The Choice

The model that underlies this entire table can be stated in a single sentence: a person, faced with a set of options the world makes available, chooses the one he most prefers. He is a maximizer — not of money, necessarily, nor of pleasure, but of preference itself, climbing to the highest point of his own ranking that lies within his reach. Everything else in the theory of choice is the working-out of this one idea. The agent has a preference ordering, a ranking of the outcomes he might obtain; he faces a feasible set, the options actually open to him given his means; and he selects from that set the option his ranking places highest. Constrained maximization, the mathematicians call it, and it is the formal skeleton of every economic decision the earlier volumes described.

One word in the model's customary name does it more harm than all its critics, and it must be disarmed before anything else can proceed. The word is rational, and in ordinary usage it carries a freight the theory never intended. To call the economic agent rational summons, in most minds, a particular and unlovely figure: cold, calculating, narrowly self-interested, a creature of spreadsheets who counts every penny and feels nothing he cannot price. Generations of critics have attacked this figure, and they have been right to, for he is a monster. But he is not the rational agent of the theory. He is a caricature, and the theory is innocent of him.

For rationality, in the precise sense the next plates will sharpen, is a property of the form of a person's preferences and not of their content. It says nothing whatever about what a person ought to want. A rational agent may want the comfort of strangers more than his own; he may want the beauty of a symphony, the quiet of a long walk, the welfare of a child, the approval of his god, and want none of these for any gain he could deposit in a bank. He may be a saint or a sensualist, an altruist or a miser, and the theory is indifferent among them. What it asks of him is only that his wants be ordered in a certain coherent way — that they not contradict themselves — and a person of the most generous and uncalculating heart may satisfy that condition as fully as any miser. Rationality is about consistency, not selfishness; about the shape of choice, not its substance.

Disarmed of its caricature, the assumption shows its true reach, which is enormous. The maximizer is not one figure among the many this table has drawn; he is the single figure beneath them all. The firm that minimized its costs was maximizing profit; the consumer who balanced the goods on the shelf was maximizing utility against a budget; the saver weighing the rate, the worker trading leisure for wages, the investor spreading his holdings against risk — each was this same agent, choosing the top of his ranking within his constraint. The theory of the firm, the theory of the household, the theory of the market: all are applications of the one model of choice. To examine the maximizer is therefore not to open a side chamber of the science but to walk into its foundation, and to ask of that foundation the question every later plate presses — whether it bears the weight the whole structure places on it.

If rationality is consistency, the question is what consistency requires, and the answer is given by two axioms so modest in appearance that one wonders they could carry a science. The first is completeness: for any two options the agent might face, he can say which he prefers, or that he is indifferent between them. He is never struck dumb; the ranking has no gaps. The second is transitivity: if he prefers a first option to a second, and the second to a third, then he prefers the first to the third. The ranking has no loops. To these the theory adds, for the sake of its mathematics, the formalities of reflexivity and continuity, but completeness and transitivity are its heart, and between them they say what it is for a set of preferences to hang together. A preference ordering satisfying them is what the theory means, and all it means, by rational.

Transitivity is the more interesting of the two, for it can be defended by an argument that turns it from an arbitrary stipulation into a condition of survival in any world where things can be traded. Suppose a man's preferences form a circle: he prefers an apple to a banana, a banana to a cherry, and — completing the loop — a cherry to an apple. He seems to violate no obvious law of reason; he simply likes things in a ring. But watch what the market does to him. He holds a cherry; since he prefers a banana, he will pay a small sum to exchange the cherry for one; since he prefers an apple to the banana, he will pay again to trade up; and since he prefers, absurdly, the cherry to the apple, he will pay a third time to return to exactly where he began. He is now three payments poorer and holds the same fruit. And nothing stops the wheel from turning again, and again, until he is penniless.

The money pump is more than a curiosity. It shows that transitivity is the price of not being exploited, that a world of exchange punishes the agent whose preferences loop, and that the discipline of the market is therefore a force pressing real preferences toward the very consistency the axiom demands. It shows, too, why a circle of preference is incoherent in a deeper sense than mere vulnerability: a ranking that loops has no top. There is no most-preferred option in a cycle, no best to choose, and so for an intransitive agent the central instruction of the whole theory — choose the best available — has no meaning, because no best exists. Transitivity is what guarantees that the instruction can be obeyed at all.

