Behavioral Economics — Heuristics, Biases, Nudges

Behavioral economics replaces the idealized homo economicus of neoclassical theory — fully rational, narrowly self-interested, with stable preferences and unlimited computational capacity — with a more realistic agent: cognitively bounded, emotionally embedded, socially motivated, and systematically biased. The field is a synthesis of cognitive psychology and economics, and it now informs everything from retirement-savings defaults to organ-donation registration to public-health messaging. Its core claim is empirical: people’s systematic deviations from expected-utility maximization are predictable, structured, and economically consequential.

This note covers the historical foundations, the prospect-theory revolution, the canonical heuristics-and-biases catalog, social preferences, intertemporal choice, choice architecture, the field-experiment turn, behavioral finance, the replication crisis, and policy applications.

---

1. Origins and intellectual lineage

1.1 Herbert Simon and bounded rationality (1955)

The first serious crack in the rational-agent framework came from Herbert Simon (Nobel 1978), whose 1955 paper “A Behavioral Model of Rational Choice” argued that real decision-makers face cognitive constraints — limited memory, limited attention, limited computational time — and therefore satisfice rather than optimize. They search for an option that exceeds an aspiration threshold and stop, rather than enumerating the full feasible set and maximizing utility over it. Simon’s framework anticipated almost everything that followed, but the field needed a sharper empirical handle before it could displace expected-utility theory in mainstream economics.

1.2 Kahneman and Tversky: the heuristics-and-biases programme

That handle arrived through the collaboration of Daniel Kahneman (Nobel 2002) and Amos Tversky (d. 1996, would have shared the Nobel). Their 1974 Science paper “Judgment under Uncertainty: Heuristics and Biases” catalogued three mental shortcuts — representativeness, availability, and anchoring-and-adjustment — and showed how each produced systematic, replicable errors in probabilistic reasoning. Their 1979 Econometrica paper “Prospect Theory: An Analysis of Decision under Risk” then provided a positive theory: a mathematically explicit, empirically grounded alternative to expected-utility theory.

1.3 Thaler and the economic turn

Richard Thaler (Nobel 2017) translated K&T’s psychological findings into economic theory. His 1980 paper “Toward a Positive Theory of Consumer Choice” introduced the endowment effect; “Mental Accounting Matters” (1999) formalized the violation of fungibility; and Nudge (2008, with Cass Sunstein) launched the field’s policy era. The 2017 Nobel committee specifically cited his work on bounded rationality, bounded self-control, and social preferences.

1.4 Other foundational figures

• Vernon Smith (Nobel 2002, shared with Kahneman) pioneered experimental economics as a methodology — induced-value theory, market experiments demonstrating efficient convergence of double-auction markets, asset-market bubbles in laboratory settings (Smith, Suchanek & Williams 1988).
• Maurice Allais (Nobel 1988) — the 1953 Allais paradox preceded prospect theory by 25 years and demonstrated independence violations.
• Robert Shiller (Nobel 2013) — behavioral asset pricing, irrational exuberance, narrative economics (2019).
• Sendhil Mullainathan & Eldar Shafir — Scarcity (2013); how cognitive load from poverty produces tunnel vision and worsens decision quality, a structural-cognitive bridge between behavioral and development economics.

---

2. The expected-utility baseline

2.1 von Neumann–Morgenstern axioms (1944)

To know what counts as a violation, one must first state the rational benchmark. John von Neumann and Oskar Morgenstern (1944, Theory of Games and Economic Behavior) showed that any agent whose preferences over lotteries satisfy four axioms — completeness, transitivity, continuity, and independence — behaves as if maximizing the expectation of some real-valued utility function u(·). That utility function is unique up to a positive affine transformation: u and a·u + b (with a > 0) represent the same preferences.

• Completeness: for any pair of lotteries L, L’, either L ≿ L’ or L’ ≿ L (or both).
• Transitivity: L ≿ L’ and L’ ≿ L” implies L ≿ L”.
• Continuity: if L ≿ L’ ≿ L”, there exists a probability p such that L’ ∼ pL + (1−p)L”.
• Independence: L ≿ L’ ⇒ pL + (1−p)L” ≿ pL’ + (1−p)L” for any L” and p ∈ (0,1].

The Allais paradox (1953) is the canonical violation of independence and was the first widely-cited piece of evidence that real preferences are not vNM-rational.

