Behavioural Economics — Notes

Heuristics, Biases and Risk Preference

1. Cognitive Psychology and Heuristics

Cognitive psychology is the science of representing and processing information in the human mind. Heuristics are informal mental shortcuts – “rules of thumb” – that reduce the cognitive load of complex decisions. Kahneman’s dual‑system thinking frames heuristics as System 1 (reflexive, fast) thinking, as opposed to System 2 (deliberate, analytical).

Why do we rely on heuristics?

They speed up cognition – decisions are made quickly.
They lower cognitive load – the brain conserves energy.

The world grows increasingly complex, yet the human mind has not evolved appreciably in 10 000 years. Hence we are forced to use fast, reflexive thinking more often. The downside: heuristics can produce systematic errors called biases.

Key terms: Cognitive psychology, heuristics, dual‑system thinking (System 1 vs System 2), biases.

2. The Four Key Judgment Heuristics (Tversky & Kahneman)

Representativeness – judging probability by how similar something is to a mental model.
Availability – judging frequency by the ease with which examples come to mind.
Anchoring – relying too heavily on an initial piece of information (the “anchor”).
Affect – making decisions based on emotional reactions.

The remainder of this sub‑section focuses on Representativeness and its associated biases.

3. Representativeness Heuristic

Representativeness is the heuristic a decision‑maker uses when she takes similarity as a proxy for probability. We ask: “Does this case resemble my mental model?” If it does, we assume a high probability – even when logical or statistical reasoning would contradict.

3.1 Conjunction Fallacy (The Linda Problem)

Linda is 31, single, outspoken, very bright. She majored in philosophy. As a student she was deeply concerned with issues of discrimination, social justice, and participated in antinuclear demonstrations.
Which is more probable?
A. Linda is a bank teller.
B. Linda is a bank teller and active in the feminist movement.

Intuition: Most people (89% of undergraduates, and even Stanford PhD students) choose B – the conjunction – because Linda’s description fits the stereotype of a feminist bank teller better than a plain bank teller.

Formal logic: The set of feminist bank tellers is a subset of bank tellers. Therefore:

$P(\text{bank teller and feminist}) \leq P(\text{bank teller})$

The conjunction fallacy occurs when a specific conjunction is judged more probable than a more general condition – a direct conflict between intuitive (System 1) thinking and probability theory.

Exam tip: A subset can never be more likely than the set that contains it. The Linda problem is the classic demonstration of the conjunction fallacy.

3.2 Insensitivity to Base Rates

Base‑rate neglect – ignoring the prior probability of a category when evaluating a specific case.

Example 1 – Earthquake:

A. A massive earthquake somewhere in India next year killing >1000 people.
B. A massive earthquake in Uttarakhand next year causing avalanches, killing >1000 people.

Most people judge B more likely, even though Uttarakhand is a subset of India. The added detail (“causing avalanches”) makes the scenario more representative of an Uttarakhand earthquake, so it feels more plausible – but:

$P(\text{earthquake in India}) \geq P(\text{earthquake in Uttarakhand})$

Example 2 – Dick the engineer/lawyer:

Treatment	Base rate	Description of Dick	Estimated chance Dick is an engineer
1	70 engineers / 30 lawyers	“Mathematically inclined, likes puzzles” (stereotypical engineer)	~90%
2	30 engineers / 70 lawyers	Same description	~90%
3	70/30 or 30/70	Neutral description (nothing diagnostic)	~50%

In treatments 1 and 2, participants ignore the completely different base rates and rely on the representativeness of the description. In treatment 3, they also ignore base rates (should be 70% or 30%), reverting to an uninformed 50‑50 guess.

3.3 Insensitivity to Sample Size

People fail to account for the fact that smaller samples produce more variable outcomes.

Hospital problem:

Large hospital: 45 births/day; Small hospital: 15 births/day.
For a year, each recorded days with >60% boys. Which hospital recorded more such days?
A. Larger B. Smaller C. About the same

Most choose C. The correct answer is B – the smaller hospital. Reason: the variance of the sample proportion is larger when $n$ is small:

$\text{Var}(\hat{p}) = \frac{p(1-p)}{n}$

Worked analogy: Compare the chance of ≥60% heads in coin flips.

Number of flips ( $n$ )	Probability of ≥60% heads
10	≈ 17%
3000	≈ 0.000001%

In a small sample, large deviations from the population parameter are far more likely. Because people rely on representativeness – 60% feels “random” regardless of sample size – they ignore sample size.