So the axioms, modest as they look, do indispensable work: they are what make preference into something an agent can maximize. And much follows from them. The next plate shows that a preference ordering obeying these conditions can be captured entirely by a single numerical function — that consistency in ranking is exactly what permits the wants to be written as utility, and the maximizer's climb to be drawn as the ascent of a measurable hill.

A preference ordering that obeys the axioms can be captured, entirely and without remainder, by a single numerical function. Assign to each option a number, larger for the more preferred, and the agent's choosing of the best becomes the maximizing of that number. The number is called utility, and the agent who climbs his preference ranking may now be drawn as climbing a measurable hill, seeking its summit within the fence of his budget. This is the bridge from the bare relation of preference to the calculus of maximization, and it is what lets the spare model of the first plate become a quantitative science.

But the most important thing about this utility is what it is not, and the history of the idea is the history of a great subtraction. To the founders of the felicific tradition — to Bentham above all, and after him Jevons and the early Marshall — utility was a substance: a real quantity of pleasure or satisfaction, present in the mind in measurable amounts, addable across the goods a person consumed and, more boldly, comparable across persons, so that one might in principle sum the happiness of a nation and act to enlarge it. This was cardinal utility, and it carried economics toward a vision of the social good as the greatest aggregate of measurable pleasure. The trouble was that no one could measure it. Pleasure has no units; there is no instrument that reads the satisfaction in a mind, no scale on which my enjoyment of bread may be weighed against yours.

The escape, worked out by Pareto and Fisher and brought to its sharp modern form by Hicks and Allen, was to discover that economics had never needed the substance at all. For everything the theory of demand requires — that the agent rank bundles, and choose the highest-ranked he can afford — only the order of the utility numbers matters, never their magnitudes. To say that one bundle yields utility ten and another utility five is to say only that the first is preferred; it is not to say it is twice as good, for the numbers might as well have been a hundred and one, or any others that preserve the ranking. Utility, on this ordinal view, is not a reading taken from the mind but a convenient index of a preference that is itself the only real thing. The numbers are a notation for the ranking, and any relabeling that keeps the order keeps the meaning. With this the embarrassment of unmeasurable pleasure simply dissolved: economics could speak of utility while disclaiming all knowledge of how much anyone enjoyed anything.

The geometry of the figure is the whole theory in a glance. An indifference curve joins the bundles the agent ranks equally; the family of such curves, nested and rising away from the origin, is a complete picture of his preferences; and his choice is the point where the highest curve he can reach just grazes the line of what he can afford. That grazing — the tangency of budget and indifference curve — is the optimum, the summit of the affordable hill, and it is nothing other than the maximizer of the opening plate rendered in two dimensions. Everything the earlier volumes drew from the consumer, the downward-sloping demand and the response to every change in price and income, can be read off the way this tangency slides as the budget line tilts and shifts. Utility, the ordinal index of a consistent preference, is all the apparatus needs.

This refinement is rightly counted among the triumphs of the field: it gave demand theory a foundation that asked nothing it could not observe and claimed nothing it could not defend. The next plate pushes the same instinct one step further, asking whether even the ranking must be assumed, or whether it too might be read off from the only thing the world ever shows us — the choices people actually make.

The ordinal turn had banished the measuring of pleasure, but it left preference itself still hidden in the mind, an unobservable from which observable choices were supposed to flow. A bolder spirit asked whether even this residue of the invisible could be dispensed with — whether the whole theory might be rebuilt on what alone can be seen, the choices people actually make. The answer, given by the young Samuelson in 1938 and elaborated for a generation after, was the doctrine of revealed preference, and it reversed the direction of the entire enterprise. Where the older theory began with preferences and deduced choices, revealed preference begins with choices and infers preferences. A man's wants are not peered at through introspection; they are read backward from his deeds.

The basic move is disarmingly simple. Suppose a person, able to afford either bundle, chooses A when B lay equally within his means. Then he has revealed a preference for A over B — not stated it, not introspected it, but shown it in the only currency that counts, the currency of action. Preference becomes a summary of choice rather than its hidden cause, and the theory is grounded at last in the observable. The ambition behind the move was the positivist temper of its decade, the same impulse that drove behaviorism in psychology: to purge science of unverifiable talk about inner states and rebuild it on public, repeatable observation. Utility and preference were not to be denied, exactly, but demoted — kept as convenient constructs, summaries of choice behavior, rather than as causes known by introspection.

