that could be measured. William Stanley Jevons and Alfred Marshall wrote as though satisfaction came in units — units later named, half in earnest and half in jest, "utils" — so that one might say not merely that a person preferred the first loaf to the second, but that the first yielded twice the utility of the second, or seven units against three. This is cardinal utility: utility as a quantity like length or weight or temperature, possessed of magnitudes and ratios, addable and divisible, in principle as measurable as anything else the sciences weigh. The ambition was nothing less than to make happiness numerical.

On this view the utility a person draws from a good has a definite shape as the quantity grows. Total utility rises with each additional unit, but at a diminishing rate — for the marginal utility of each successive unit is less than the last, the diminishing marginal utility of an earlier volume. The figure draws the consequence: the total-utility curve climbs and flattens, concave to the quantity axis, its slope — the marginal utility — falling toward zero as satiation approaches. This concavity is the cardinal expression of a basic fact about wants, that they are sated; and it is no idle shape, for it grounds the downward slope of demand, as the volume on the margin showed, and it will ground, in a later plate, the aversion to risk. The bend in the curve does real work.

The trouble was that the unit was never found. Temperature has its thermometer and weight its balance; utility had no instrument, no agreed zero, no scale on which two people's readings — or even one person's, across two goods — could be set side by side. Worse, the cardinal magnitudes appeared to do no work that mere ranking could not do as well: to derive the demand curve one needs only that marginal utility falls, never by exactly how much. A measurement that no instrument can take and that changes no prediction began to look like a metaphysical extravagance — an unmeasured measurement, a number with no observable consequence. The suspicion hardened into a program, the subject of the next plate, which asked how much of the cardinal apparatus could simply be discarded.

The answer, Vilfredo Pareto saw, and John Hicks and Eugen Slutsky made exact, is: almost nothing. To build a theory of demand one does not need to measure utility at all — one needs only to rank. It is enough that a person can say which of two bundles they prefer, or that they are indifferent between them; the magnitudes the cardinalists chased are never used. Utility, on this account, is purely ordinal: only the order of the bundles matters, not the spacing between them, and any renumbering that preserves the order — any increasing transformation of the utility function — yields exactly the same theory. The numbers lose their meaning as quantities and keep only their meaning as a sequence.

The instrument of this austerity is the indifference curve and the map it belongs to. An indifference curve joins all the bundles among which a person is indifferent — equally preferred, equally good in their eyes. A whole family of such curves, the indifference map, represents the entire preference ordering, as the figure shows: each curve is a level of satisfaction, and the curves further from the origin are the more preferred, that ranking being the only thing the map asserts. No numbers are written on the curves; were one to label them, any increasing labels would serve. From this map alone — together with a budget line — the consumer's choice follows, at the tangency where the marginal rate of substitution meets the price ratio, exactly as the volume on the margin established. The whole of riskless demand is rebuilt on the bare ranking of bundles.

Utility itself is not abolished but demoted. It survives as a representation — a function that assigns higher numbers to more-preferred bundles — but the numbers are now mere labels, like the ordinals on a list of finishers in a race, which record who came first and second and third and say nothing of the margins between them. One may speak of a utility function still, for convenience, but it is understood to be one of infinitely many that would do as well, all agreeing on the order and differing in everything else. Preference is the primitive; utility is its shadow. The subject has traded a measurement it could never make for a ranking it can, and lost nothing in the theory of demand by the trade.

If utility is merely a representation of preference, the natural question is what a preference must be like for such a representation to exist — for a utility function to assign its numbers consistently with the ranking. The answer is a small set of conditions on the preference relation, the axioms of rational choice, together with a theorem that ties them to a utility function. The axioms are few and look modest, and on them the entire formal edifice of choice is balanced.

Two are central. The first is completeness: for any two bundles, the chooser has a view — prefers the one, prefers the other, or is indifferent — with no gaps, no pair on which preference simply runs out. The second is transitivity: if a person prefers A to B and B to C, then they prefer A to C; preferences do not run in circles. With a technical condition of continuity added, these axioms yield the representation theorem — the guarantee that a utility function exists assigning higher numbers to more-preferred bundles. And they define what the economist means by rationality, which is a remarkably spare thing: to be rational is only to have a complete and transitive ranking, a coherent and circle-free ordering of the options. It is not to choose wisely, nor to want good things; it is only to choose consistently.