Real‑world application: Ad agencies use “4 out of 5 dentists recommend…” knowing consumers are insensitive to the (often tiny) sample size.

3.4 Misconceptions of Chance

A mistaken belief that random sequences must “look random” (i.e., have no runs), leading to the gambler’s fallacy: the expectation that after a streak of one outcome, the opposite outcome is due.

Coin toss examples:

After 5 consecutive tails, people predict a head on the 6th toss. (Correct: each toss is independent; $P(H)=0.5$ .)
Which sequence is more likely?
$\text{Red: } H\,T\,H\,T\,H\,T$
$\text{Blue: } H\,H\,H\,H\,H\,T$
Most choose red, because it “looks random”. In truth, both sequences are equally likely ( $(0.5)^6$ ).

Again, the root is representativeness: the red sequence resembles a typical random outcome, so it is judged more probable.

Consequential real‑world applications:

Fertility decisions: A household with two daughters, desiring a son, may try for another child because they believe a boy is “due”. Seema Jayachandran and Rohini Pande show this affects child health and sex ratios, contributing to Amartya Sen’s “missing women” in South Asia. The mathematician Laplace noted the same fallacy among French parents 150 years ago.
Gambler’s fallacy: After ten losing poker hands, a player believes a win is imminent. After winning a lottery, a player changes their “lucky numbers” because they think the same numbers are unlikely to come up again – ignoring that the lottery draws with replacement.
Sports commentary: “The law of averages will catch up” – e.g., a cricketer who has scored a 50 in four consecutive matches is said to be less likely to do so in the fifth.

Key Takeaways

Heuristics are System 1 shortcuts (fast, low effort); biases are mistakes that arise from them.
Representativeness judges probability by similarity: “seems like” = “is likely”.
Conjunction fallacy: $P(A \cap B) \leq P(A)$ , but representativeness can violate this (Linda problem).
Base‑rate neglect: detailed, representative scenarios cause people to ignore prior probabilities (Dick example).
Sample‑size neglect: small samples are more variable; people treat all sample sizes as equally informative (hospital problem).
Misconception of chance leads to the gambler’s fallacy – expecting reversals after streaks – with serious consequences in gambling, fertility decisions, and sports commentary.

Exam tip: For any question involving probability judgments, ask: Is the person being fooled by representativeness? Look for “similarity” overriding logical or statistical reasoning – that is the signature of this family of biases.

Regression to the Mean

Regression to the mean is the statistical phenomenon that after an extreme outcome, the next outcome tends to be closer to the average. Intuitively: when performance depends partly on luck, the best and worst performances are often flukes — and subsequent attempts will "regress" toward the person’s true ability level.

Kahneman’s flight-trainer story illustrates how easily regression is misinterpreted as causation. Flight instructors observed that:

Praise after an exceptionally smooth landing was typically followed by a poorer landing.
Harsh criticism after a rough landing was typically followed by improvement.

The instructor’s conclusion: punishment works, praise doesn’t. Kahneman recognised this as regression to the mean. Landing a fighter jet is highly variable. Those who performed extremely well were simply lucky; the next attempt was likely to be closer to their average (worse). Those who performed poorly were unlucky; the next attempt was likely to be closer to their average (better). The instructor’s praise or punishment had nothing to do with the change.

flowchart LR
    A[Extreme high performance<br/>(luck)] --> B[Next attempt<br/>closer to mean<br/>worse]
    C[Extreme low performance<br/>(bad luck)] --> D[Next attempt<br/>closer to mean<br/>better]

Common examples of regression to the mean

Domain	“Rule”	Explanation
Sports	Great rookies have less impressive second seasons (sophomore jinx)	Beginners often have lucky streaks; skill level regresses
Family	Gifted children have less successful siblings	One child’s extreme talent partly from random genetic or environmental factors; siblings tend toward average
Business	Firms with outstanding profits one year do less well the next	Extraordinary results often include a strong luck component
Everyday	“Beginner’s luck” → poor subsequent performance	Initial success was above true ability; next outcome regresses

Exam tip: The key insight is that the instructor assigned a causal interpretation to a statistical pattern. Any time you see praise or punishment followed by a change, check whether regression to the mean is the real explanation.

Relation to representativeness heuristic

The representativeness heuristic assumes future outcomes are directly predictable from past outcomes (perfect correlation). Regression to the mean shows that extreme past outcomes are often not representative of future ones — especially in early stages of a process, outcomes tend to shift toward the mean over time.