Consistency, banished from the mind, returns as a condition on the deeds. The Weak Axiom of Revealed Preference requires that a chooser not contradict himself across situations: if he picks A over an affordable B today, he must never pick B over an affordable A tomorrow. A strengthened version forbids the longer cycles, restoring transitivity in the language of choice. These axioms are the empirical heart of the theory, for they are testable: one can watch a household's purchases across changing prices and incomes and ask whether the pattern obeys them. And a deep result, refined by Houthakker and Afriat, closes the circle: choices that satisfy the axioms can always be rationalized by some utility function, as if the chooser had been maximizing one all along. Observable consistency in choosing is exactly equivalent to the existence of a preference ordering behind it. The theory had made itself operational — a science of behavior, testable against the record, owing nothing to the inspection of minds.

As an instrument this is among the most elegant constructions in the field, and the demand theory of the earlier volumes rests on it more securely than on any introspective foundation. But elegance has concealed a slide whose consequences run through everything that follows in this volume. In making preference the summary of choice, revealed preference has made the two identical by definition. What a person prefers simply is what he chooses; there is, now, no daylight between them. The eleventh plate will show how this identification threatens to drain the theory's central claim of all its content. Here it is enough to mark the hinge: the operational triumph that grounded the theory in behavior did so by defining away the distance between what we choose and what we want — and a great deal of human unhappiness lives in exactly that distance.

The choices that matter most are made in the dark. The investment whose return is unknown, the career whose prospects are uncertain, the insurance bought against a misfortune that may never come, the very crossing of a street — each is a choice not among outcomes but among gambles, lotteries over outcomes whose resolution lies with chance. The theory of choice must therefore extend from the certain to the risky, and the question is what the rational agent maximizes when he cannot know which way the world will fall.

The natural first answer is expected value: weigh each money outcome by its probability, add, and choose the gamble with the largest sum. It is the right way to price a wager repeated endlessly, and it is wrong as a theory of choice, as a puzzle posed in Saint Petersburg three centuries ago made plain. Consider a coin tossed until it first shows heads; if that happens on the first toss you are paid two, on the second four, on the third eight, the prize doubling each time. The expected value of this gamble is two times one-half plus four times one-quarter plus eight times one-eighth, and so on without end — a sum of one-after-one forever, which is to say infinite. Expected value therefore counsels you to stake your entire fortune, and more, for the privilege of playing. Yet no sane person would pay more than a trifle. The maximand cannot be expected value, for expected value commands what no one will do.

Bernoulli, who posed the puzzle, also solved it, and his solution is the foundation of all that follows. People do not value money in proportion to its amount; they value it less and less at the margin, the second thousand meaning less than the first, the millionth dollar a rounding error to a man who has millions. Utility is concave in wealth. And what the rational agent maximizes is not expected value but expected utility — each outcome weighed by its probability, but counted in utility rather than dollars. Under a concave utility the doubling prizes of the Petersburg gamble are tamed by the shrinking worth of each, the infinite sum collapses to a modest figure, and the willingness to pay a trifle is explained. Two centuries later von Neumann and Morgenstern, in the same volume that founded the theory of games, gave the idea its rigorous form: from a handful of axioms on the ranking of lotteries they proved that a consistent agent must choose as if maximizing the expected value of some utility function. Expected utility is to choice under risk what ordinal utility is to choice under certainty — the precise content of rationality in the presence of chance.

The figure shows the shape of every such decision: a fork the agent governs, a fork chance governs, and at the tips the outcomes whose utilities, weighed by their probabilities, decide the matter. To choose well under risk is to fold the tree back from its tips, replacing each chance fork with the expected utility behind it, until the agent's own fork can be settled by comparing two numbers. It is a procedure of great power and reach, the engine of the theory of insurance, of portfolios, of every decision the earlier volumes made under a cloud of chance. And it is, as the next plate will show, the soil from which the whole phenomenon of risk aversion grows — for the concavity that tamed the Petersburg gamble is the very same curvature that makes a man pay to be rid of risk.

For all its elegance the theory rests on one axiom more vulnerable than the rest, and its vulnerability has been the most productive fault line in the modern study of choice. The next plates will mine it.

