The Bias

A bias, in the sense this volume gives the word, is not simply a mistake. People have always made mistakes, and the rational model never claimed they did not; a theory of how a coherent agent ought to choose was never refuted by the observation that flesh-and-blood agents sometimes choose badly. The claim that founds behavioral economics is sharper and stranger than that people err. It is that they err systematically — that the departures from the model are not a random scatter of slips but a patterned tilt, leaning the same direction across persons and occasions, predictable in advance by anyone who knows the pattern. A bias is an error with a shape.

Everything turns on the distinction between error that is random and error that is systematic, and it is worth being slow about it, because the whole standing of the field depends on which kind human error is. Suppose people's mistakes were random — scattered evenly around the right answer, some too high and some too low, with no prevailing lean. Then the orthodox theory would have little to fear, for random errors cancel. Average across a crowd of noisy choosers and the noise washes out, leaving the rational answer behind; the market price still reflects the wisdom of the many, the typical consumer still behaves as the model says, and the rationality of the aggregate survives the irrationality of its parts. This is the great defense the older economics raised against every report of human folly: it does not matter, the errors cancel, the model holds in the large. The defense is sound — but only for noise.

Against systematic error it collapses entirely. When the mistakes all lean one way — when nearly everyone overweights the dramatic, feels a loss more keenly than the equal gain, grasps at the sooner reward — the errors do not cancel; they reinforce. The average is not pulled back to the rational answer but dragged away from it in the direction everyone leans. The bias survives aggregation, persists where noise would dissipate, moves prices that noise would leave alone, and stands waiting to be exploited by anyone who sees it. To show that human error is systematic is therefore to show that it does not wash out — that it reaches past the individual into the market and the aggregate, the very places the older economics had declared it could not reach. This is why the catalogue assembled in these plates is not a footnote about human frailty but a claim upon the foundations.

The discovery was long resisted, for economics had built its rigor on the rational agent and did not surrender him gladly. It took the patient accumulation of evidence — the heuristics and biases mapped by Kahneman and Tversky through the nineteen-seventies, the anomalies pressed by Thaler through the eighties, a generation of experiments each showing the same lean in the same place — to force the rational chooser to share his table with the actual mind. The recognition came slowly and then, with two Nobel prizes a decade and a half apart, decisively. What had been a quarrel at the edge of the discipline became, in the end, the discipline's overdue encounter with the psychology it had spent a century abstracting away. This volume stands to the volume on the choice as anatomy to architecture: where that volume drew the model and conceded, at its seams, that people slip from it, this one studies the slips themselves, maps their direction, and offers in prospect theory a positive account of the mind that makes them.

Before the biases can be catalogued, the machine that makes them must be described, and the description that has organized the field divides the mind into two ways of working. The first is fast: automatic, effortless, always running, the source of the impression that arrives before you have decided to form it. It recognizes a face, completes the phrase bread and, reads anger in a voice, knows that two and two are four, and supplies, without any sense of labor, the great majority of the judgments a person makes in a day. The second is slow: deliberate, effortful, the seat of conscious reasoning and self-control, the faculty that multiplies seventeen by twenty-four, checks a syllogism, or holds the tongue. Kahneman named them, drably and deliberately, System One and System Two, and the drabness was a kindness, for any livelier name would have tempted us to think we knew them.

The relation between the two is the key to everything that follows, and it is not a partnership of equals. The fast system runs the show; the slow system, which fancies itself in charge, is in truth a lazy auditor that mostly rubber-stamps what the fast system hands it. Reasoning is effortful and the slow system conserves its effort, engaging fully only when the fast system stumbles into something it flags as hard or wrong. Most of the time it does not stumble, and most of the time, therefore, the deliberate mind simply endorses the intuition the automatic mind produced, adding a veneer of reasons after the fact. We are, far more than we like to believe, creatures of the first system narrating in the voice of the second.

And the first system reaches its verdicts by a method that is the engine of every bias in this volume. Confronted with a hard question — how probable is this, how frequent is that, how much is this worth — it does not labor toward the hard answer. It quietly substitutes an easier question whose answer is ready to hand, and reports that instead. Asked how likely an event is, it answers how easily an example springs to mind. Asked how probable it is that a person belongs to a category, it answers how much the person resembles the category's type. This is attribute substitution, the swapping of a hard question for a tractable one, and it is ingenious, for the easy answer usually does correlate with the hard one — things that come readily to mind usually are common, things that resemble a type usually do belong to it. The heuristic works. It is fast where deliberation would be slow, frugal where calculation would be costly, and right often enough to have been worth evolving.

