Behavioral economics, oriented

Dual-process theories split thinking into two types. System 1 is fast and automatic and leans on heuristics, mental shortcuts that save effort. System 2 is slow and deliberate and can correct a quick answer, but needs time and mental resources. The split is a model, not a settled fact; Gigerenzer, for one, rejects it. Shortcuts are not always costly: on many data sets, two simple rules that ignore some information predicted new cases better than linear regression, especially when the training sample was small.

Errors come when an easier question stands in for the one asked. Availability judges frequency by how easily examples come to mind, so flood insurance purchases rose in recently flooded areas and fell back as memories faded. Representativeness judges probability by resemblance to a typical case, and base rates can get neglected: in the classic taxi problem, people answer 80 percent, the witness's accuracy, when combining it with how few cabs are blue gives about 41 percent, though critics note the setup matters. Two events together can never be more probable than either alone, yet people often rate the pair higher; asking which is more frequent rather than more probable cut that error from 77 to 27 percent in one study. A statement seen before is rated more likely true, a small but reliable effect for true and false statements alike. An extreme reading tends to be followed by one nearer the average even without intervention, so a drop after a program is not proof it worked; an equivalent comparison group separates real change from that drift.

A known outcome seems more predictable than it seemed beforehand, and the same medical decision was rated better when it ended well, even by people who said outcomes should not count. Overconfidence includes overprecision, being more certain than your accuracy justifies.

The trap is believing that slowing down cures all of this. A second look catches some errors, but on one reflection test, time spent barely predicted accuracy. Say you are estimating how long a report will take, though similar reports ran late; underestimating anyway is the planning fallacy. The outside view starts from how long similar reports took, which works like a base rate, and in three experiments a second prediction was less biased when the earlier task had been the same task.

Rule: when an answer comes easily, ask which easier question you just answered, and find the base rate before you trust the details of the case.

Expected utility theory scores each option as a probability-weighted average of how much its outcomes are valued, and many behavioral findings are framed as departures from it, though whether they are errors or reasonable preferences is debated. In Kahneman and Tversky's prospect theory, outcomes are coded as gains or losses from a reference point, such as current wealth, an aspiration or an imagined comparison. Each extra unit matters less the farther it lies from that point, which predicts playing safe with gains and gambling with losses. In the classic disease problem, most people chose the sure option when outcomes were described as lives saved and the gamble when the same outcomes were described as deaths, a pattern a 47-country survey replicated with substantial variation between nations.

A loss weighs more than an equal gain on average, but less than the classic estimate of 2.25 times: a 2024 meta-analysis put the mean near 2 and the median near 1.7, and some experiments found no loss aversion for small stakes. Small stated probabilities get extra weight, one proposed explanation for both insuring against rare losses and betting on rare gains, though people who learn about rare outcomes through experience tend to underweight them instead. Past costs cannot be recovered whichever way you decide, yet they still pull people toward continuing what they started. Ownership is said to add value too: the endowment effect is valuing an item more once you own it, classically because giving it up feels like a loss. Its cause is disputed, since in one meta-analysis and three experiments owning did not change beliefs about an item's quality or price, and the effect vanished among Kanak participants tested in a Kanak setting.

The trap is treating the starting point as neutral. Because the status quo is often the reference point and losses loom larger, the option already in place can start with an edge. One survey exposed that edge by pairing familiar drugs with invented ones of the same properties: the familiar legal ones drew more support for staying legal. Say you are deciding whether to keep a service you already pay for. Picture it offered to you today as a new option beside the alternatives, and ask whether you would pick it.

Rule: name your reference point, notice whether a gain or a loss is doing the deciding, and judge what you already hold as if it were new.

Classic decision theory assumes description invariance: two wordings of the same outcome should not change a choice. Framing research shows they often do, in a predictable direction. Beef labeled 80 percent lean is also 20 percent fat; in a meta-analysis of food studies, such framing moved attitudes a lot and intentions almost not at all. Reacting to a frame is not always irrational, since wording can carry real information about what the speaker expects or prefers. Some well-known results proved weaker: stressing what you lose by not acting, rather than what you gain by acting, made almost no average difference across 165 effect sizes. And the classic finding that far more people would make a trip to save 5 dollars on a 15-dollar item than on a 125-dollar one failed in a 2025 replication, though the authors caution that this does not prove the effect is gone.

Mental accounting treats money by its source or its earmark, even though every dollar buys the same things. People who lost a 10-dollar theater ticket were less willing to buy another than people who lost a 10-dollar bill, a result a 2025 replication confirmed with a smaller gap, and mental budgets are also thought to aid self-control. Evidence that a prior win makes people take bigger risks is mixed: a 2025 replication found no such increase after an imagined 30-dollar win, while a meta-analysis of 36 studies of unexpected income found a low-to-moderate effect, weak and nonsignificant in field experiments. How you pay matters as well: cash is held to cause the most pain of paying, and cards and phone payments are linked to more spending, though the mechanism is disputed. The pleasure of a good deal shifts with setting, too; people would pay more for the same soda from a fancy resort hotel than from a run-down grocery store, a small effect.