Key takeaways

Regression to the mean: extreme outcomes are likely followed by outcomes closer to the average.
It often appears as a causal effect when it is purely statistical.
Common in sports, business, education, and any domain with variable performance.
Confusing it with cause-effect is a classic bias.

Conjunction Fallacy

The conjunction fallacy occurs when people judge the joint occurrence of two events (A and B) as more probable than one of the individual events alone — violating the laws of probability. Intuitively: a specific, detailed story can feel more representative and therefore more likely than a generic one, even though the generic one must be at least as probable.

In a rational world, for any two events A and B: $\Pr(A \cap B) \leq \Pr(A) \quad\text{and}\quad \Pr(A \cap B) \leq \Pr(B)$

The Linda problem demonstrates this systematically. Subjects were given a description: “Linda is 31 years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was very concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.”

Then asked to rank the probability of several statements. Two key options:

Linda is a bank teller.
Linda is a bank teller and active in the feminist movement.

The second (conjunction) was judged more likely than the first by a large majority — even though it is a subset of the first. The conjunction “feminist bank teller” appears more representative of the description than “bank teller” alone.

Key takeaways

Conjunction fallacy: $\Pr(A \cap B) > \Pr(A)$ is impossible, but people commit it when the conjunction is more representative.
Driven by the representativeness heuristic.
Classic demonstration: the Linda problem.

Confirmation Bias

Confirmation bias is the tendency to seek, interpret, and remember evidence in a way that confirms one’s pre-existing beliefs. Intuitively: once you decide something, you look for reasons you are right — not reasons you might be wrong.

The car-buying contrast

Scientific approach: Gather reasons, then decide.
Confirmation bias approach: Decide first, then find reasons to justify that decision.

Lord, Ross & Lepper’s death-penalty experiment

Subjects were given two sets of studies — one supporting, one opposing the effectiveness of capital punishment. Both proponents and opponents of the death penalty evaluated the evidence:

Proponents found the anti-death-penalty evidence methodologically flawed and unpersuasive.
Opponents found the pro-death-penalty evidence methodologically flawed and the anti-penalty evidence persuasive.

After reading, both groups became more convinced of their original positions.

Why does confirmation bias exist?

The mental cost of processing a contrarian viewpoint is higher than processing a confirmatory one. Fischer et al. (2008) showed that mentally fatigued individuals were more likely to choose articles consistent with their viewpoint.

Yes Man vs. Devil’s Advocate (Thaler)

Role	Message	Likely survival	Value to firm
Yes Man	“Here are 10 reasons your idea is terrific.”	More likely to survive — managers with confirmation bias prefer confirming feedback.	Low — reinforces flawed ideas.
Devil’s Advocate	“Here are 10 reasons your idea is terrible.”	Less likely to survive — contrarian feedback is mentally costly and uncomfortable.	High — exposes flaws, improves decisions.

Exam tip: The Yes-Man/Devil’s-Advocate tradeoff is a classic application. Think about real organisations: do their internal cultures reward honest critics or loyal cheerleaders?

Key takeaways

Confirmation bias: seeking and favouring evidence that supports existing beliefs.
Mentally easier to process confirming than disconfirming information.
Fuels polarisation and poor decision-making in organisations.
Devil’s advocates may add more value but struggle to survive when leaders suffer from confirmation bias.

Availability Heuristic

Availability heuristic is the tendency to judge the probability or frequency of an event by the ease with which instances come to mind. Easily recalled events feel more common, even if they are not.

Evidence from experiments

Causes of death ranking: Participants rank war and starvation as leading causes of death, but WHO data shows cancer, lung diseases, and respiratory infections are actually more common. The vivid, media‑heavy coverage of war and drought makes them retrievable → overestimated.
Famous names list: A list with equal numbers of men and women, but one version contains more famous men, the other more famous women. Participants rate whichever gender had the more famous names as more numerous – because famous names are easier to recall.

Salience

Salience – the property of drawing attention – drives availability. Something becomes salient when it is:

Familiar
Important to the observer
Part of personal experience
Recent

Performance evaluation bias: Managers evaluating employees tend to give disproportionate weight to vivid, recent incidents (positive or negative) because those are easiest to recall. An employee who “goofed up in August” may be penalised heavily in an October review, even if overall performance was good.

Retrievability bias

A classic demonstration: Are there more English words with r as the first letter (e.g., road) or as the third letter (e.g., car)? Most people say first‑letter words are more numerous, but letters like r and k actually appear more often in the third position. The bias arises because our mental search engine retrieves words starting with a given letter far more easily than words with it in the third position.