The bias is the shadow the heuristic casts. For the easy question's answer correlates with the hard one's only usually, not always, and where the correlation breaks the substitution misfires — in exactly the same direction, for everyone, every time. The vivid but rare event comes easily to mind and is judged common; the person who resembles a type but belongs to a tiny minority is assigned to the type against the odds. The biases to come are therefore not a miscellaneous list of human defects to be memorized one by one. They are the predictable failure-modes of a small repertoire of heuristics, and to understand the heuristic is to know in advance where, and in which direction, the error will fall. That is the order beneath the catalogue, and it is what makes the catalogue a science rather than an inventory.

The first of the great heuristics answers the question how likely? with the question how easily can I recall an instance? This is availability, and like every heuristic it begins as good sense. Things that happen often do, as a rule, leave more traces in memory; the common event is the one you have met before, and so the ease with which examples spring to mind is a serviceable gauge of how frequent a thing really is. Asked whether dogs or dingoes are the more common, you need run no census — instances of dogs flood in, instances of dingoes do not, and the flood gives you the answer. Most of the time, availability and frequency march together, and the shortcut delivers.

They part company whenever something other than frequency governs what comes to mind, and many things do. Vividness governs it: a death by fire is more memorable than a death by diabetes. Recency governs it: the crash reported this morning is fresher than the steady toll of the road. Emotion governs it, and above all the news governs it, for the news is in the business of reporting the unusual — a plane that lands safely is not a story, a plane that falls is — and a mind fed on the unusual comes to believe the unusual is usual. Where these forces inflate the availability of an event beyond its true frequency, the heuristic misfires in a fixed direction: the dramatic, the recent, the reported, and the frightening are judged more common than they are, and the mundane, the silent, the unreported, and the slow are judged rarer. We do not misjudge risk at random. We misjudge it toward the lurid.

The consequences are neither small nor confined to the laboratory. We dread the terrorist and ignore the staircase, fear the shark and not the sun, buy the lottery ticket whose jackpot we can vividly imagine and neglect the retirement we cannot. Whole public policies follow the same lean, lavishing attention and money on the salient catastrophe while the statistical one, which kills far more and photographs far worse, goes unaddressed. And the modern information environment pours fuel on the fire, for a press and a feed that select relentlessly for the rare and the shocking are availability machines, manufacturing the impression that the world is composed of the very events that are, by construction, the least representative of it. The citizen who watches the most news is not the best informed about the frequencies of life but, in a precise and measurable way, the most misled about them.

Yet before the mind is convicted, the heuristic deserves its defense, which is the defense the closing plates will generalize.

The second great heuristic answers the question how probable is it that this belongs to that? with the question how much does this resemble a typical that? This is representativeness, judgment by similarity to a type. Shown a quiet, orderly, detail-loving man and asked whether he is more likely a librarian or a salesman, the mind does not reach for the numbers; it measures him against its stereotypes and, finding him a better match for the librarian, answers accordingly. As ever the shortcut has a sound core, for things that resemble a type usually do belong to it, and similarity is a fair first guess at category. As ever it fails where the resemblance and the probability come apart — and they come apart in ways that are precise, repeatable, and instructive.

They come apart, first, because resemblance is blind to the base rate — to how many of each kind there are to begin with. The orderly man resembles the librarian, but librarians are few and salesmen are many, and once the sheer preponderance of salesmen is weighed, even a bookish man is the likelier a salesman. Representativeness ignores this; the stereotype fills the mind and crowds the numbers out, so that people will confidently sort a person into a tiny category he merely resembles, against a base rate that should have overwhelmed the resemblance. And they come apart, most famously, in the puzzle of Linda. Linda is described as a philosophy student, outspoken on social justice, once active against nuclear arms. Is she more likely a bank teller, or a bank teller who is active in the feminist movement? The overwhelming majority choose the second — for the feminist detail resembles the Linda they were shown, while a bare bank teller does not. And the overwhelming majority are, demonstrably, wrong.