The trap is reacting to the unit instead of the total. In a 2025 test, a yearly membership described as merely 27 cents a day was rated more attractive than one at 100 dollars a year, and the gap shrank but held side by side. Say a plan is pitched to you in cents a day. Restate it as the yearly total, then ask what else that money could buy; in experiments, such reminders lowered willingness to buy. Describing a choice in both frames at once weakens framing effects too, though not always fully.

Rule: restate every offer in the opposite frame and as a total, then name what else the same money would buy.

A first number pulls later estimates toward itself, even when it is plainly irrelevant, and adjustment away from it often stops short. In a classic demonstration, guesses of the share of African countries in the United Nations tracked a number from a wheel of fortune. The effect is robust for numeric estimates but inconsistent for what goods are worth: in one set of experiments, arbitrary anchors shifted willingness to pay substantially only when answers were hypothetical, not when real money was at stake. The Federal Trade Commission's pricing guides treat a former price as a legitimate comparison when the product was genuinely offered at it, regularly, for a reasonably substantial period, and call a price inflated briefly to stage a cut a false bargain.

The rest of the menu works on you as well. An option tends to gain share just because it sits in the middle of the range, so adding a pricier top tier can turn the old top choice into the middle one. A decoy that is worse than one option on every attribute tends to help that option, but concerted replication attempts found the effect only in stylized lab tasks with numeric attributes, and in 3.6 million grocery wine purchases decoys raised their target's share by roughly 1 percent. The claim that more options reduce buying is disputed: one meta-analysis put the average effect near zero, while another tied overload to conditions such as complex sets, hard tasks or uncertain preferences. Charm prices such as 1.99 dollars assume buyers fix on the left digit, an effect that shows up in some data and not others. A high price can read as quality, so a fictitious drug that did nothing was judged more effective when described as expensive. Position counts too: candidates gained from being listed first in 86 percent of 84 New Hampshire election tests.

The trap is letting the display set your starting point. Say a plan comes in three tiers, with a premium one you would never buy. Decide what you need before you look at the lineup; in one study, people who chose before seeing the extra options were less pulled toward the middle. When a number is on the table, list reasons it might be too high, a step that weakened a high anchor's effect on some ratings in one experiment, though only partly.

Rule: set your own number and your own needs first, then read every printed price and middle tier as context, not as evidence of value.

Classic economic theory says a delayed reward should lose the same share of its value for each added period of delay. People instead discount the near future more steeply than the distant one, and present bias models this as extra weight on whatever is available right now. The predicted result is a reversal: asked a week ahead, people chose a healthy snack more often than when the snack was for right now. The evidence is mixed, though; in a weeks-long study with real money, reversals toward the sooner amount were not significantly more common than reversals the other way.

Only someone who foresees the conflict can plan against it. A commitment device is a voluntary arrangement that makes failing a goal costly, yet signing up is not the same as changing: over a third of middle school students opted into one tied to better conduct, and their behavior did not change. Save More Tomorrow commits part of future raises to saving; one paper calls it effective at raising saving rates, though one review argues stagnant wages blunted it. A 2026 replication of one deadline study found negligible effects. Installment framing points you at the monthly payment, which can hide that a longer loan term means more interest overall.

Forecasts of feeling miss as well. People usually predict fairly well whether an event will feel good or bad, but misjudge how intense and lasting the feeling will be, and feelings tend to drift back toward a usual level after gains and losses, though not always fully: in over 3,500 older Europeans, well-being recovered after short-term disability but adapted much less after lasting disability. Memory keeps its own score. Remembered pain tracked the worst moment and the final minutes, and most people chose to repeat a 90-second cold-water trial whose last 30 seconds were slightly warmer over a 60-second one, though this evidence comes mostly from simple episodes.

The trap is planning in a calm state and expecting the plan to survive a hot one. People tend to underestimate cravings they are not feeling; smokers who were not craving underestimated how much they would later value a cigarette. Say you resolve after a big meal to skip the snacks at work tomorrow. Plan for the hungry version of yourself, for example with friction between the impulse and the act, which cut self-reported time on a problem app in one small trial.

Rule: decide in advance, and plan for a future self who will want the sooner reward more than you expect today.

Other people's behavior is information. A descriptive norm says what most people do, and social proof treats that as evidence of what works; an injunctive norm says what people approve of. In energy field experiments, norm messages mostly produced small effects, and in German field experiments one sign stating what most people do and approve of cut paper towel use and raised mask wearing but had no effect on plastic coffee lids. Telling people an unwanted behavior is common can backfire by making it sound normal, though not every study finds that boomerang.