The thinness is the point, and it is also the vulnerability. The axioms say nothing about the content of preference — what a person ranks above what. One may prefer to ruin one's health, to act against one's own interest, to want what is trivial or self-destructive, and still be perfectly rational in this sense, so long as the wanting is complete and transitive; a coherent utility function will describe the choices of the consistent fool as faithfully as those of the sage. Rationality here governs the form of preference, its coherence, not its wisdom — a deliberate restraint that lets economics describe any consistent chooser without presuming to judge their ends. But it is also the doorway to every later critique, for the representation stands or falls with the axioms: where they are violated, no utility function exists to represent the preference, and the whole apparatus that rests on it gives way.

The ordinal theory had demoted utility to a representation of preference, but it still spoke of preference as a mental fact lying behind choice. Paul Samuelson proposed to go one step further and purge even that — to make choice itself the primitive, and preference merely its summary. Instead of beginning with preferences and deducing what a person will choose, begin with what they are observed to choose and infer the preferences. If a person selects bundle A when bundle B was equally within their means, they have revealed a preference for A over B; and this revealed relation, assembled from actual choices, is all that the theory needs or is entitled to claim. Preference is no longer the cause we posit behind choice but the pattern we read off from it.

The figure shows the elementary inference. A budget line marks what the person can afford; they choose the bundle A upon it; and there is another bundle, B, that lay within the budget and was passed over. The selection of A while B was affordable reveals A preferred to B — no feeling consulted, only a choice observed and a conclusion drawn. For such inferences to cohere into anything resembling a utility ordering, the choices must be consistent, and the requirement has a precise form, the weak axiom of revealed preference: if A is ever revealed preferred to B, then B must never, in any other budget, be revealed preferred to A. From this consistency, suitably strengthened, an entire preference ordering can be reconstructed — and behind it a utility function — built up from choices alone, with no mental state ever named.

This is operationalism brought into economics — the doctrine that a concept means no more than the observations that would test it. Utility, on this austere reading, is not a hidden satisfaction that choice reveals; it is nothing whatever but a compact description of consistent choosing, carrying no surplus meaning about any feeling. The appeal is rigor and observability: every statement about utility now redeems itself in choices one could, in principle, sit and watch. The theory becomes, in effect, behaviorist — an account of what people do, scrupulously silent about what they feel. It is the high-water mark of utility-as-fiction, the concept stripped of all psychology and surviving only as the shadow that behavior casts.

Consider a wager. A fair coin is tossed until it first falls heads; if the first head appears on the nth toss, the prize is two raised to the nth power — two ducats, or four, or eight, doubling with each toss the head is delayed. What is this game worth? Its expected value is the sum, over every possible round, of that round's probability times its prize — and since each round contributes one half times two, a constant, and the rounds are infinite in number, the sum grows without bound. The game is worth, on average, an infinite amount. Yet set it before anyone and ask what they would pay to play, and the answer is a small sum — a few ducats, no more. Here is the St. Petersburg paradox, which baffled the early students of probability: if people chose by expected value, they would surrender any fortune for this game, and not one of them will.

Daniel Bernoulli found the resolution, and it is the seed of the whole modern theory. People weigh not the expected money but the expected utility, and the utility of money diminishes at the margin, along the concave curve of an earlier plate. A prize twice as large is not twice as good; the thousandth ducat adds far less than the first. So the doubling prizes — two, four, eight, growing without limit — translate into utilities that grow ever more slowly, and the expected utility of the game, unlike its expected value, is finite. The figure sets the two against each other: round upon round, the expected value climbs forever, while the expected utility bends and settles toward a finite ceiling. That finite ceiling is what a person will pay. Measure the game in utility rather than in money, and the paradox simply dissolves.

Bernoulli's insight founded the theory of choice under risk. When outcomes are uncertain, rank the options not by expected value but by expected utility — the probability-weighted average of the utilities of the possible outcomes. This is the principle that governs the whole of economics where chance enters: the buying of insurance, the holding of risky assets, the premium demanded for bearing uncertainty, the subject of the volume on the premium. And it restores to the cardinal utility of the second plate a genuine and indispensable role — for to take an average of utilities, one must have utilities that are magnitudes and not merely an order; an expected value cannot be taken over a ranking. Risk is the arena in which cardinal utility, idle for riskless choice, earns its keep. The next plate sets the principle on firm axiomatic ground.

Bernoulli's rule was an insight; John von Neumann and Oskar Morgenstern made it a theorem. They laid down axioms not on preferences over bundles but on preferences over gambles — completeness and transitivity once more, a continuity condition, and a decisive independence axiom, which says that a preference between two gambles should be unchanged if each is combined, in the same way, with some third gamble. From these they proved that a chooser who obeys them behaves exactly as if maximizing the expected value of some utility function, and that this function is unique up to its scale and its zero. The cardinal utility the ordinalists had banished returns here with real content: in the domain of risk, the magnitudes of utility are fixed — up to the choice of scale, as temperature is — by the pattern of choices among gambles.