The same misreading of resemblance fathers a whole brood of errors. It produces the gambler's fallacy, the conviction that after a long run of red the black is somehow due — for a mixed sequence looks more representative of chance than a monotonous one, though the wheel has no memory and each spin stands alone. It produces an insensitivity to sample size, by which a result from three cases is trusted as readily as a result from three thousand, since both equally resemble the truth they are meant to estimate, though only the larger sample has earned the trust. It produces the seeing of patterns in noise, the streaks and hot hands and meaningful coincidences with which the mind, demanding that randomness look random, overwrites the genuine article. In each, a judgment that should rest on the structure of probability rests instead on a feeling of similarity, and the feeling misleads in a direction one can predict.

The conjunction fallacy is among the most cited results in all of social science, and it is also among the most contested — a contest the honest reader must weigh before convicting the mind.

The third heuristic concerns not likelihood but magnitude — the estimation of an unknown quantity, a length, a price, a count — and it works by starting somewhere and moving. The mind takes an initial value, an anchor, and adjusts away from it toward what it judges the truth to be. The adjustment, however, is reliably insufficient. People stop too soon, at the near edge of the range they find plausible rather than its center, and the final estimate is left clinging to the anchor it set out from. To anchor and adjust is sensible; to adjust too little is the bias, and the bias is large, stubborn, and almost impossible to will away.

What lifts anchoring from a mild tendency to a genuine astonishment is that the anchor need carry no information whatever, and can be known by the person to carry none, and will drag the estimate all the same. In the experiment that fixed the phenomenon, Tversky and Kahneman sat their subjects before a wheel of fortune, secretly rigged to stop at either ten or sixty-five, and had each watch it spin before asking an unrelated question: what percentage of the nations in the United Nations are African? Those who had watched the wheel land on sixty-five guessed, on average, near forty-five percent; those who had watched it land on ten guessed near twenty-five. A number everyone in the room had just seen produced by a spinning wheel — a number with no more bearing on African geography than the weather — moved the answer by twenty points. The subjects were not fools and did not believe the wheel knew anything. The anchor worked beneath belief, where reasons do not reach.

Once the eye is trained, the drag is everywhere. Ask whether Gandhi died before or after the age of a hundred and forty and people guess his true age higher than if you had named nine; the absurd comparison is an anchor still. Roll dice in view of a judge and the sentence he hands down tilts toward the number the dice showed, though he would indignantly deny the dice had any part in justice. A house sells nearer its listing price than its worth, because the listing anchors the appraisal of the very professionals paid to see past it. Invited to bid on a bottle of wine after writing down the last two digits of an identification number, people who wrote high digits bid more than those who wrote low — a finding so improbable it had to be repeated before it was believed. The mind, asked for a magnitude, fastens on whatever magnitude is nearest to hand and cannot let go.

The mechanism is twofold. The deliberate mind does adjust, but adjustment is effortful, and the lazy auditor halts at the first answer it can call acceptable, which lies on the anchor's side of the truth. And beneath the adjusting, the anchor quietly rigs the evidence: entertaining the hypothesis that the figure might be sixty-five summons to mind the facts that would fit sixty-five, so that the very act of considering the anchor stacks the deck in its favor. Whatever the machinery, the consequence for the world is plain and is not lost on those who profit by it. The opening offer anchors the whole negotiation, which is why the party who names the first number, even an outrageous one, so often does better than the party who waits. The struck-through former price anchors the sale price beneath it, turning an ordinary tag into a bargain by the mere proximity of a larger number. The most expensive dish on the menu sells few portions and earns its place by making the second-most-expensive look reasonable. In a thousand transactions a day, the first number on the table quietly governs the last.

The heuristics so far have governed judgment — the estimation of how likely, how frequent, how large. The plate turns now to choice, and to the theory that did for the description of risky choice what the heuristics did for the description of judgment: it replaced an elegant account of how a rational agent ought to choose with a homelier account of how people actually do. This is prospect theory, set down by Kahneman and Tversky in nineteen seventy-nine, and it is the apparatus the volume on the choice named and deliberately withheld, promising it here. It rests on three departures from the expected-utility model, and the first is the foundation on which the other two stand.

The foundation is that people do not evaluate outcomes as final states of wealth, but as gains and losses from a reference point. The expected-utility agent, offered a prospect, asked what total fortune it would leave him holding and valued that; the actual chooser asks whether the prospect is a gain or a loss against where he stands now, and values the change. He does not think I shall possess one million and forty thousand; he thinks I shall be forty thousand up, or ten thousand down. The carrier of value is the movement, not the position — and this single shift, from levels to changes, is the seam the previous volume exposed under the name of framing, now load-bearing. For if value lives in changes from a reference, then everything depends on where the reference is set, and the same objective outcome can be a gain or a loss according to the point it is measured from.