2.2 Why violations matter

Expected utility is normatively defensible (a Dutch-book argument can be assembled against any agent who violates it), so the existence of robust violations is genuinely surprising. It also has structural consequences: if independence fails, you cannot reduce compound lotteries to simple ones, dynamic-programming approaches to choice break down, and “rational” market prices become ill-defined.

---

3. Prospect theory (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992)

The 1979 Econometrica paper proposed three core departures from EU. The 1992 cumulative variant fixed mathematical issues with how the original handled stochastic dominance.

3.1 Reference dependence and the value function

Utility is defined not over final wealth states but over gains and losses relative to a reference point (typically the status quo). The value function v(·) has three properties:

1. Reference-dependence: v(0) = 0; the same dollar feels different depending on whether you are above or below the reference.
2. Diminishing sensitivity: v is concave for gains (x > 0) and convex for losses (x < 0). A second $100gainedmatterslessthanthefirst;asecond$100 lost also matters less than the first.
3. Loss aversion: the function is steeper for losses than for gains. The canonical estimate is λ ≈ 2.25 (Tversky & Kahneman 1992) — losing $100feelsroughly2.25\times asbadasgaining$100 feels good.

A common parameterization:

• v(x) = x^α for x ≥ 0
• v(x) = −λ(−x)^β for x < 0 With α = β ≈ 0.88 and λ ≈ 2.25 fitting median experimental data.

3.2 Probability weighting

People do not use objective probabilities p; they use decision weights π(p) derived from a weighting function w(·). w is inverse-S-shaped:

• Small probabilities are overweighted (lottery tickets, terrorism fears, insurance against rare disasters).
• Moderate-to-large probabilities are underweighted.
• w(0) = 0 and w(1) = 1, but there are discontinuities at the endpoints.

A common functional form (Tversky-Kahneman 1992): w(p) = p^γ / [p^γ + (1−p)^γ]^(1/γ), with γ ≈ 0.61 for gains and ≈ 0.69 for losses.

3.3 Four-fold pattern of risk attitudes

Combining the value function and probability weighting yields the fourfold pattern:

	High probability	Low probability
Gains	Risk averse (sure gain over lottery)	Risk seeking (lottery tickets)
Losses	Risk seeking (gamble to avoid sure loss)	Risk averse (buy insurance)

This single table explains both the casino and the insurance company — agents on the same day buying both.

---

4. Heuristics and biases

K&T’s research programme catalogued mental shortcuts. They are usually adaptive (cheap, fast, often accurate), but they fail predictably.

4.1 Anchoring

Initial values exert disproportionate influence on subsequent estimates. K&T’s wheel-of-fortune study (1974) spun a rigged wheel landing on 10 or 65, then asked subjects to estimate the percentage of African nations in the UN. Median estimates: 25% vs 45%. The anchor was visibly random, irrelevant, and unforgettable — and still moved answers.

4.2 Availability heuristic

Frequency is judged by ease of retrieval. People overestimate causes of death that are vivid (homicide, plane crashes) and underestimate prosaic ones (diabetes, stroke). Media coverage shapes risk perception more than statistics.

4.3 Representativeness

Probability is judged by similarity to a stereotype.

• Linda problem (Tversky & Kahneman 1982). Linda is described as bright, philosophy-major, anti-nuclear activist. Subjects rank “Linda is a bank teller and active in the feminist movement” as more probable than “Linda is a bank teller” — a textbook conjunction fallacy (P(A∩B) ≤ P(A) always). Around 85% of subjects commit it.
• Base-rate neglect: in the classic taxicab problem, given a witness who is 80% reliable and a city where 15% of cabs are Blue, most subjects estimate P(Blue | witness says Blue) ≈ 80%. The correct Bayesian answer is ≈ 41%.

4.4 Sunk-cost fallacy

Costs already incurred should be irrelevant to future decisions, yet they consistently are not. The Concorde was kept alive for decades by sunk-cost reasoning; individuals finish bad meals “because they paid for them”; firms double down on failing projects.

4.5 Confirmation bias and hindsight bias

Confirmation bias: seek and weight evidence supporting current beliefs (Wason 4-card task, 1960). Hindsight bias: “I knew it all along” — after learning an outcome, people misremember their prior probability estimates as closer to the truth (Fischhoff 1975).