People pay to avoid risk. They buy insurance at a price above the average loss it covers; they spread their wealth across many ventures rather than stake it on one; they prefer, almost universally, a sure sum to a gamble worth the same on average. This aversion is among the most pervasive facts of economic life, and the previous plate has already supplied its cause, though the cause deserves to be drawn. It is the concavity of utility — the diminishing worth of each added dollar — and from that single curvature the whole disposition to shun risk, and the entire industry built upon it, can be derived.

The figure shows the mechanism exactly. Take a gamble that pays a low wealth or a high one on the toss of a fair coin. Its expected wealth is the point midway between them on the floor of the diagram; its expected utility is the point midway between the two utilities, which lies on the straight chord joining the curve's two ends. Because the curve bows upward — because it is concave — the chord runs below it, and so the expected utility of the gamble falls below the utility of its expected wealth. The agent, that is, would rather have the average wealth for certain than face the gamble that yields it on average. That preference, read straight off the geometry, is risk aversion, and it is present for any concave utility whatever, however gentle the bow.

The construction names the two quantities on which the economics of risk turns. The certainty equivalent is the sure wealth that would leave the agent exactly as well off as the gamble — found by carrying the gamble's expected utility back to the curve and reading the wealth beneath it. It lies short of the mean, and the shortfall, the risk premium, is the price the agent will pay to be rid of the risk: the amount by which he will accept less than the average, for certain, rather than face the spread. Here is the whole secret of insurance, the subject of an earlier volume. The risk-averse will pay a premium above their expected loss to hand the risk to someone else; the insurer, pooling many independent risks so that the average is nearly certain, bears it cheaply; and the premium between is the room in which both parties gain. The same curvature explains why the prudent diversify, spreading wealth across uncorrelated ventures to shrink the spread without shrinking the average, as the volume on the stock described; why safe assets are dear and risky ones must promise more to be held; why the whole apparatus of hedging and pooling exists. One bend in a curve, and the architecture of risk-bearing stands explained.

It is an account of real beauty and real reach, and it is also, at the scale where most of our risk-taking actually occurs, demonstrably wrong — a failure that points directly toward the volume to come.

The strains of the axioms were strains of consistency; the strain this plate examines is graver, for it is a strain of possibility. To maximize is to survey the whole field of options, evaluate each by its consequences, and select the best — and for almost any choice that matters, no mind can do it. The options are too many, the consequences too uncertain, the future too branching. A person buying a house cannot inspect every house; a firm setting a price cannot compute the profit-maximizing figure across every contingency; a young man choosing a calling cannot trace the life each would yield. The optimum exists, perhaps, as a point in some vast space, but it is not findable, and a theory that instructs the agent to find it instructs him to do what cannot be done.

It was Herbert Simon, more than anyone, who turned this obvious difficulty into a research program and gave it a name. Rationality, he argued, is bounded — hemmed in on every side by the limits of the available information, the capacity of the mind to process it, and the time there is to decide. The agent of the textbook, who optimizes against the whole world in an instant and at no cost, is a fiction; the real agent thinks with a small and costly instrument under pressure of the clock. And the real agent, Simon observed, does not maximize. He satisfices: he forms an aspiration, a level of outcome that would be good enough, and searches only until he finds an option that meets it, and then he stops. He does not find the best house; he finds a house he can live in and buys it. He does not compute the optimal price; he marks his cost up by a customary margin. The maximizer is replaced by a creature who copes.

This is not a confession of irrationality but a redescription of what rationality, for a finite being, can mean. Simon drew the distinction sharply: substantive rationality is choosing the objectively best outcome, which is what the theory assumes and what no bounded agent achieves; procedural rationality is deciding by sensible procedures given one's limits, which is what a wise bounded agent does. The shift of attention is from the optimality of the result to the intelligence of the process. And the procedures bounded agents use — the rules of thumb, the heuristics, the customary markups and the standing routines — are not the marks of stupidity they are sometimes taken for. They are often beautifully adapted to their task, fast and frugal where full computation would be slow and ruinous, and a later school under Gigerenzer has shown that simple heuristics can outperform elaborate optimization in the uncertain worlds for which they evolved. Intelligence, for a bounded creature, lies in matching the procedure to the environment, not in a computation no environment rewards.