The second departure is the asymmetry that gives the plate its name. The curve that maps gains and losses into value is not symmetric about the reference: it is markedly steeper on the side of losses than on the side of gains. A loss of a hundred wounds more than a gain of a hundred pleases, by a factor that the experiments place near two — losses loom roughly twice as large as equivalent gains. At the reference point the curve does not pass smoothly through; it bends, sharply, kinked downward, so that the first step into loss costs more value than the first step into gain confers. This is loss aversion, and it is the most consequential single fact behavioral economics has established, for from it a remarkable range of conduct follows.

The third departure completes the curve's shape: sensitivity diminishes as one moves away from the reference in either direction. The value function is concave over gains, as the earlier plate on risk aversion already found — each further pound of winning adds a little less delight — and, in the discovery that surprised even its authors, convex over losses, each further pound of loss adding a little less pain. The consequence is strange and verifiable. Because the gain-arm is concave, people are risk-averse in gains, preferring a sure win to a gamble of equal expectation; but because the loss-arm is convex, they turn risk-seeking in losses, preferring the gamble to a sure loss, throwing good money after bad in the hope of clawing back to the reference. The same person who will not risk a certain gain will eagerly risk a certain loss — and from this curvature, joined to the weighting of probability that the next plate supplies, falls the celebrated fourfold pattern: cautious with likely gains and unlikely losses, bold with unlikely gains and likely losses. One shape of one curve organizes the buying of insurance and the buying of lottery tickets, the holding of losers and the selling of winners, the whole tangle that expected utility had declared incoherent.

Prospect theory matters beyond any single prediction, for it is the nearest thing the field possesses to a positive theory — not a bias to be added to a list but a structured replacement for the utility function, descriptively faithful and astonishingly general. It is the standing answer to the charge that behavioral economics only tears down. And yet it carries, at its very foundation, a difficulty it has never fully resolved.

The value function told how outcomes are felt; it remains to tell how probabilities are reckoned, and here prospect theory makes its second break with the orthodox account. The expected-utility agent multiplied the value of each outcome by its true probability and summed: a ten-percent chance counted for one-tenth. The actual chooser, prospect theory holds, does not use the probability but a distorted image of it — a decision weight, got by passing the probability through a function that bends it out of true. He does not weight a one-percent chance by one percent. He weights it by something larger, and a ninety-percent chance by something smaller, and the pattern of these distortions is as regular as the value function's kink.

The pattern is an inverse curve, an elongated S laid against the diagonal of honest weighting. At the low end, small probabilities are overweighted: the slim chance looms larger in the mind than in the world, so that a one-in-a-thousand prospect is felt as though it were one in a few hundred. At the high end, moderate and large probabilities are underweighted: the strong chance is given less than its due, the near-certainty discounted. Between, the curve crosses the diagonal once, near a third, where for a moment weight and probability agree. And at the very ends the curve does something the diagonal cannot, leaping to meet certainty and impossibility — for the step from a possibility to a sure thing, or from impossibility to a bare chance, carries a weight out of all proportion to the sliver of probability it spans.

The reward for this complication is that behavior long thought incoherent falls into order. Consider the lottery and the insurance policy, which the older theory could not house under one roof: a single person who buys both seems at once a lover of risk and a hater of it, risk-seeking for the jackpot and risk-averse for the catastrophe. Prospect theory dissolves the contradiction at a stroke. Both the jackpot and the catastrophe are rare; both their probabilities are overweighted; and a person who feels a one-in-a-million chance as though it were larger will pay over the odds to chase the prize and over the odds to escape the ruin. The lottery and the policy are not two attitudes to risk but one attitude to small probabilities, applied once to a gain and once to a loss. And the same curve, behaving near certainty, resolves the paradox Maurice Allais had thrown at expected utility a generation before — for the leap in weight as a near-certainty becomes a certainty is exactly the disproportionate craving for sureness that Allais's choosers displayed and the orthodox theory forbade.

With value function and weighting function in hand, prospect theory stands as the most complete descriptive account of risky choice that economics has — and its very completeness invites the hardest question of the volume.