4.6 Planning fallacy and overconfidence

Buehler, Griffin & Ross (1994): students predict thesis completion in 34 days on average; actual is 56 days. Overconfidence appears in three forms (Moore & Healy 2008): overestimation (of your own ability), overplacement (above-average effect — 88% of US drivers think they are above-median), and overprecision (confidence intervals too narrow).

The planning fallacy generalizes to project management (“Hofstadter’s Law: it always takes longer than you expect, even when you take into account Hofstadter’s Law”), software estimation (Standish Group CHAOS reports consistently show ~70% of large IT projects over schedule and budget), and infrastructure megaprojects (Bent Flyvbjerg’s Oxford program on cost overruns: ~9 in 10 megaprojects overrun by an average of 28%).

4.7 Affect heuristic and emotion in judgment

Slovic, Finucane, Peters & MacGregor (2007): risk and benefit judgments are negatively correlated in subjective ratings but positively correlated in reality (high-benefit activities are usually high-risk). The affect heuristic says global feeling about an object substitutes for analytic risk-benefit decomposition. Implications include the dread risk bias — outsized fear of nuclear power, terrorism, and pandemics relative to base rates.

4.8 Dual-process theory: System 1 / System 2

Kahneman’s Thinking, Fast and Slow (2011) packaged the field for general audiences around a two-system model:

• System 1: fast, automatic, associative, intuitive, parallel, effort-free.
• System 2: slow, deliberative, rule-based, serial, effortful.

System 1 generates impressions and intuitions; System 2 endorses or overrides them. Most biases arise when System 1 produces a confident wrong answer and System 2 fails to engage. The model is descriptive shorthand rather than a neuroanatomical claim and has drawn critique (Keren & Schul 2009 — single vs dual-process debate; Melnikoff & Bargh 2018) but remains useful pedagogy.

---

5. Framing effects

5.1 The Asian-disease problem (Tversky & Kahneman 1981)

A disease is expected to kill 600 people. Two framings:

• Gain framing: Program A saves 200 for certain; Program B has 1/3 chance of saving all 600, 2/3 chance of saving none. Majority chooses A.
• Loss framing: Program C lets 400 die for certain; Program D has 1/3 chance no one dies, 2/3 chance all 600 die. Majority chooses D.

A = C and B = D in expected outcomes — only the description differs. Preferences reverse with framing. This is a direct violation of description invariance, one of the implicit axioms behind rational choice.

5.2 Goal framing and message framing

Health communication research distinguishes:

• Goal framing: emphasizing the consequences of acting (gain) vs not acting (loss). Loss frames are typically more persuasive for detection behaviors (mammograms, HIV testing); gain frames for prevention (sunscreen, exercise) — Rothman & Salovey 1997.
• Attribute framing: “90% lean” vs “10% fat” — same product, different evaluations.
• Risky-choice framing: as in the Asian-disease problem.

5.3 Narrow vs broad framing

Investors evaluating portfolios stock-by-stock (narrow) vs as a whole (broad) reach different conclusions because of loss aversion. Read & Loewenstein (1995) on snack choice: subjects asked once for seven days of snacks pick a variety; subjects asked daily mostly pick the same favorite. Broad framing increases diversification; narrow framing increases risk aversion.

---

6. Endowment effect

Kahneman, Knetsch & Thaler (1990) ran the canonical mug experiment: half a class is given a coffee mug; half is not. The endowed half is asked their minimum selling price (WTA — willingness to accept); the unendowed half is asked their maximum buying price (WTP — willingness to pay). Standard theory predicts both are equal (the mug’s reservation value should not depend on who possesses it). Empirically, median WTA ≈ $7;medianWTP\approx$3 — roughly the loss-aversion ratio. Trade volume is far below the predicted 50%.

Endowment effects vanish for experienced traders (List 2003), for fungible goods, and for tokens with explicit exchange value. They appear strongest for goods consumed via possession.

---

7. Mental accounting (Thaler 1985, 1999)

People organize money into psychological accounts that violate fungibility — the principle that a dollar is a dollar, regardless of source or label.

• Source-dependence: tax-refund dollars are spent more freely than salary dollars (Shefrin & Thaler 1988).
• Payment decoupling: prepaid vacations feel free at the time of consumption; pay-as-you-go casinos hurt with every pull.
• Sunk costs in narrow accounts: theatergoers refuse to buy a replacement ticket after losing one ($20)butreadilybuyaticketafterlosing$20 cash — same dollar amount, different mental account.
• Hedonic editing: bundle losses, segregate gains.