Exam tip: Availability is easily confused with representativeness. Remember: availability is about ease of recall (influenced by salience, recency, vividness), not about similarity to a prototype.

Key takeaways – Availability

Availability heuristic: frequency/probability judged by ease of recall.
Driven by salience – familiarity, personal experience, recency, vividness.
Leads to overestimating vivid causes of death and underestimating common diseases.
Causes performance evaluation to overweight recent or vivid episodes.
Retrievability bias (e.g., first‑letter vs. third‑letter words) is a direct consequence.

Anchoring Heuristic

Anchoring heuristic is the tendency to make estimates by starting from an initial value (the anchor) and then adjusting insufficiently to reach a final answer.

Famous demonstration: product of 1–8

Participants were asked to quickly estimate the product of $1 \times 2 \times 3 \times \cdots \times 8$ (correct answer = 40,320). One group saw the ascending sequence (1,2,3,…); their average guess was 512. Another group saw the descending sequence (8,7,6,…); their average guess was 2,250. The initial few numbers acted as an anchor, and adjustments were far from the true value.

Multiple anchoring experiments

Question	Anchor 1 (low)	Mean guess 1	Anchor 2 (high)	Mean guess 2
When was the telephone invented?	1850	1870	1920	1900
Height of Mount Everest (feet)	2,000	8,000	45,500	42,550

In every case, the anchor pulled the average guess toward itself, even though anchors were assigned randomly.

Why anchoring works

Lack of information – in ambiguous situations, the anchor provides a natural starting point; adjustments away from it are typically insufficient.
Confirmatory thinking – the anchor activates information that is consistent with it (similar to confirmation bias); people find it easier to think of reasons why the anchor might be right than why it might be wrong.

Real‑world implications

Consumer choice (Mussweiler & Strack): After guessing the average new‑car price in Germany – some were given a high anchor (40,000 DM), others a low anchor (20,000 DM) – participants were asked to name four car brands. The high‑anchor group listed Mercedes, BMW; the low‑anchor group listed Golf, Volkswagen. Exposing consumers to high prices can raise willingness to pay.
MBA salary negotiations: Companies often ask for pre‑MBA salary as an anchor. Students who bargain aggressively create a higher anchor for themselves. Research shows women bargain less than men, leading to a lower starting‑salary anchor – a possible contributor to the gender wage gap.

Exam tip: Anchoring affects even domain experts. The effect is robust and may operate subconsciously. To counter it, deliberately consider an opposite anchor before making a final estimate.

Key takeaways – Anchoring

Anchoring: start with an initial value (anchor), adjust (insufficiently) to a final answer.
Common in numerical estimation tasks (product of numbers, historical dates, heights).
Works because of lack of information and confirmatory search.
Powerful in consumer behaviour – extreme price anchors shift willingness to pay.
Applies to salary negotiations, performance appraisals, and any context with ambiguous values.

Affect Heuristic

Affect heuristic refers to the quick “good” or “bad” emotional response (affect) that arises before higher‑order reasoning engages. Many judgments are actually driven by this gut feeling.

Everyday examples

Recruiter sorting CVs: Positive affect may arise if a candidate shares the same university, subject, or hostel block. Negative affect can come from unrelated features. The emotional tag influences the sorting decision.
Affective polarization: In modern politics, people not only hold opposing views but also actively dislike supporters of the opposite party. This emotional response (affect) colours all subsequent political judgments.

The smoking paradox – from positive to negative affect

1960s – Positive affect: The tobacco industry used the Marlboro Man campaign to associate smoking with masculinity, freedom, and a rugged lifestyle. The product was sold via lifestyle imagery, bypassing risk information. Positive affect made people underestimate the health dangers.
2000s – Negative affect: Policy makers reversed the technique. Cigarette packets now carry gory health‑warning images that evoke disgust and fear. Video campaigns create negative emotional triggers. The goal is to make the product feel bad before any rational cost‑benefit analysis.

The affect heuristic is a core part of System 1 thinking: fast, automatic, emotional. It explains why factual information alone often fails to change behaviour – emotions must be engaged.

Exam tip: Affect is the newest of the four heuristics. It often overlaps with availability (vivid images are both affect‑laden and easy to recall). Examiners may ask you to distinguish them: affect is the emotional evaluation, availability is the ease of recall.