Loss aversion does not stay in the laboratory of gambles; it reaches into the having of things and the holding of money, and there it produces a family of departures so pervasive that economic life is built around them. The first and most studied is the endowment effect: the moment a thing becomes yours, you value it more than you did the moment before. In the experiment that named it, mugs were handed to half the people in a room and the room invited to trade. Standard theory was clear about what should happen — the mugs were distributed at random, tastes were not, so roughly half should change hands as they found their way to those who wanted them most. Almost none did. The owners, asked the least they would take to part with a mug, named about twice what the non-owners, asked the most they would pay to get one, would offer. The mug had not changed. Ownership had, and ownership had doubled its worth.

The cause is the kink of the previous plates. Once the mug is yours, it enters the reference point; parting with it is now a loss, and losses loom large, while acquiring one is a mere gain, and gains are discounted. The seller and the buyer are not disagreeing about the mug; they are standing on opposite sides of the reference, one facing a loss and one a gain, and loss aversion drives a wedge between what it takes to surrender a thing and what it takes to obtain it. This wedge has a consequence that reaches back into the foundations laid in earlier volumes. The Coase theorem had promised that when bargaining is cheap, the initial assignment of a right does not matter, since it will be traded to whoever values it most and the same efficient end will be reached either way. The endowment effect breaks the promise: whoever is handed the right values it more for being handed it, and will not trade it away, so the assignment sticks and the allocation we begin with is, very often, the allocation we keep. Where you start determines where you stay.

From the same root grow two more departures. There is the bias toward the status quo, the disproportionate pull of things as they are, because any change from the present state must be reckoned partly as a loss and the loss outweighs the matching gain — so the default option, whatever it happens to be, enjoys an advantage it has not earned, and inertia rules choices from insurance plans to the allocation of a pension. And there is the honoring of sunk costs, the refusal to treat money already spent and gone as the irrelevance the rational agent knows it to be. He finishes the tasteless meal because he paid for it, sits to the end of the tedious film, pours fresh money into the failing venture because so much has already gone in — for to abandon the thing is to confess the loss and feel its full weight, while to persist is to keep the loss unrealized a little longer, a comfort the convex loss-arm makes him pay for dearly.

And money itself, which the theory treats as the most fungible thing there is — a pound the same pound whatever its history — the mind insists on sorting into accounts. People keep mental ledgers: the grocery money and the holiday fund and the winnings from the wager are held in separate compartments, spent by different rules, guarded with different zeal. A windfall is squandered where an equal wage would be saved; money labelled winnings is gambled with a freedom money labelled savings never sees; a household runs a balance on a costly card while keeping cash untouched in a jar marked for something else, though paying the card would plainly leave it richer. This is mental accounting, and it violates the plainest axiom of all, that a pound is a pound. The reference point and the loss it defines have, by this plate's end, shown themselves to be not a curiosity of the gambling table but the hidden architecture of how people own, hold, and spend.

Of all the biases, the one that costs the most is the one that feels the least like a bias, because it feels like competence: overconfidence, the steady overestimation of how much we know, how well we judge, and how reliably the future will unfold as we expect. It wears three faces. The first is the conviction of being above average — most drivers rate themselves better than most drivers, most managers more capable than their peers, most of us more honest, more skilled, and, with a fine irony, less biased than the people around us, in proportions that arithmetic forbids. The second is the planning fallacy, the serene underestimation of how long a task will take and how much it will cost, undimmed by a lifetime of tasks that ran late and over: the renovation that doubles its schedule, the software that ships a year behind, the dissertation eternally a month from done. We plan for the morning when everything goes right, a morning that statistically never comes.

The third face is the most precisely measured, and the most damning, for it concerns not what we believe but how sure we are of it. Ask people factual questions and have them mark their confidence, and a systematic gap opens between the confidence claimed and the accuracy delivered. The things they are ninety percent sure of turn out true some seventy or seventy-five percent of the time; the confidence intervals they draw around their estimates are far too narrow, so that the truth falls outside their stated range with a frequency that should have shamed the range. This is miscalibration, and it does not wander; it leans always toward too much certainty, and the leaning grows as the certainty grows.

Two further habits guard the conceit against correction. Confirmation bias sets us seeking the evidence that flatters our beliefs and sliding past the evidence that would test them, so that a hypothesis once formed is defended rather than examined and grows more certain on a diet of its own confirmations. And hindsight bias, once an event has happened, rewrites our memory of what we expected, persuading us that we knew it all along — which poisons the very learning that might have taught us humility, for a mind that believes it foresaw the last surprise sees no reason to brace for the next. Confidence, shielded from disconfirmation by the one and from the lessons of error by the other, compounds untouched.