---

8. Intertemporal choice and self-control

8.1 Hyperbolic vs exponential discounting

Exponential discounting (the rational benchmark) values future utility at U·δ^t for some δ ∈ (0,1). It is dynamically consistent: preferences between two future options do not flip as time passes.

Empirically, people exhibit present bias — a sharp discount for immediate vs near-future rewards, then a much shallower discount thereafter. Laibson (1997) proposed the quasi-hyperbolic β-δ model:

• U(c₀, c₁, c₂, …) = u(c₀) + β · Σ_{t≥1} δ^t · u(c_t), with β ∈ (0,1).

If β = 1, the agent is exponential. If β < 1, the agent is present-biased. Estimates put β ≈ 0.5–0.8 for typical adults. The model captures both procrastination (cost in present, benefit in future → β multiplies the benefit → underdo it) and overconsumption (benefit in present, cost in future → overdo it).

8.2 Commitment devices

Present-biased agents predict their future selves will defect and demand commitment. Examples:

• Ulysses contracts: stickK.com lets users put money at risk against a future goal.
• Christmas clubs: low-liquidity savings accounts that prevent withdrawal.
• Save More Tomorrow (SMarT) — Thaler & Benartzi (2004): employees commit in advance to increase 401(k) contributions automatically with each future raise. Three years after enrollment, contribution rates rose from 3.5% to 13.6% in the original study; most employees stayed enrolled.

8.3 Dual-self / planner-doer models

Thaler & Shefrin (1981) modeled the agent as a long-run planner and a series of short-run doers. The planner can set rules, defaults, and constraints; the doer chooses moment-to-moment. This is a structural rather than just descriptive model of self-control.

---

9. Social preferences

Pure self-interest is empirically false. People care about others’ payoffs, about fairness, and about reciprocity.

9.1 Ultimatum game (Güth, Schmittberger & Schwarze 1982)

Proposer is given $10 and offers a split (x, 10−x) to Responder. Responder accepts (both get the offer) or rejects (both get 0). Subgame-perfect Nash equilibrium for self-interested agents: Proposer offers ε, Responder accepts. Empirically:

• Median offer is ≈ 40–50%.
• Offers below 20% are rejected ≈ 50% of the time.
• The pattern is robust across hundreds of replications and most cultures (though magnitudes vary — Henrich et al. 2001 cross-cultural study).

9.2 Dictator game

Strip the Responder’s option to reject. Pure rational self-interest predicts dictators keep everything. Empirically, mean donation is ≈ 20–30% of the endowment; modal split is often 50/50 or 0/100 bimodal.

9.3 Trust game (Berg, Dickhaut & McCabe 1995)

Player 1 sends some amount x of $10toPlayer2;theexperimentertriplesit;Player2returnssomeamounty.Standardtheorypredictsx=y=0.Empirically:averagex\approx$5, average y ≈ $4.5. Trust is extended and partially reciprocated.

9.4 Public-goods game

N players contribute privately to a public pot; the pot is multiplied (by, say, 1.6) and split equally. Free-riding dominates individually; cooperation maximizes total payoff. Empirically: initial contributions ≈ 40–60%, declining toward 0 over rounds unless punishment is allowed (Fehr & Gächter 2000) — in which case high cooperation is sustained.

9.5 Inequity aversion

• Fehr & Schmidt (1999): U_i = x_i − α_i · max(x_j − x_i, 0) − β_i · max(x_i − x_j, 0). Disadvantageous inequity hurts more than advantageous inequity (α > β); both reduce utility.
• Bolton & Ockenfels (2000): ERC (Equity, Reciprocity, Competition) — agents value their absolute payoff and their relative share, with diminishing returns to both.

9.6 Reciprocity and gift-exchange

Akerlof (1982) and Fehr, Kirchsteiger & Riedl (1993) modeled labor markets as gift exchange: above-market wages are reciprocated with above-minimum effort. Experimental gift-exchange games reliably show wage-effort correlations that pure self-interest cannot generate.

9.7 Altruistic punishment

Fehr & Gächter (2002, Nature): subjects in public-goods games pay private costs to punish free riders even in one-shot interactions with no reputation channel. Punishment opportunities collapse free-riding and sustain near-100% cooperation. The willingness to “burn money to make a point” is a robust phenomenon across cultures, though magnitudes vary (Henrich et al. 2006 cross-cultural follow-up).