Key takeaways – Affect

Affect heuristic: judgments driven by an immediate emotional “good/bad” response.
Occurs automatically, before reasoning (System 1).
Examples: recruiter bias from shared background, affective political polarization.
Demonstrated powerfully in smoking – positive affect in Marlboro ads reduced perceived risk; negative affect in health warnings increased perceived risk.
Policy makers now intentionally use negative affect to counteract harmful behaviours.

Overconfidence

Overconfidence – called the "mother of all biases" – is the systematic tendency for a person's subjective confidence in their judgments to exceed the objective accuracy of those judgments. It is considered the most robust finding in the psychology of judgment (Taylor). Its consequences are severe: wars, stock-market bubbles, strikes, unnecessary lawsuits, and major accidents (Challenger, Chernobyl) have all been linked to overconfidence.

Three distinct forms of overconfidence are distinguished:

Form	Core question	Definition
Overprecision	"Are we too sure about our judgments?"	Confidence intervals are too narrow relative to the true distribution of outcomes.
Overestimation	"Are we better, smarter, more attractive than we actually are?"	Self-assessment exceeds true ability or performance.
Overplacement	"Are we better, smarter, more attractive than others?"	Viewing oneself as above average relative to a peer group.

Overprecision

Intuition: When you are 98% sure a number lies inside a range, you should be wrong only 2% of the time. In reality, people are wrong far more often – their "confidence bands" are too tight.

Key studies

Alpert & Raiffa (1982): Participants gave best estimates and 98% confidence intervals for unknown quantities (e.g., Walmart's 2010 revenue, world population in 2012). Of about 100 participants making 10 judgments each, most correctly bracketed only 3–7 of the true values – far fewer than the 98% they claimed. This shows systematic overprecision: people are far surer of their knowledge than they should be.
McKenzie et al. (UCSD): To test whether unfamiliarity drives the effect, students (experts on UCSD campus) and computer programmers (experts on IT industry) were asked questions about both domains. Overprecision persisted in both groups, but was reduced when the context was familiar. Familiarity helps but does not eliminate the bias.
Ben-David et al. (2013, QJE): Over 13,000 business executives were asked for 80% confidence intervals for future stock market returns (10-year panel). The realized return fell inside their 80% bands only 36% of the time. This miscalibration was worst during periods of high market uncertainty and also predicted miscalibration about the executives' own firm performance.

Real-life implications of overprecision

A legal counsel 98% confident of winning will push for litigation instead of settlement.
A marketing plan based on a sales forecast with an overly narrow confidence band will lack a Plan B – leaving the firm exposed.

Debiasing strategies

The "crowd within": Herzog & Hertwig (2009) asked people to generate multiple estimates of the same outcome and then average them. The averaged estimate was more accurate – a form of "wisdom of the crowd" applied within one person.

Exam tip: Actively seeking others' perspectives gives a wider sense of where the truth lies – a practical antidote to overprecision.
Explicitly consider alternative outcomes: Harran et al. showed that asking people to state the chance of failure (e.g., "what is the percentage chance of failure?") after they have stated a 98% chance of success forces recalibration and improves accuracy.

Key takeaways (overprecision)

Overprecision = confidence intervals that are too narrow for the true level of knowledge.
Robustly observed even among experts in familiar domains, though familiarity reduces the magnitude.
Real economic costs: bad legal advice, missing Plan B, misjudged firm performance.
Debiasing: use multiple estimates (crowd within) and explicitly prompt the probability of failure.

Overestimation

Intuition: We tend to think we are better across many domains than we really are – not relative to others, but relative to our own actual ability.

Manifestations

Self-enhancement: Viewing oneself positively rather than accurately – leads to inflated perceptions of own performance, ability, and talent.
Planning fallacy: Believing one will deliver a project on time when one consistently fails to do so. In work teams, this leads to unrealistic deadlines.
Optimism bias: Overestimation of how positive the future will be. Culturally captured in phrases like Akash Kusum (Bengali) and Khayali Pulav (Hindi) – "imaginary rice". Self-help books often reinforce this bias.

Economic consequences of overestimation

Effort: A person who overestimates their ability is less likely to put in effort for an upcoming exam (because they believe they already know enough).
Savings: Overestimation leads to under-saving – the person believes they can generate sufficient resources in the future, so they save too little today.

Exam tip: Note the contrast: overestimation can reduce effort (because you think you don't need to try) while overplacement can increase risk-taking (because you think you are better than rivals). Both distort economic decisions.

Key takeaways (overestimation)

Overestimation = inflated view of one's own absolute ability or future outcomes.
Leads to planning fallacy, self-enhancement, and optimism bias.
Real consequences: under-saving, insufficient effort, unrealistic deadlines.