The costs fall where the stakes are highest. The overconfident trader, certain he knows what the market does not, trades too often and too large, and the most active traders are reliably the worst rewarded; the overconfident chief executive overpays for the acquisition and overbuilds the capacity; the overconfident forecaster and general and planner march their institutions into the gap between the morning they imagined and the one that came. Overconfidence is the bias that scales — harmless in trivia, ruinous in command.

The biases have been catalogued one mind at a time, but economics cares less about minds than about prices, and a price is the work of a market, not a person. So the decisive question is not whether individuals are biased — the foregoing plates have settled that they are — but whether their biases survive the market, or are washed out of it before they reach the price. The orthodox tradition gave a confident answer, and it is the market's version of the defense that opened this volume. Grant that many traders are biased; still, the argument runs, the rational few will see the mispricing the biased create and pounce on it, buying what fear has made too cheap and selling what greed has made too dear, and in the act of profiting they will push the price back to its proper value. The biased lose money to the rational and are bled from the market; the rational, enriched, come to set the price. The market is efficient though its traders are not, for arbitrage launders the folly out of the price and leaves only the truth.

It is a beautiful argument, and it rests on an assumption that does not hold: that arbitrage is riskless, costless, and unlimited. It is none of these. The arbitrageur who sells an overpriced asset short bears the risk that its true value rises against him; worse, he bears the risk that the mispricing widens before it closes — that the overpriced grows dearer still, the bubble fatter, as the very biases he is betting against drive other traders to pile in. The market, in Keynes's warning, can stay irrational longer than the arbitrageur can stay solvent. And solvency is precisely what fails him, for the arbitrageur trades on borrowed conviction and others' capital, and his backers, seeing his losses mount as the mispricing worsens, withdraw their money at the very moment it is most needed — so that arbitrage capital drains away exactly when the mispricing is widest and the opportunity greatest. The corrective force the orthodox story relied upon is real but bounded, and where it reaches its bound the bias is not laundered out of the price. It is priced in.

The evidence that biases reach the price is now a field of its own, behavioral finance, and it reads the markets for the fingerprints of the minds catalogued here. The premium that stocks have paid over bonds is far larger than risk alone can justify, and matches almost exactly what loss aversion, applied afresh each year to a portfolio watched too closely, would demand — the puzzle of the earlier volume's plate on risk, answered in the idiom of this one. Prices move far more than the news about fundamentals can warrant, an excess volatility that efficient valuation forbids. Momentum carries winners on and reversal drags them back, in patterns no fundamental tracks. And above it all stand the bubbles — the tulip, the company of the South Seas, the dot-com mania, the houses that could only rise — each a season in which price and value parted company before a watching world, and arbitrage stood by, unable or unwilling to close a gap that grew until it burst. The biases do not vanish at the threshold of the market. They tilt the cycle, swell the spread, and inflate the stock, reaching back to touch the volumes that thought prices were made of cooler stuff.

And yet — for the honesty of the table requires the and-yet, and here it is sharper than anywhere — whether the markets are truly inefficient is among the hardest questions in all of economics to settle, and may be unsettleable in principle.

If the mind departs from its own interest in regular and predictable ways, a possibility opens that the older economics had bolted shut: the architecture of a choice can be arranged so that the predictable departure carries the chooser toward his good rather than away from it. This is the practical harvest of the whole field, and it begins from an observation that is harder to dodge than it first appears. Every choice must be presented somehow — in some order, with some option foremost, against some backdrop, under some default — and there is no neutral way to present it, no arrangement that fails to shape what is chosen. Someone designs the form, sets the menu, writes the default; that someone is a choice architect whether the title is claimed or not; and since the architecture cannot help influencing, the question is only whether it will influence by accident or by design, carelessly or for the chooser's benefit.

The instrument of choice is the nudge: any feature of the architecture that alters behavior predictably without forbidding any option or materially changing its costs. Its purest form is the default, and the default is mighty for a reason this volume has already given — the bias toward the status quo, the inertia that leaves the preset option in place because moving from it is felt as a loss and trouble besides. Set the default well, and most people will accept it, freely and at no cost to their liberty, for the door to every other option stands open. The most striking demonstration concerns the gift of organs after death. Where the law makes donation a thing one must opt into, by an act of registration, scarcely more than a tenth of a population enrolls; where it makes donation the default, from which one may freely opt out, enrollment climbs toward nine in ten. The people are not different and their feelings about donation are not different. Only the default is different, and the default decides.