9.8 Pro-social warm glow vs pure altruism

Andreoni (1989, 1990) distinguished:

• Pure altruism: utility depends on the welfare of others; my donation and a stranger’s donation are perfect substitutes; full crowd-out by government transfers.
• Warm glow / impure altruism: I get utility from the act of giving itself; substitution is incomplete; partial crowd-out.

Empirical crowd-out estimates from charitable giving are typically 20-40% — consistent with impure altruism, inconsistent with pure altruism.

---

10. Choice architecture and nudges

10.1 Defaults

Johnson & Goldstein (2003) compared organ-donor registration rates across countries:

• Opt-in (Germany 12%, UK 17%, US 28%): “check this box to donate.”
• Opt-out (Austria 99%, France 99%, Hungary 99%): “check this box if you do NOT want to donate.”

The intervention is purely architectural — the choice set, the cost of choice, and the underlying preferences are unchanged. Effect sizes of 60–80 percentage points are typical. Defaults work via (1) inertia, (2) implicit endorsement, (3) reference-point setting.

10.2 Madrian-Shea 401(k) defaults (2001)

A large US employer switched 401(k) enrollment from opt-in to opt-out. Participation jumped from ≈ 49% to ≈ 86% within three months of hire. Most employees stayed at the default contribution rate and default fund — the “path of least resistance.”

10.3 Libertarian paternalism

Thaler & Sunstein’s Nudge (2008) framed the design philosophy: arrange the choice environment so that the predictably-biased default is also the welfare-maximizing default, without removing options. Critics (e.g. Hausman & Welch 2010) argue this is still manipulation; defenders argue every choice context has some architecture and the only question is whether it is designed deliberately.

10.4 Sludge

The dual of nudges: choice-architecture frictions that make beneficial actions harder (insurance-claim paperwork, subscription cancellation flows, eligibility-verification hurdles). Sludge audits are now common in policy design.

10.5 The NUDGES taxonomy

Thaler & Sunstein offer a memorable taxonomy of nudge tools:

• N: iNcentives — make benefits salient.
• U: Understand mappings — translate technical outputs into life consequences (kWh → dollars; fat grams → minutes of exercise).
• D: Defaults — the highest-effect-size tool.
• G: Give feedback — energy-use dashboards, calorie counters.
• E: Expect error — design for slips and lapses.
• S: Structure complex choices — sort filters, recommendations, tiered menus.

10.6 Salience and attention

DellaVigna (2009) “Psychology and Economics: Evidence from the Field” surveys salience effects:

• Tax salience (Chetty, Looney & Kroft 2009): tax-inclusive vs tax-exclusive prices change demand even when the tax is otherwise known.
• Shrouded attributes (Gabaix & Laibson 2006): hidden fees, drip pricing.
• Inattentive consumers: limited price comparison, sticky brand choices.

---

11. Field experiments and the credibility revolution

11.1 RCTs in development economics

Banerjee, Duflo & Kremer (Nobel 2019) industrialized randomized controlled trials in development economics through J-PAL (Abdul Latif Jameel Poverty Action Lab, MIT). Major findings:

• Deworming in Kenya: low-cost mass deworming improved school attendance by ≈ 25% (Miguel & Kremer 2004).
• Conditional cash transfers: Mexico’s Progresa/Oportunidades programme (Levy 1997) gave cash payments to families conditional on children’s school attendance and health checkups. School enrollment rose 8–15 pp; the design has since been replicated in 60+ countries.
• Microcredit: Banerjee, Karlan, Zinman and others found much smaller effects on poverty than the original Grameen Bank narrative suggested.

11.2 Audit studies

Bertrand & Mullainathan (2004) sent ≈ 5,000 fictitious résumés to job ads in Boston and Chicago, randomly assigning either “white-sounding” (Emily, Greg) or “black-sounding” (Lakisha, Jamal) names. White names received callbacks ≈ 50% more often, equivalent to ≈ 8 additional years of experience. This was the first widely-cited audit study using résumé randomization.

11.3 Energy nudges

Allcott & Mullainathan (2010) evaluated Opower Home Energy Reports — printed mailers comparing a household’s electricity use to its neighbors’. Effect: ≈ 2% reduction in consumption, robust across millions of households. Cost-effective relative to traditional efficiency programs.