Overplacement

Intuition: The "better-than-average effect" – most people think they are above average at skills like driving, leading, or social competence.

Classic evidence

93% of American drivers rated themselves as more skillful than the median driver.
CAT (MBA entrance) students: 60% rated themselves in the top 10% of ability to get along with others; 25% placed themselves in the top 1% – a statistical impossibility.

Real-world consequences

Litigation: Thompson & Loewenstein (1992) – because parties overplace their chances of winning, they fight too hard, hold on too long, and overpay lawyers.
Entrepreneurship: Entrepreneurs who believe they are more capable than competitors enter new markets even when objective success chances are low.
Mergers & Acquisitions: Malmendier & Tate found that CEOs who overplace their ability relative to other managers pursue M&A deals that destroy shareholder value – decisions driven by overplacement, not rational calculation.

Key takeaways (overplacement)

Overplacement = the "better-than-average" effect.
Ubiquitous in self-reports of skill, driving, academic ability.
Drives excessive litigation, inefficient market entry, and value-destroying M&A.

Summary: Overconfidence – the three forms

flowchart LR
  A[Overconfidence] --> B[Overprecision]
  A --> C[Overestimation]
  A --> D[Overplacement]
  B --> E[Confidence intervals too narrow]
  C --> F[Actually better than true ability]
  D --> G[Better than others]

Overarching key takeaways

Overconfidence is pervasive and has severe real-world consequences across wars, markets, litigation, and corporate decisions.
The three forms (overprecision, overestimation, overplacement) are distinct but often co-occur.
Debiasing tools exist: generating multiple estimates, explicitly considering failure probabilities, and seeking external perspectives.
Overprecision persists even among experts and in familiar domains – it is not merely a lack of knowledge.
Overestimation can reduce effort and savings; overplacement fuels excessive competition and risk-taking.

Definition and Intuition

Exponential growth bias (EGB) is the pervasive tendency to linearize exponential functions when assessing them intuitively. People mentally replace the true exponential curve with a straight line, leading to a systematic underestimation of future values given a present value, or overestimation of present values given a future value.

EGB is not a simple arithmetic mistake — it reflects a deep cognitive failure to grasp how repeated multiplication (compounding) behaves. In daily life it explains why milk suddenly boils over, why cities like Bangalore gridlock, and why water hyacinth can choke a pond overnight: the growth rate appears slow at first, then accelerates faster than intuition can track.

Classic Example: Sissa and the Chessboard

The story of Sissa ibn Dahir (legendary inventor of chaturanga, a precursor to chess) captures EGB perfectly. Asked for a reward, Sissa requested:
1 grain on the first square, 2 on the second, 4 on the third, 8 on the fourth … doubling each time. The king, thinking only of the early small numbers, agreed. By the 64th square the required grains totalled $2^{63}$ — more than the entire empire’s wealth. The king’s intuition linearised the doubling; EGB made him bankrupt.

Formal representation:
Number of grains on square $n$ : $2^{(n-1)}$ .
Total after 64 squares: $\sum_{i=0}^{63} 2^{i} = 2^{64} - 1$ — astronomically large.

Consequences of EGB

Domain	Effect of EGB	Mechanism
Borrowing	Increased borrowing (Stango & Zinman)	Underestimation of future debt → present borrowing feels cheaper than it is
Savings	Reduced precautionary saving (Levy & Tasoff, 2015)	Lower perceived future debt → higher current consumption
Risk perception	Lower perceived future risk	Underestimation of exponential growth (e.g., cases, pollution) → less precautionary action

The underlying causal chain:

flowchart LR
  A[EGB] --> B[Underestimation of future values]
  B --> C[lower perceived risk / future burden]
  C --> D[More borrowing, less saving, less compliance]

Exam tip: EGB is a cornerstone behaviour that links financial decisions (borrowing/saving) to health decisions (compliance). The chessboard story is a classic exam starter.

Key takeaways

EGB = tendency to linearise exponential growth → systematic underestimation.
Drives over-borrowing, under-saving, and poor risk anticipation.
Evident in everyday phenomena (boiling milk, traffic) and formal puzzles (chessboard).
Foundational for understanding how people misjudge compound interest and epidemic spread.

Study 1: EGB and COVID-19 Compliance

Design. Participants saw the confirmed COVID-19 case count for weeks 1, 2, and 3 (from Germany, USA, France, Spain) during both early and later phases of the pandemic. They predicted the total cases for weeks 4 and 5. Incentive: small payment if prediction was within 5% of the actual number. After prediction, participants completed a survey measuring three compliance indices:

Actual realised compliance
Appropriateness of norm violation
Agreeableness with government’s performance

Results.