From this the architects built a philosophy and a practice. The philosophy, libertarian paternalism, claims to square a circle: paternalist in steering people toward what is good for them, libertarian in leaving every option open and every exit free, it promises help without coercion, a guiding hand that never closes into a fist. The practice spread with remarkable speed — automatic enrollment that turned reluctant savers into savers by the simple expedient of enrolling them unless they refused, units of behavioral insight installed in governments to apply the same art to health and energy and the paying of taxes, a whole apparatus of gentle steering erected on the findings these plates have surveyed. It is, by any worldly measure, the field's great success, the place where the science of the actual mind reached out of the journals and changed how institutions treat the people in their care.

But the nudge carries a worry it has not dispelled, and the worry runs straight back to the deepest question of the volume on the choice. The nudge steers toward what the architect judges to be the chooser's good — and this volume has spent eleven plates establishing that the chooser's preferences are constructed, framed, anchored, and reference-bound, which is precisely why they could be nudged in the first place. But then which of his shifting preferences is the true one, the one the architect should serve? If his choices are too biased to reveal his good — if revealed preference is broken, as the volume on the choice confessed it might be — then the architect cannot read the chooser's real interest off the chooser's behavior; and if not off his behavior, then off what? The same demolition that licenses the nudge undermines the standard by which the nudge would be judged good. Classical welfare economics had a clean answer to what was good for a person: whatever he chose, for he was sovereign over his own ends. Behavioral economics has shown, more vividly than any critic before it, that what is chosen and what is good come apart — and in showing it has knocked away the only welfare standard economics had, without supplying another. One cannot simply say the architect should steer toward what the chooser would pick if unbiased, for there is no unique unbiased choice to pick — strip away which bias, against which reference point? — and the rational self who would choose it is a fiction, the very fiction the field congratulated itself on exposing. The nudge, the field's proudest achievement, rests on a theory of welfare the field itself has reduced to rubble.

The volume set out to study the ways the actual mind departs from the model of it, and the departures, gathered, make a long and orderly list. There was the founding distinction, that the errors are systematic and not random, and so do not cancel but compound. There was the divided mind that makes them, fast and slow, and the heuristics by which the fast mind answers hard questions with easy ones. There were the three heuristics of judgment — availability, which mistakes the recalled for the frequent; representativeness, which mistakes the similar for the probable; anchoring, which cannot let go of whatever number it started from. There was prospect theory, the great positive achievement, with its value bent at the reference and steeper into loss, and its probabilities bent out of true. There were the departures that follow — the endowment that ownership confers, the inertia of the status quo, the honoring of sunk costs, the compartments of mental accounting, the conceit of the overconfident. And there were the consequences, traced into the market that arbitrage cannot fully discipline and the policy that the nudge would reform. The actual mind, which a century of economics had abstracted away, has been drawn in detail, and the detail is the volume's gift.

The order beneath the list is the volume's argument, and it is worth stating plainly at the close. Every error catalogued here has a direction, and the directions are not a thousand separate quirks but the predictable issue of a few mechanisms — the substitution of easy questions for hard, the dominion of a reference point, the asymmetry of loss and gain. To know the mechanism is to forecast the error before it is made, and a thing forecast is a thing understood, at least in part. And because the errors lean the same way for everyone, they survive the aggregation that would have dissolved mere noise: they reach past the single mind into the price, the bubble, the savings rate of a nation, the very places the older economics had declared the irrational could not reach. This is why the catalogue is a contribution to economics and not only to psychology. The mind's departures are not confined to the mind. They are abroad in the world, in the numbers the discipline most cares about.

Set beside the volume on the choice, this one completes a single inquiry conducted from two sides. That volume built the rational chooser in his elegance and, honest at the seams, confessed the places he strained against the human original; this volume has walked those seams one by one, mapped the strain, measured its direction, and offered in prospect theory a positive account of the creature who strains. The one asked where the model breaks; the other has answered, here and here and here, and by this much and toward that side. Model and critique, they are not rivals but halves — the architecture and the anatomy of the same chooser, the one indispensable to stating what he is supposed to be, the other to recording what he is. Neither stands complete without the other, and the table has now set down both.