The consequences reach back into the earlier volumes. The firm of the orthodox theory maximized a profit it could not compute; the firm that Simon's followers actually studied — the behavioral firm of Cyert and March — followed routines, met targets, satisficed, and changed its rules only when its aspirations went unmet. Whole markets, peopled by such satisficers, need not behave as markets of maximizers would. There is a famous defense, owed to Friedman, that the falsity of the assumption need not matter: competition will weed out the firms that stray too far from the optimum, so the survivors will behave as if they maximized, whatever their actual procedures. The defense has force where selection is sharp and the optimum well-defined. It has far less where competition is slack, the environment shifting, and the optimum itself unknowable — which is to say, across much of economic life. The maximizing firm may be a fair prediction of the survivors in a brutal market and a poor description of the agents in a gentle one.

The theory assumes that a preference exists before the choice, fixed and waiting to be consulted, and that the choice merely reads it off. The third and perhaps deepest seam is the discovery that this is frequently false — that the preference is not consulted but constructed, assembled in the moment of choosing under the heavy influence of how the choice happens to be presented. Change the presentation and you change the preference, which means the preference was never sitting there to begin with. The phenomenon is called framing, and once seen it cannot be unseen.

The canonical demonstration, from Tversky and Kahneman, concerns a public-health choice between two programs to combat a disease expected to kill six hundred. Described in lives saved — one program saves two hundred for certain, the other offers a one-third chance of saving all six hundred and a two-thirds chance of saving none — most people take the certain rescue of two hundred. Described in lives lost — one program lets four hundred die for certain, the other offers a one-third chance that none die and a two-thirds chance that all six hundred die — most people reject the certain loss and gamble. But the two descriptions are the same programs in the same numbers: two hundred saved is four hundred lost, exactly. Nothing has changed but the words, and the choice has reversed. People are risk-averse when outcomes are framed as gains and risk-seeking when the identical outcomes are framed as losses, and the frame is doing the choosing.

This violates a principle the theory cannot do without, sometimes called invariance or extensionality: that a rational preference depends only on what the outcomes actually are, never on how they are described. The same final state, however worded, should command the same choice; a mind that flips with the phrasing is responding to something that, by the theory's lights, is not there. And the source of the flipping, Kahneman and Tversky saw, is that people do not evaluate outcomes as final states at all. They evaluate them as changes — as gains or losses measured from a reference point — and the reference point is exactly what the framing sets.

The damage to the theory is structural, not marginal. Orthodox choice is defined over final states — total wealth, total consumption, the full description of where one ends up — and it presumes that the ranking of those states is stable. Reference-dependence relocates choice to the realm of changes, gains and losses reckoned from a baseline, and makes the baseline movable by a word. A preference ordering anchored to a reference that the framing can shift is not the fixed ordering the theory requires; it is a family of orderings, one per frame, with no privileged member among them. The complete and stable ranking of the opening plates dissolves into something contextual and constructed, and the maximizer is revealed, at least often, to be composing his preferences as he goes. The full apparatus that captures this — the value function bent differently for gains and losses, the overweighting of small probabilities — belongs to the volume on the bias, which is built upon it. Here the seam itself is the lesson: the agent does not have the preferences the theory says he consults.

And if preferences are written by the frame, the question of who holds the pen becomes the most pressing in all of economics.

A man resolves on Sunday to begin his diet on Monday, and on Monday he eats the cake. He intends, all year, to start saving next month, and next month he spends. He means to write the chapter tomorrow, and tomorrow he means it again. These are not failures of knowledge — he knows the cake is unwise, the saving prudent, the chapter due — but failures of a stranger kind: the self who plans and the self who acts want different things, and the planner loses. The theory of choice, which knows only one agent with one stable set of wants, has no room for this most familiar of human experiences, and the fourth seam is the place where time pulls the single agent apart.

Choice across time requires discounting, since a reward later is worth less than the same reward now, and the rational form of discounting is exponential: a constant rate per period of delay, the future fading at a steady proportional pace. Its cardinal virtue, established by Samuelson and explored in the volume on the rate, is consistency. Under exponential discounting the ranking of two future dates never changes as the present advances toward them; what you prefer for the distant future, you will still prefer when the future arrives. The exponential agent makes plans he will keep.