The shape of that utility function encodes a person's attitude to risk, and the figure draws the central case. Let the utility of wealth be concave — bending as marginal utility diminishes — and take a fair gamble between a lower and a higher wealth. The expected utility of the gamble lies on the chord joining the two outcomes; the utility of the gamble's expected value lies on the curve above it; and because the curve bends away from the chord, the second exceeds the first. The person therefore prefers the sure expected value to the gamble — this is risk aversion. The sure wealth that would yield the same utility as the gamble, the certainty equivalent, falls short of the expected wealth, and the gap between them is the risk premium: the amount of expected wealth a person will sacrifice to be quit of the risk. Aversion to risk is nothing other than the concavity of the curve, made visible as the distance between a chord and an arc.

From this single bend a great deal follows. It explains why we insure — why a person will pay a premium above the expected loss to convert the small chance of a large loss into the certainty of a small one, because the utility rescued in the bad state outweighs the utility spent on the premium. It explains the premium that risky assets must offer to be held at all, the subject of the volume on the premium: a higher expected return is the price of persuading a risk-averse holder to bear the variance. It explains diversification, the hunger for safety, the willingness to pay for the certain over the merely probable. The one fact that the utility of wealth bends — that the first dollars matter more than the last — organizes the entire economics of risk, as diminishing marginal utility organizes the whole of demand. Concavity is to risk what satiation is to wanting: the curvature that explains the conduct.

Maurice Allais built his objection out of the theory's own materials. He set down two pairs of gambles and showed that most people — including expert decision theorists who knew the axioms by heart — choose in a pattern that no expected-utility function can rationalize, a pattern that violates the independence axiom. The mechanism is the pull of certainty: when one option offers a sure outcome, people cling to it in a way that, when the same prizes are offered with that certainty removed, reverses. The choices are deliberate, stable, and all but universal — and flatly inconsistent with the rule that descends from Bernoulli and von Neumann and Morgenstern. As a description of how people actually choose under risk, expected utility is false, and false not at the margins but at the center.

Daniel Ellsberg exhibited a second violation, about uncertainty rather than risk. People sharply distinguish risk, where the odds are known — a coin one knows to be fair — from ambiguity, where the odds are unknown — an urn of wholly unknown composition — and they shun the ambiguous. Offered a bet on the known-fair coin and an equivalent bet on the unknown urn, most prefer the known, and prefer it in a way that proves mutually inconsistent across a set of such bets. Expected utility, and the subjective-probability extension of it that Leonard Savage had built to cover unknown odds, requires that a person assign some definite probability to every event and act upon it. Ambiguity aversion is the refusal to do so — the treatment of "I have no basis for a probability" as itself a reason to avoid the bet, worse than any particular odds one could name. The world's uncertainty, people insist by their choices, is not always a probability.

These were not isolated curiosities but the leading edge of a wave. Daniel Kahneman and Amos Tversky gathered the violations into prospect theory, which sets aside expected utility and describes instead how people in fact choose: they read outcomes as gains and losses from a reference point rather than as levels of final wealth; they feel a loss more keenly than an equal gain, the asymmetry of loss aversion; they overweight the small probabilities and underweight the large; and they are moved by the mere framing of a choice, preferring or rejecting the same prospect according to how it is described. Prospect theory is descriptively far superior to expected utility and deliberately silent on what is rational — it reports how people do choose, not how they ought to. The gap between the two, between the rational ideal and the actual mind, is the whole subject of the volumes on behavior, to which this plate is the threshold.

Bentham's project required adding utility across people. The greatest happiness of the greatest number is a sum — my pleasure joined to yours, pooled into a social total to be made as large as possible — and classical welfare economics inherited the arithmetic. The standard case for redistribution rested on it: a dollar yields more utility to a poor person than to a rich one, by diminishing marginal utility, so moving it from rich to poor raises the total, and the transfer is justified by the sum it increases. Every such argument presupposes that utilities are comparable across persons — that there exists a common scale on which my satisfaction and yours can be laid side by side and the one judged greater or less than the other.

The ordinal revolution destroyed the common scale. If a person's utility is merely their own private ranking of bundles, with no magnitudes and no anchored unit, then there is simply no fact of the matter as to whether my utility exceeds yours. The question is not difficult but meaningless — as empty as asking whether my ordering of the colors is numerically higher than yours. Lionel Robbins pressed exactly this in the nineteen-thirties, and pressed it with positivist severity: interpersonal comparisons of utility, he argued, are not propositions that evidence could settle but value judgments wearing the dress of science, and an economics that means to be a science must put them aside. The argument was widely accepted, and its acceptance cut economics off from the utilitarian summing of happiness that had been, since Bentham, its ethical engine.