11.4 Karlan & Zinman on microcredit

Karlan & Zinman (2010, 2011) ran RCTs on consumer credit in South Africa and the Philippines, finding limited long-run welfare gains and significant heterogeneity — undermining boosterish claims of microcredit as a poverty silver bullet.

---

12. Behavioral finance

12.1 Equity premium puzzle

Mehra & Prescott (1985): US equities returned ≈ 6 pp/year more than safe bonds (1889–1978). Standard CRRA utility cannot rationalize this gap without implausibly high risk aversion (γ > 30). Benartzi & Thaler (1995) proposed myopic loss aversion as an explanation: investors evaluate portfolios too frequently (annually rather than over a multi-decade horizon) and feel each year’s losses with λ ≈ 2.25.

12.2 Disposition effect (Shefrin & Statman 1985)

Retail investors sell winners too early and hold losers too long. Mechanism: realizing a gain feels good (concave gain region of prospect theory); realizing a loss feels terrible (convex loss region — better to gamble and hope to break even). Odean (1998) documented the effect using 10,000 brokerage accounts.

12.3 Momentum and reversal

• Jegadeesh & Titman (1993): stocks that outperformed over the past 3–12 months continue to outperform over the next 3–12 months.
• De Bondt & Thaler (1985): stocks that underperformed over 3–5 years tend to reverse over the next 3–5 years.

Both are inconsistent with weak-form market efficiency. Behavioral explanations include underreaction (anchoring on prior expectations) and overreaction (representativeness).

12.4 Home bias and herding

Investors overweight domestic equities relative to mean-variance optimality (French & Poterba 1991). They also herd — copying analyst recommendations, momentum-trading, social-media-driven flows (the GameStop episode in early 2021 is a recent case).

12.5 Bubbles

Robert Shiller (Nobel 2013) — Irrational Exuberance (2000) — documented dot-com and US housing bubbles with cyclically-adjusted price-earnings (CAPE) ratios. His broader thesis: asset prices are more volatile than fundamentals can justify (Shiller 1981 variance-bounds test) — direct evidence against the efficient-market hypothesis in its strong form.

12.6 Limits to arbitrage

Shleifer & Vishny (1997) “The Limits of Arbitrage”: even if some traders are rational, mispricings can persist because arbitrageurs face capital constraints, agency conflicts (clients withdraw funds after losses, forcing liquidation at the worst time — noise-trader risk), and short-sale constraints. The Long-Term Capital Management collapse (1998), the dot-com run-up (1999-2000), and quant-fund losses (August 2007) are textbook cases.

12.7 Behavioral asset-pricing models

• De Long, Shleifer, Summers & Waldmann (1990) “Noise Trader Risk in Financial Markets” — sentiment-driven price deviations and persistent excess volatility.
• Barberis, Shleifer & Vishny (1998) — under/overreaction model based on representativeness and conservatism.
• Hong & Stein (1999) — gradual information diffusion plus momentum-trading newswatchers.
• Daniel, Hirshleifer & Subrahmanyam (1998) — overconfidence and self-attribution bias.

---

13. Critiques and the replication crisis

13.1 WEIRD samples

Henrich, Heine & Norenzayan (2010) “The weirdest people in the world?” noted that most psychology experiments draw from Western, Educated, Industrialized, Rich, Democratic samples — about 5% of the global population — and yet generalize claims to “human nature.” Their cross-cultural ultimatum-game data show large variation in fairness norms across societies.

13.2 The 2015 reproducibility crisis

The Open Science Collaboration (2015) replicated 100 prominent psychology studies; only ≈ 36% reproduced with significant effects in the same direction. Effect sizes were on average half the original. The crisis has hit social psychology hardest and several behavioral-economics-adjacent findings:

• Ego depletion (Baumeister 1998) — willpower as a depletable resource — failed a 23-lab pre-registered replication (Hagger et al. 2016).
• Social priming — e.g. “elderly” words slowing walking (Bargh 1996) — has largely failed to replicate.
• Power-pose effects (Carney, Cuddy & Yap 2010) on hormones did not replicate (Ranehill et al. 2015).

Behavioral economics proper has fared better — prospect theory, loss aversion, default effects, and ultimatum-game patterns replicate consistently — but the field has tightened standards: pre-registration, multi-lab consortia, registered reports, and open data are now routine.