People do not discount exponentially. They discount hyperbolically, as the figure shows — steeply across the near delays and gently across the far ones, so that the value of a reward collapses in the first moments of waiting and then declines only slowly thereafter. This shape, documented by Strotz, Ainslie, and Laibson, is present bias: an outsized pull toward whatever is immediate. And its consequence is precisely the Sunday-to-Monday reversal. Seen from a distance, the larger-later reward dominates the smaller-sooner — health next year beats cake next year, and you sincerely plan accordingly. But as the moment of the cake approaches, the steep near-segment of the hyperbola lifts its present value above the receding health, the ranking flips, and you eat. Your preferences reverse as time passes, not because you learned anything, but because the curve is bent. You are, in effect, a succession of selves with conflicting wants — a farsighted planner who legislates for the future and a myopic doer who governs the present — and the two are at war, with self-control the name of the planner's struggle to bind the doer.

The agent who foresees his own inconsistency has one recourse, and it is among the oldest devices of prudence: he binds his future self in advance. Ulysses, who had himself lashed to the mast so that he might hear the Sirens and yet not steer toward them, is the patron of every commitment device since — the savings club one cannot raid, the gym fee paid in advance, the tobacco locked away, the deadline announced to those who will shame its breach. The volume on the contract met this figure under the name of the mast, and here is its psychological root: a planner, knowing the doer will betray him, reaches forward to remove the doer's options before the doer can exercise them. That such devices are wanted at all, that people pay to restrict their own future freedom, is something the theory of a single coherent agent cannot explain, for why would a unified maximizer ever pay to have fewer choices? The demand for self-binding is the signature of a self at war with itself.

And where there is a self at war, there is a market ready to take a side.

The framing of the eighth plate moved preferences within the moment of choice; this plate moves them across a life. For the theory makes one assumption about preferences so basic that it is rarely stated: that they are given — brought to the market ready-made from somewhere outside it, fixed primitives that the economy takes as data and never questions. Stigler and Becker made the assumption a motto, declaring that tastes are not to be disputed, that one should treat them as stable and shared and explain behavior by prices and incomes rather than by changes in what people want. The motto has real methodological discipline behind it. But it rests on a fiction, for tastes are not given. They are made — and frequently made by the very processes the theory treats them as prior to.

The most pointed version of the charge is Galbraith's, and it strikes at the heart of the case for production. An economy, the orthodox story runs, is to be praised for satisfying wants: the more wants met, the more welfare produced. But a vast portion of a modern economy's effort goes not into satisfying wants that exist but into manufacturing wants that do not — the whole apparatus of advertising and marketing exists for no other purpose. And if the wants are produced by the same machine that then satisfies them, the praise collapses into circularity. Galbraith called this the dependence effect: the wants depend on the process that fills them. To create an itch in order to sell the scratch is no service to the scratched; it is an industry that produces its own demand and congratulates itself on meeting it. The satisfaction is real only if the want was, and the want, increasingly, is the product.

Beneath the engineered wants lie the inherited ones. What a person desires is in large part what his upbringing, his class, his nation, and his age have taught him to desire — the food that tastes right, the success worth wanting, the life that counts as a life — and the volume on the norm traced how deeply the unwritten rules of a community shape the wants of those raised inside it. Preferences are social products before they are individual data. And there is a darker mechanism still, which Jon Elster anatomized under the name of adaptive preference and the older fable called sour grapes. People adjust their wants to their possibilities: the fox who cannot reach the grapes concludes that they are sour; the person denied some good comes, in time, genuinely not to want it; the deprived scale their desires down to the meager range of the attainable, and call the result contentment. This is ruinous for any claim that satisfying preferences measures welfare, because it means the preferences of the deprived have been bent by the deprivation itself. Meet the diminished wants of someone whose wants were diminished by his want, and you have produced not welfare but the quiet of resignation — the contented slave, whose satisfaction indicts the measure that records it as well-being.

Even the gentlest case tells against the assumption. Many of the things we choose are goods whose worth we cannot know until we have had them — the unfamiliar music, the acquired taste, the calling one cannot judge from outside. For these the informed preference the theory needs does not exist before the choice; it is formed by the consuming, not consulted prior to it, and the agent learns what he wants by getting it. Preferences are an output of experience as much as an input to choice. From every side, then — engineered, inherited, adaptive, learned — the same conclusion presses: the wants the theory treats as the exogenous foundation of everything are themselves endogenous, shaped within the economic process they are supposed to stand outside. And a foundation that is produced by the building it supports is no foundation at all.