Following others can even be rational: in an information cascade, people deciding in sequence copy those before them, so an option can lock in after most private evidence points the other way. The Securities and Exchange Commission warns that jumping on the bandwagon with heavily promoted stocks may mean significant losses. In a replication of Asch's line-judging study, people matched the group's wrong answer on 33 percent of critical trials, and most answers stayed independent. Pooling 21 of Milgram's conditions, 43.6 percent of participants continued to the maximum shock, and obedience varied with how legitimate, directive and consistent the experimenter was. Each added shared trait raised how likable a stranger seemed, though those studies measured liking, not compliance, and Cialdini proposes that shared group identity makes people more willing to comply. A favor is thought to create a felt obligation to repay, yet prepaid incentives made no difference in one survey trial, and the foot-in-the-door effect, though one review found it more often than chance allows, fails in many studies and occasionally reverses. In negotiation experiments, people routinely turn down unfair offers that are better than nothing.

The trap is reading a display of popularity as independent evidence. Say a product page shows thousands of glowing reviews, a countdown timer and a note that the item is trending. The Federal Trade Commission has described review sites a seller secretly controls, comparison sites that rank companies by what they are paid, fake followers for sale and countdown timers on offers that are not really time-limited, and its 2024 rule targets fake reviews. Ask who produced each signal and what they gain if you believe it.

Rule: before you follow the crowd, check whether the crowd is independent, informed and real.

Choice architecture is how a decision is laid out: how many options there are, which one is the default, how each is described and where it sits. Thaler and Sunstein argue that no layout is neutral, since every arrangement makes some choices more likely. A nudge changes the layout to shift behavior predictably without banning an option or significantly changing incentives. Defaults stick partly because people tend not to act to change the status quo: when a state enrolled child care providers in a text-message program by default, fewer than 3 percent of those reached opted out, and the Department of Labor says automatic enrollment raises retirement plan participation.

Measured at scale, the effects are small on average, and published averages appear inflated by publication bias. A large meta-analysis of over 200 studies reported an average effect, but a reanalysis correcting for publication bias found no clear evidence of one. Trials run by two large US nudge units, where every trial was counted, averaged about 1.4 percent against 8.7 percent in academic papers, small but still significant. Text reminders raised booster uptake by about 1 percentage point in a pharmacy megastudy, while reminders before passport appointments did not change no-shows. When Virginia let rental-aid applicants attest to their income instead of submitting documents, approvals rose by more than 10 percentage points, yet new applications did not rise significantly. Opt-out countries had about six times higher organ donor registration, yet registration does not guarantee more donations. Critics ask whether nudges respect autonomy and whether a focus on individual choice pulls attention from structural causes.

The trap is forgetting that the same tools can serve the seller. The Federal Trade Commission lists a default that is good for the company but not the user among dark patterns. Say you are signing up for a free trial and a box is already checked. Agency guidance warns that the trial will probably turn into a charge if you do not cancel in time, and that a pre-checked box may authorize further charges or data sharing. Because opting out takes more effort than going along, the starting setting often becomes the final one.

Rule: at every sign-up, ask what happens if you do nothing, and change any default that serves someone else's goal instead of yours.

Dark patterns are interface designs that trick or manipulate people into choices they would not otherwise make, often by exploiting cognitive biases or delaying needed information. They are common, and the tactics tend to work more strongly together. The Federal Trade Commission's list includes fees revealed late in checkout, where buyers not shown fees up front spent about 20 percent more; fake scarcity, such as a countdown that resets when it runs out; confirmshaming, which words the decline option so that saying no feels foolish; and cancellation mazes, including one alleged path of six to nine screens. In one study, more than twice as many people accepted a free trial when the monthly charge appeared only in small gray print. Since May 2025 a commission rule has required the total price up front for live-event tickets and short-term lodging, and the rules are still changing.

In real trading, investors tend to sell winners too soon and hold losers too long, and in a classic study of 78,000 brokerage households, frequent traders netted less than infrequent ones. Recent winning stocks tend to keep outperforming for months, which some explain as slow reaction to gradually arriving information and others link to changing risk exposure. Mispricing can persist when trading costs or short-selling limits block its correction, though persistence is not proof of error.

Read the field's own findings with the same care. Many published behavioral results shrink or vanish when repeated: one project repeating about 100 psychology studies replicated roughly 40 percent, though no such sample gives a single rate for the field. Ego depletion, the claim that self-control draws on a resource that gets used up, went from a moderate effect in early published studies to tiny or null effects in preregistered multilab tests, and it remains contested. A well-known 1996 report that an old-age stereotype made people walk more slowly did not hold up when a replication used automated timing.

The trap is trusting one memorable study. Say you read that holding an expansive pose changes your hormones and your appetite for risk. The original study reported that, but later evidence has not supported those claims. Findings are argued to be less likely true when studies are small, effects are small and analyses are flexible.

Rule: before you click, find the total price, the end date and the way out; before you believe a finding, ask whether it replicated.