13.3 Methodological reforms

• Pre-registration of hypotheses and analysis plans (e.g. AsPredicted, OSF).
• Many Labs replication consortia.
• Effect-size focus over p-values (the American Statistical Association statement on p-values, 2016).
• Higher significance thresholds — Benjamin et al. (2018) proposed α = 0.005 for “discovery”; 0.05 for “suggestive.”

---

14. Policy applications

14.1 The UK Behavioural Insights Team (BIT)

The “Nudge Unit,” founded 2010 inside the UK Cabinet Office under David Halpern, has run hundreds of RCTs. Notable findings:

• Tax compliance letters: telling delinquent UK taxpayers that “9 out of 10 people in your area have already paid” raised payment rates by ≈ 5 pp.
• Court fine collection: text-message reminders cut bailiff dispatch by ≈ 30%.
• Organ donation prompts at the moment of driver’s-license renewal.

BIT has since spun out into a social-purpose company and seeded similar units worldwide.

14.2 US Social and Behavioral Sciences Team / OEA

Established under the Obama administration (Executive Order 13707, 2015); produced annual reports on behavioral interventions across federal agencies. Now lives in the Office of Evaluation Sciences (OEA) at the GSA.

14.3 Australia’s BETA

The Behavioural Economics Team of the Australian Government runs analogous interventions, with strong work in superannuation, organ donation, and small-business compliance.

14.4 Retirement savings

• Madrian-Shea 2001 auto-enrollment.
• Save More Tomorrow (Thaler & Benartzi 2004).
• UK pensions auto-enrollment (2012 onward): participation rose from ≈ 55% to > 88% of eligible workers within a decade.

14.5 Health and tax

• Organ-donation defaults (Johnson & Goldstein 2003): the single highest-effect-size behavioral policy intervention on record.
• Smoking and obesity: graphic warning labels, default portion sizing, sugar taxes (combining behavioral and Pigouvian rationales).
• Tax compliance (HMRC, IRS): social-norm letters, simplified forms, salience of penalties.

14.6 EAST and the practitioner framework

BIT’s EAST framework distills the practitioner advice:

• Easy: reduce friction, use defaults, simplify language.
• Attractive: personalize, use imagery, draw attention.
• Social: leverage norms, networks, reciprocity.
• Timely: prompt at moments of receptivity (birthdays, life events, just before a decision).

14.7 Consumer financial protection

The US Consumer Financial Protection Bureau (founded 2010 post-Dodd-Frank) embeds behavioral analysis in regulation:

• Mortgage disclosure simplification (Know Before You Owe).
• Overdraft opt-in requirement (Regulation E amendments).
• Credit card CARD Act 2009: prominent disclosures of payoff time and total interest if minimum payments are made — directly inspired by mental accounting and hyperbolic discounting research.

---

15. Critique of nudges

The nudge era has not been uncriticized.

• Effect sizes are modest: the largest meta-analyses (DellaVigna & Linos 2022; Mertens et al. 2022) found average effect sizes around d ≈ 0.04–0.08 — far smaller than initially celebrated — and significant publication bias.
• Scaling problems: laboratory nudges often shrink dramatically when rolled out at population scale.
• Ethical concerns: even libertarian paternalism manipulates the choice environment; transparency is non-trivial; revealed-preference welfare criteria are unclear when preferences are themselves shaped by the architecture.
• Sludge audits are now part of the standard policy toolkit (Sunstein 2022 Sludge) — recognition that bad choice architecture often does as much damage as good architecture can repair.
• Structural critique (e.g. Chater & Loewenstein 2022, “The i-frame and the s-frame”): nudges focus on individual-level interventions and may distract from systemic reform (regulation, taxation, antitrust).

---

16. Behavioral macroeconomics and behavioral game theory

16.1 Animal spirits and macro applications

Akerlof & Shiller’s Animal Spirits (2009) argues that confidence, fairness, money illusion, corruption, and stories drive macro fluctuations in ways standard DSGE models cannot capture. Gabaix (2020) built a behavioral New Keynesian model in which agents have bounded attention to macro variables, generating realistic dampening of monetary-policy effects, hump-shaped impulse responses, and resolution of the forward-guidance puzzle.