The median prediction for week 4 and week 5 was significantly lower than the actual number.
A bias measure was computed:
$bias_{t|3} = \text{prediction error relative to week 3 (log-scale used)}$
Averaged across weeks 4 and 5, the bias was positive and significantly different from zero — confirming EGB.
EGB was negatively correlated with compliance behaviour: participants with higher EGB were less likely to wear masks, social distance, etc., and were more likely to view norm violations as acceptable.

Interpretation. EGB reduces perceived future risk → lower motivation to comply with public health measures.

Study 2: Reducing EGB with Behavioural Nudges

This follow-up experiment tested whether simple, behaviourally informed interventions could reduce or eliminate EGB and thereby adjust economic expectations (optimism about future macroeconomy).

Experimental design.

Treatment	Key feature	What participants predicted
Baseline	Standard prediction task	Cases on Day 35 given Days 0, 5, 10
Step	Predict sequentially for Days 15, 20, 25, 30, 35	Same data, repeated predictions
Step + Feedback (numbers)	After each step, receive numeric prediction error	Same as Step
Step + Feedback (graph)	After each step, receive graphical feedback of error	Same as Step
Step + Forecast	Before predicting, shown a statistical-model forecast range	Same as Step

All treatments were identical in stage 2: participants invested a hypothetical amount in a stock market of the fictitious country, with returns based on the actual Day‑35 market performance.

Outcome variables.

Bias (prediction error) and EGB (derived from bias)
Economic expectations (measured via RBI’s current situation index, future expectation index, and Michigan Consumer Survey items)

Key results.

EGB present. Baseline predictions were far from actual — large bias.
Step reduced EGB; Feedback (both number and graph) and Forecast virtually eliminated EGB (bias not significantly different from zero).
Economic expectations were lower (more realistic / less optimistic) in Step, Feedback, and Forecast treatments compared to Baseline.

Caveat for comparison. Only Day‑35 predictions are comparable across all treatments because Baseline lacks intermediate predictions. For comparisons among Step‑based treatments, the average bias over Days 15–35 can be used.

Exam tip: Behavioural nudges that force step‑by‑step prediction or provide feedback (especially graphical) can correct EGB. This is a key example of “debiasing” through choice architecture.

Key takeaways

EGB is robust (observed in COVID‑19 case predictions across countries).
Step‑by‑step prediction reduces EGB; adding feedback or a model forecast can eliminate it.
Reducing EGB makes economic expectations less optimistic and improves compliance intentions.
Behaviourally informed policy design: simple changes in how information is presented can correct systematic bias.

Hindsight Bias

Hindsight bias is the tendency to perceive past events as having been more predictable than they actually were before they occurred. In plain language: after something happens, people readily believe “I knew it all along.” This bias distorts how we evaluate decisions and limits genuine learning.

Classic examples from everyday life

Financial crisis 2008 — Many policymakers later claimed they “knew the crisis was inevitable,” yet the majority of macroeconomists did not see it coming. The question: was it truly knowable?
Cricket, 1986 Sharjah Cup final — India vs. Pakistan; last ball, four runs needed. Captain gave the ball to Chetan Sharma; Javed Miandad hit a six. Retrospectively, “everyone” knew Chetan Sharma should not have bowled that ball.
Driving with a spouse — At an unmarked fork your spouse turns right, you end up on the wrong route. You say, “I knew you should have turned left.” Did you really know?
Hiring disagreement (Ali and Rani) — After Shyam’s poor performance, Ali says “there was a lot of evidence Shyam would be lousy,” though he had previously agreed to go with Rani’s view.
Marketing presentation — A senior VP says, “I could have told you these results,” dismissing a six‑month consumer study.

Evidence: Fischhoff (1975)

Fischhoff experimentally demonstrated the gap between hindsight and foresight.

Participants read a passage about the British vs. Gurkha war of 1814.
They were divided into five groups, each given a different “result” of the war:
- British won
- Gurkhas won
- Military stalemate with no peace settlement
- Military stalemate with peace settlement
- Baseline: result not shared
All groups were then asked to assess the probability of each outcome without the benefit of that knowledge.
Result: Participants in every group claimed they would have reached the same conclusion regardless of which outcome they were shown. In a rational world, this is impossible — it’s driven by hindsight bias.