Economics learned to say a surprising amount without ever comparing utilities, chiefly through the criterion Pareto had supplied: one outcome is better than another if it makes at least one person better off and no one worse off. This ranks outcomes without ever weighing one person's gain against another's loss — but it ranks very few, for almost every policy that matters helps some and hurts others, and on all of those the criterion is silent. The renunciation of interpersonal comparison purchased scientific respectability at the price of ethical muteness: an economics that can certify a free lunch, where someone gains and no one loses, but cannot pronounce on a trade-off, where one gains and another loses — which is to say, on nearly every question a society actually faces. Whether this restraint is rigor or abdication, and whether some comparison must quietly be let back in for economics to say anything useful about distribution at all, is the burden of the volume on the share.

Grant the renunciation of the last plate; suppose we will not compare the magnitudes of anyone's satisfaction, and have before us only each person's ordinal ranking of the options. Can we at least combine those individual rankings into a single, coherent social ranking — a fair and consistent statement of what the group prefers? This is the problem of social choice, and democratic voting is its everyday instance: given everyone's preferences over the candidates, produce the choice of society. One asks for a rule of aggregation that is fair, sensible, and yields a coherent ordering for the whole.

The trouble surfaces at once, in a paradox known to Condorcet two centuries ago. Let three voters rank three options in rotated orders — the first preferring A then B then C, the second B then C then A, the third C then A then B. Then a majority prefers A to B, a different majority prefers B to C, and a third majority prefers C to A: the social preference runs in a circle, with no option that beats all others, as the figure shows. Majority rule, faced with more than two options, can manufacture a society that prefers A to B to C and back to A — an incoherent ordering, and one open to manipulation by whoever sets the order in which the votes are taken. The simplest and most natural rule of aggregation fails to deliver a coherent social preference at all.

Kenneth Arrow asked whether some cleverer rule might escape the trap — might turn every possible profile of individual rankings into a coherent social ranking while meeting a few minimal demands of fairness and sense. The demands are mild: unanimity, that if everyone prefers A to B then so does society; independence of irrelevant alternatives, that society's ranking of A against B depend only on how individuals rank A against B, and not on their views of some third option; and non-dictatorship, that no single person's preference always prevail regardless of everyone else's. Arrow proved that no rule can satisfy all of these at once — that the only rules guaranteeing a coherent social ordering over all profiles are dictatorships. This is his impossibility theorem, among the deepest results in all the social sciences: the aggregation of individual preferences into a fair and coherent social preference is, in general, not merely difficult but impossible.

The whole apparatus has assumed, silently, that what we choose is what is good for us — that the decision utility revealed in choice is the same as the well-being it is meant to serve. Daniel Kahneman drove the two apart. He revived Bentham's older and more literal notion of utility as felt experience, the pleasure and pain actually undergone, and set it against the decision utility that governs choice. Experienced utility is the quality of life as it is lived, moment to moment; decision utility is the weight an outcome carries when one chooses among prospects. Standard theory simply equates them — the thing we pursue and the thing we feel. Kahneman asked whether they are in fact the same, and found, in study after study, that they are not.

The wedge is driven through memory. We do not choose on the basis of experience directly, for experience is gone the moment it occurs; we choose on the basis of our memory of experience, and memory is a poor and systematically biased recorder. The figure shows the central finding. The remembered quality of an episode is governed not by the sum of the experience over its duration but by two moments alone — its peak and its end, the peak-end rule — and is strikingly insensitive to how long it lasted. An ordeal twice as long but otherwise identical is remembered as no worse; an episode that ends well is remembered fondly though most of it was wretched. The experiencing self lives through every moment in equal measure; the remembering self, keeping only the peak and the close, is the self that does the choosing — and it routinely overrules the self that actually lived the life.

The consequence is unsettling. If the remembering self systematically misjudges the experiencing self, then our choices, made by the rememberer, do not reliably maximize our experienced well-being. We choose the holiday that will photograph well over the one we will enjoy; we prolong an experience for the sake of an ending we will misremember; we pursue what decision utility rates highly and experienced utility fails to deliver. This drives a wedge into the very foundation of welfare economics. If utility-as-choice and utility-as-experience come apart, then revealed preference does not track well-being, and the identification of what people want with what is good for them — the assumption beneath the entire subject, the one the revealed-preference plate had already strained — fails outright. Which utility, then, should policy serve: the choices people make, or the lives they actually experience? The question is open, and it reaches back to the volume on growth, where the same gap haunts the broken link between rising income and reported happiness.