16.2 Behavioral game theory

Camerer’s Behavioral Game Theory (2003) integrates social preferences and bounded rationality into strategic settings:

• Level-k / cognitive hierarchy (Stahl & Wilson 1995; Camerer, Ho & Chong 2004): level-0 players randomize; level-k players best-respond to level-(k-1). Empirical median is around level-1 or level-2 — a far cry from infinite-depth common knowledge of rationality. The beauty contest game (“guess 2/3 of the average”) is the canonical demonstration.
• Quantal response equilibrium (McKelvey & Palfrey 1995, 1998): players choose with probability proportional to exp(λ·u), giving a noisy best-response equilibrium that nests Nash as λ → ∞ and uniform random play as λ → 0.

(See game-theory §16 for cross-references.)

17. Open questions and frontiers

• Welfare without revealed preference: if preferences are reference-dependent and frame-sensitive, what is the right welfare criterion? Bernheim & Rangel (2009) propose libertarian welfare with multiple reference frames.
• Machine learning meets behavioral economics: predicting choice from large datasets (Mullainathan & Spiess 2017), detecting biases at scale, personalized nudges, algorithmic recommender systems that exploit hyperbolic discounting (autoplay, infinite scroll).
• Behavioral industrial organization: firms exploit consumer biases (drip pricing, default subscriptions, hyperbolic loyalty programs); regulation must anticipate this. Gabaix-Laibson “shrouded attributes” model is the workhorse.
• Cross-cultural and developmental: when do biases emerge in children? How do they vary across cultures and economic conditions? Scarcity (Mullainathan & Shafir 2013) argues that the cognitive bandwidth tax of poverty can produce 13-IQ-point swings — comparable to a night without sleep.
• Neuroeconomics: fMRI and neural correlates of loss aversion (Tom et al. 2007 — amygdala asymmetry), temporal discounting (McClure et al. 2004 — limbic vs prefrontal trade-offs), and trust (oxytocin studies, Zak et al.). A neural-circuits foundation for the behavioral findings, though replication concerns apply here too.
• Generative AI as choice architect: large language models are emerging as personalized nudgers, both for good (compliance assistants, therapy chatbots) and ill (dark-pattern automation). The line between empowerment and manipulation gets thinner.

Behavioral economics began as a critique and is now a mature, methodologically rigorous, policy-engaged discipline. Its central insight — that human decision-making is systematically non-classical, and that this matters at policy scale — has reshaped economics, finance, marketing, public health, and design.

18. Glossary of key terms

• Allais paradox: a pair of lottery choices that systematically violates the independence axiom.
• Anchoring: numerical estimates pulled toward an irrelevant initial value.
• Availability heuristic: probability judged by ease of recall.
• Choice architecture: the design of the environment in which decisions are made.
• Dictator game: one-sided allocation game; tests pure altruism without strategic component.
• Endowment effect: WTA > WTP gap arising from loss aversion over possessions.
• Framing effect: preferences depend on how options are described.
• Hyperbolic discounting: present-biased intertemporal preferences inconsistent with exponential discounting.
• Loss aversion: losses loom roughly 2.25× as large as equivalent gains.
• Mental accounting: psychological compartmentalization of money violating fungibility.
• Nudge: a change in choice architecture that alters behavior without restricting options.
• Prospect theory: K&T 1979 model with reference-dependence, loss aversion, and probability weighting.
• Reference dependence: utility evaluated relative to a status quo or expectation.
• Sludge: friction in choice architecture that impedes welfare-enhancing action.
• Status-quo bias: tendency to favor the current state regardless of utility.
• System 1 / System 2: dual-process labels for fast-automatic vs slow-deliberative cognition.
• Ultimatum game: take-it-or-leave-it allocation game; tests fairness and inequity aversion.
• WEIRD: Western, Educated, Industrialized, Rich, Democratic sample bias in psychology research.

---

Adjacent

• microeconomics-foundations — utility theory, consumer choice, the rational benchmark behavioral economics critiques.
• macroeconomics-foundations — animal spirits, behavioral New Keynesian models, expectations formation.
• game-theory — ultimatum and dictator games, social preferences in strategic settings, bounded-rationality solution concepts.
• portfolio-theory — equity premium puzzle, disposition effect, behavioral asset pricing.
• probability-foundations — Bayesian reasoning, base rates, the conjunction rule that the Linda problem violates.
• statistics-and-inference — replication crisis methodology, pre-registration, effect-size estimation.