The seams so far have been empirical — places where the theory's predictions fail against what people do. The last and most unsettling seam is logical, and it concerns whether the theory makes a prediction at all. Its central claim is that people choose what is best for them. Press on the word best and the trouble appears. How is the best identified? By the doctrine of revealed preference, the best for a person is what he prefers, and what he prefers is read off what he chooses. So the best is defined as the chosen. Now read the central claim back with that definition in place: people choose what is best for them becomes people choose what they choose. The grand proposition on which a science was built turns out, on inspection, to be a circle — true necessarily, because it could not be false, and saying, for that very reason, nothing.

The emptiness is the price of the operational triumph. By defining preference as the summary of choice, the fourth plate made the proposition that agents maximize preference true by construction — and therefore vacuous, for any behavior whatsoever can be exhibited as the maximizing of some preference, simply by reading the preference off the behavior to be explained. A man who buys the dearer good and a man who buys the cheaper are each maximizing; a man who is consistent and a man who reverses himself can each be fitted with a preference that rationalizes what he did, if the preference is allowed to be whatever the deed requires. A theory able to accommodate every possible observation has thereby forbidden none, and a theory that forbids nothing predicts nothing. This is the hollow at the center, and the rhetoric of rational choice — its talk of a great empirical discovery that people pursue their interests — papers over it.

The hollow is widened by an equivocation the theory practices whenever it is pressed. Attacked as a false description of how people actually decide — by the framing, the present bias, the manufactured wants of the preceding plates — its defenders retreat to higher ground: the model was never meant as description, they say, only as a normative benchmark, an account of how a coherent agent ought to choose. Attacked as an empty tautology that describes nothing, they advance again: it is a substantive empirical theory with real predictive content. It cannot be both, yet it is defended as both, sliding between the empirical and the definitional, the descriptive and the normative, so that every objection meets whichever face deflects it. The maneuver keeps the theory safe from refutation, but safety from refutation is exactly the disease, not the cure.

The theory is rescued from total vacuity by one thing, and it is worth being exact about what. The consistency axioms have genuine empirical bite: completeness and transitivity, and their counterparts in revealed choice, do rule certain patterns out — the cycles, the reversals across frames, the intransitivities. So the claim that an agent maximizes a stable and consistent preference is not empty; it is testable, and it can fail, which is precisely why the laboratory could find it failing. The empirical content of the whole theory lives in the consistency it demands, together with a few substantive predictions like the downward slope of demand that follow from it. But this rescue is also the theory's deepest embarrassment, for it locates the content exactly where the evidence is most damning. The theory means something only insofar as it requires consistency; it requires consistency only where the behavioral findings show people to be inconsistent. And so the fork on which every seam of this volume converges: keep the consistency that gives the theory content, and the evidence falsifies it; drop the consistency to fit the evidence, and the theory dissolves into the tautology that people do what they do. False, or empty. The theory of rational choice must be one or the other, and the art of using it well is knowing, in each application, which it has become.

The volume began with a single sentence — that a person chooses the option he most prefers — and built from it an apparatus of extraordinary reach. We watched the spare relation of preference acquire its axioms of completeness and transitivity, enforced by the money pump; become a utility that ranks without measuring, the ordinal index that freed economics from weighing pleasure; ground itself in the choices it could observe, by the doctrine of revealed preference; and extend into the dark of uncertainty through expected utility, from whose concave curve the whole architecture of risk-bearing — insurance, the premium, diversification — fell out almost for free. This is no small construction. It is the engine beneath every price the earlier families drew, the common form of the firm and the household and the market, general enough that Becker could carry it into crime and marriage and politics, rigorous enough to yield predictions a science can test, and serviceable at once as a description of how people choose and a standard for how a coherent agent should. Whatever its troubles, a model that does all this has earned its central place, and no honest accounting can begin by despising it.

Yet the seams are real, and they are not a scatter of isolated defects to be patched one by one. The axioms that define rationality are violated in regular ways — by the Allais and Ellsberg paradoxes, by the incommensurable goods that completeness cannot rank. The maximization the theory demands exceeds any mind, and Simon's bounded agent satisfices where the model says he optimizes. The stable preferences the theory consults are often constructed in the moment, swung by a frame and anchored to a reference point that whoever describes the choice can move. They reverse across time, the farsighted planner forever betrayed by the myopic doer, so that people pay to bind their own future hands. And they are manufactured across a life — by the advertiser who produces the want he then fills, by the culture that teaches what to desire, by the deprivation that bends desire down to fit. Beneath all of these lies the logical seam: that the central claim, with the best defined as the chosen, threatens to collapse into the empty proposition that people do what they do.