Utility is the want behind value, the end that choice is meant to serve, the quantity that the whole subject tracks and no instrument can read. Across eleven plates the inquiry has circled it: the want and the scale; the ranking and the axioms; the preference read from choice; the wager and the expected and the anomaly; the incomparable, the impossible, the experienced. It remains to ask what the concept is, taken whole, and whether it can carry the weight that an entire science has laid upon it.

Can it be measured? The question has no single answer, and the volume's first arc was the discovery of why. For riskless choice, mere ranking suffices and the cardinal magnitudes are idle — the ordinal turn was right that one need never measure the satisfaction of a loaf to derive the demand for it. But where risk enters, a cardinal utility returns with genuine content, its magnitudes pinned down by choices among gambles and its concavity the very measure of aversion to risk. Utility is ordinal where order is all that is asked of it and cardinal where magnitudes do work; the answer depends entirely on the question, and the long confusion was the failure to see that it must.

Is it something felt, or only something inferred? Here the volume found a standing contest. Bentham meant a felt state, a pleasure or pain genuinely undergone; revealed preference meant only a summary of consistent choice, with no feeling behind it. The operational view bought its rigor by identifying utility with choice — and in the same stroke lost the power to say that anyone had ever chosen badly, for one maximizes one's utility by definition in the act of choosing. The hedonic view restored the felt state, and with it the possibility of error, but at the cost of the measurability the operationalists had prized. Utility hangs suspended between the two: felt but inferred, real but reachable only through the behavior that casts its shadow.

Is it the same as well-being? The subject's foundation is the quiet identification of what we choose with what is good for us, and the experienced-utility plate drove the two apart. The remembering self, which makes our choices, systematically misjudges the experiencing self, which lives our life; we choose for peaks and endings and misremember the rest. If choice and welfare diverge, then revealed preference does not track well-being, and the whole edifice rests upon an assumption that its own findings unsettle. The most basic premise of welfare economics — that the satisfied preference is the served interest — turns out to be not a truth but a hope.

Can it be shared? Bentham's sum required comparing utilities across persons; the ordinal turn made the comparison meaningless; Robbins made the renunciation orthodox; and Arrow proved that even the bare rankings that remain cannot, in general, be coherently aggregated into a social judgment at all. Economics can certify a free lunch, where one gains and none lose, but it cannot adjudicate a trade-off, where one gains and another loses — and the questions of justice that require the weighing of one life against another are precisely the ones it has disqualified itself from answering. The quantity invented to be the measure of all value resists both measurement and addition.

What, then, is known? A great deal, and securely: a rigorous theory of choice — preference, representation, expected utility — that organizes demand, risk, insurance, and the pricing of assets, built upon observable choice and freed of an unmeasurable feeling. And what is not known shadows each achievement. The axioms on which the theory stands are systematically violated, as the anomalies show; the concept cannot be compared across persons or aggregated without contradiction, as the impossibilities show; and it may not even track the well-being it was invented to capture, as the experienced wedge shows. The most fundamental quantity in all of economics is the one the subject holds least securely in its hand.

The Utility is the third of The Quantities, and the deepest of them — the value that the others presuppose. It stands beneath The Margin, whose marginal utility is utility at the edge and whose ordinal turn this volume carried to its end; beneath The Surplus, which measures in money the utility a market delivers to those who trade in it; beneath The Stock, where utility is spread across time and discounting weighs the present want against the future. And it reaches throughout the atlas — to The Premium, where the concavity of utility prices the bearing of risk; to The Choice and The Bias, where the axioms break and prospect theory takes the field; to The Share, where the impossibility of comparison disarms every judgment of distribution. Utility is the concept on which the subjective theory of value finally rests, and the one it can least securely hold.

What is utility? The want that price is built to measure — and a quantity that resists, at every turn, the weight the subject puts upon it: felt but unmeasured, inferred but unobserved, ranked but incomparable, chosen but not certainly good for the chooser, summed only at the cost of contradiction. It is at once indispensable and a ghost. The marginal revolution made value subjective, and this volume has been the reckoning with what that meant — that the foundation of the entire subject is a quantity lodged in a single mind, available to no instrument and to no other mind, which we approach only through the choices it leaves behind like footprints in snow. The quantity, like every quantity this atlas weighs, is a verdict on a hidden contest: between value as felt and value as chosen, between the cardinal and the ordinal, between what we want and what is good for us, between the single mind that ranks and the many minds that cannot be summed. The argument remains open.