With this volume the family of the Behaviors closes. Four volumes have theorized the one figure on whom every price in this work depends: the choice, which is his model, the rational agent maximizing under constraint; the bias, which is his critique, the systematic record of how the living mind departs from that model; the game, which is his strategic extension, the chooser set against other choosers as canny as himself; and the norm, which is his social alternative, the chooser bending to the unwritten rule that no maximizing could derive. They are four ways of seeing the same creature — as optimizer, as erring mind, as strategist, as member — and no one of them is the whole truth, for the chooser is at once all four, and the disagreements among them are not errors to be resolved but facets to be held together. The Behaviors are the last family the table needed to give its prices a chooser worthy of the name, and they are now entire.

What, then, is the cost the volume's question asked after — the cost of the mind departing from the model in regular ways? It is reckoned on three ledgers, and the volume has totaled each. For the individual, bias is a private tax: the savings not made, the debts run up, the health neglected, the thousand small mistakes that a sharper creature would not make and that this one makes predictably, against his own interest, to the profit of whoever has read the pattern. For the market, bias is a public distortion that the limits of arbitrage permit to stand: prices that wander from values, volatility past what the news can warrant, bubbles that inflate before a watching world because the force that should correct them is bounded and its capital flees as it is needed. And for the polity, bias is at once an opportunity and a snare — the opportunity of the nudge, which can move millions toward their own professed good at no cost to their liberty, and the snare of a paternalism that presumes to know a good it can no longer read from choice. The mind's regular departures are not free. Someone pays, on every ledger, and the volume has named who.

The verdict the table owes must be double, for the truth is double. Behavioral economics is a triumph, and the triumph should not be grudged: it dragged the actual mind into a science that had defined it out of existence, it is rich with fact where the old theory was a beautiful abstraction, it predicts and explains anomalies the old theory had ruled impossible, and it has reached out of the seminar to change how governments and firms and citizens conduct themselves. No honest accounting can call it anything but one of the great advances of the discipline's modern history. And yet it is incomplete in a way that its critics have named justly and that this volume has conceded at every turn. It is, in the unkind phrase, a butterfly collection — a dazzling cabinet of biases pinned and labeled, where the theory it dethroned had offered a single principle from which all behavior flowed. Its finest unification, prospect theory, hangs on a reference point it cannot itself fix and describes the anomalies more than it explains them. Its reach into the markets is contested and, by the joint-hypothesis problem, perhaps permanently undecidable. It is a research program caught in mid-stride, magnificent in its findings and still wanting the parsimony, the single organizing idea, that would make of its catalogue a theory. It unseated a tidy fiction and has not yet built a cathedral to replace the temple it pulled down.

But the gravest thing the volume leaves is not an unfinished science; it is an unhealed wound in the foundations of value, and it is the same wound the volume on the choice first opened. Behavioral economics has demonstrated, more vividly and more rigorously than any moralist or critic before it, that what a person chooses and what is good for him come apart — that choice, framed and anchored and reference-bound, does not reliably reveal welfare. This is a finding of the first importance, and it is corrosive, for the welfare standard that economics had leaned upon for a century was exactly the identification of the chosen with the good: the doctrine of revealed preference, the sovereignty of the consumer, the reading of a person's interest off his purchases, which the volumes on utility and on the choice traced to its root. Knock away the premise that choice reveals welfare, and that whole apparatus falls — and behavioral economics has knocked it away. What it has not done, what it cannot yet do, is raise another in its place. There is no agreed criterion of a person's good once his choices are disqualified from defining it, because the rational self whose choices might have counted is a fiction, and there is no single, principled way to strip the biases out — strip which, against what reference, restoring what self that never existed? The field has sharpened the oldest question of the table to a fine point and left it standing: what is good for a person, if not what he chooses? Having met the actual mind at last, economics finds it harder to serve than the tractable phantom it replaced. We have learned, past doubt, that the model was wrong about the mind. We have not learned what the mind's good consists in, nor what it would be to do well by it.

And so the volume ends where the honest ones must, with the encounter incomplete. The actual mind has entered economics and brought its beautiful disorder in with it; the science is the richer for the meeting and the foundations are the shakier, and both are true at once. The chooser of the whole table is not the maximizer the table presumed — this much the catalogue has established beyond recall — and what, in the maximizer's place, he is, and what it would mean to reckon his welfare and respect his choosing, the table cannot yet say. The argument remains open.