These seams converge, and where they meet is the fork on which the whole volume has come to rest. The empirical content of rational choice lives in the consistency it requires — in the axioms that forbid cycles and reversals, the conditions that make preference something an agent can be said to maximize at all. But it is exactly that consistency the behavioral findings deny, for framing and present bias and the rest are nothing other than the inconsistencies the axioms rule out. So the theory faces two roads and no third. Keep the consistency that gives it teeth, and the evidence falsifies it. Loosen the consistency to absorb the evidence, and it dissolves into the tautology that explains everything and forecasts nothing. False, or empty: the art of using the model well is, in each application, to know which of the two it has quietly become.

The conclusion this volume reaches is therefore not a verdict but a discipline, a refusal of both the verdicts that beckon. To discard the model because it is false would be folly: it remains the indispensable first approximation, the frame within which even its sharpest critics state their corrections, the baseline departures must be measured against, and across enormous stretches of behavior — where stakes are high, choices repeated, and competition sharp — it predicts well enough to be the right tool. But to believe the model, to take it for a literal account of the human chooser, would be a deeper folly, for the seams show plainly that it is an idealization, a deliberate and fruitful simplification whose worth lies precisely in its distance from the messy original. The wisdom is to hold both at once: to use the maximizing agent as the powerful fiction he is, and to keep always in view the catalogue of places where the fiction parts from the fact. A model one neither worships nor discards, but wields with a map of its own failures in hand — that is the whole of what this volume has to teach about how to think with rational choice.

And it is fitting that this should be the third plate of the last family, for the chooser examined here is the figure on whom that family turns. He is the agent The Game set against other agents like himself, each maximizing into the others' maximizing until their choices locked into equilibrium. He is the agent The Norm watched submit to the unwritten rule, choosing not always the optimum but the done thing, the expected thing, the thing the community would not punish. He is the agent whose systematic and predictable errors — the framing, the anchoring, the loss aversion, the present bias glimpsed here only in outline — the volume on The Bias exists to catalogue in full. Rational choice is the model; the game is its extension into strategy; the norm is its rival and supplement; the bias is its critique. To have laid the model bare, in its power and its strain alike, is to have set the foundation on which the other three Behaviors are built, and to have shown why a science of behavior needs all four.

The stakes of the inquiry run far past method, into the whole normative edifice the science has raised. For the most consequential conclusions of economics — that the consumer is sovereign, his wants the authoritative measure of his good; that competitive markets are efficient, allocating resources to their highest valued uses; that what people freely choose is, by that fact, what serves them — rest entirely on the rationality of the chooser. Let that rationality be the idealization this volume has found it to be, and the conclusions lose their footing in the same motion. Every seam was also a crack in consumer sovereignty: if preferences are framed, the market that frames them is not serving wants but writing them; if they are present-biased, the firm is not meeting a stable demand but choosing, for profit, the myopic self over the prudent one; if they are manufactured, the economy is congratulating itself for satisfying the appetites it manufactured. The optimality of markets is only as secure as the rationality of the agents who populate them, and the agent, examined honestly, is not quite rational enough to bear the weight that doctrine places on him. This is not a brief against markets, which remain the most powerful instrument of coordination yet found; it is a refusal to let their defense rest on a psychology the evidence will not support.

There remains the question the model cannot answer in its own terms, the one toward which every plate has tended. The theory takes the best for a person to be what he chooses, and what he chooses to reveal what he prefers, and his preference to define his good. But choosing is not the same as flourishing, and the gap between them is where this volume must finally leave its reader. The smoker chooses the cigarette and is not thereby served; the adapted poor choose the narrow life their poverty allowed them to want; the present-biased self chooses against the self that will inherit the morning; the manipulated consumer chooses what he was arranged to choose. In each, what is chosen and what is good have come apart, and the model, which knows only the chosen, cannot so much as see the difference — for it defined the good as the chosen at the outset, and a theory cannot find what its definitions have hidden. The volume on utility began to ask what well-being is, beneath the preferences that are its rough and ready proxy; the volume on the share asked whose well-being a just economy should weigh. This volume has shown only that the foundational identification — that a person's good is the satisfaction of whatever he happens to prefer — is too thin to bear examination, and that beneath the elegant science of choice lies an older and harder question it was built to avoid: not what a person wants, nor what he chooses, but what is genuinely good for him, and whether the two were ever the same thing. The argument remains open.