Why it matters: pros and cons

Upsides	Downsides
Flattering — makes you feel your judgment is better than it is	Limits learning from the past — if you think you predicted it, you won’t analyze what went wrong
Allows you to criticize others’ lack of foresight	Eliminates objective evaluation of decisions — e.g., a low‑risk surgery that goes wrong leads a jury to believe the operation was obviously risky, making it impossible to judge the pre‑surgery decision fairly

Exam tip: Hindsight bias is often tested with the Fischhoff experiment. Remember the key finding: different groups given different “outcomes” still believed they would have predicted that same outcome. The bias arises partly from imperfect recall – we misremember our earlier predictions.

Key takeaways

Hindsight bias = “I knew it all along” phenomenon; events seem more predictable after the fact.
It limits learning and prevents fair evaluation of past decisions.
Fischhoff’s war study is the classic empirical demonstration.
The bias is flattering but harmful for decision‑making.

Curse of Knowledge

Curse of knowledge is the difficulty of un knowing something once you know it — and the resulting inability to imagine how others, who lack that knowledge, perceive a situation. It causes experts to overestimate what novices understand.

Everyday examples

Fresh PhDs becoming assistant professors struggle to explain concepts simply to undergraduates — they cannot “unlearn” five years of deep expertise.
Product designers often include advanced features that the average user finds too complex, underestimating the learning curve.

Empirical evidence: Kysar (cognitive psychology)

Participants read: “David had dinner at a restaurant based on a friend’s recommendation.”

Treatment 1: David enjoyed the meal.
Treatment 2: David disliked the meal.
Both groups then read: “David wrote to his friend: about the restaurant it was marvellous just marvellous.”
Question: Would the friend read the comment as sincere or sarcastic?
Result:
- Treatment 1 (David enjoyed) → “sincere”
- Treatment 2 (David disliked) → “sarcastic”
Why is this interesting? Because David’s friend did not know how David’s meal went — but the participants did. They could not ignore that information when predicting the friend’s interpretation. This is the curse of knowledge in action.

Organizational implications

A great deal of disappointment in organisations arises from poor communication — we mistakenly believe that our ambiguous messages are clear to others. A known remedy is to adopt a mindset of focusing on differences (between people and objects) rather than similarities. People who focus on differences are better at taking others’ visual perspective and less likely to project their own private information onto others.

Key takeaways

Curse of knowledge = inability to forget what you know when judging what others know.
It leads to over‑complex products, poor teaching, and communication failures.
Kysar’s restaurant study shows how knowledge of an outcome biases interpretation of a message.
A practical fix: focus on differences, not similarities, to reduce projection.

Changes vs. Levels

Intuition: Standard neoclassical economics says satisfaction depends on absolute levels — the number of mangoes you eat or the money you earn. But real human happiness responds to changes from a reference point, not just final levels.

Thought experiment: Same final score (95/100) from (a) 95 on both midterm and final, or (b) 65 on midterm and 95 on final. Most people feel happier in (b) — the change matters.
Ali vs. Rani: Ali (poor, ₹500/month) suddenly earns ₹1 crore; Rani (rich, ₹1 crore) earns ₹500. The final income is nearly the same, but Ali’s happiness is far greater. Humans compare to a reference point; “econs” (rational agents) do not.
Perception parallel: Hot water / cold water experiment — left hand in hot, right hand in cold, then both in lukewarm. Sensation depends on the change from prior state.
Pain management: Cancer patients are told to accept pain as the “new normal” — the reference point shifts to reduce emotional trauma.

Reference Points

The change is computed from some reference point. Candidates:

Expected outcome (what one anticipates)
Fair outcome / Status quo / Legal entitlement
Average outcome in society
Personal goal / Social hero’s outcome

Key takeaways

People care about changes, not absolute levels.
Satisfaction is measured relative to a reference point.
Reference points are shaped by expectations, fairness, status quo, etc.
This idea underlies the first key principle of prospect theory: changes matter more than levels.

Losses and Gains

The second key principle: losses matter more than gains. In standard utility theory, the pleasure of a gain equals the pain of an equal loss (symmetry). Prospect theory rejects symmetry.

Mug Experiment (Kahneman, Knetsch, Thaler)

Sellers (given a mug) see selling as a loss — reference point = possession.
Buyers (no mug) see buying as a gain — reference point = non-possession.
Results: Willingness to Accept (WTA) ≈ $7, **Willingness to Pay (WTP)** ≈$ 3.
Implication: the value function is steeper in the loss domain than in the gain domain.

Duke Basketball Tickets (Carmon & Ariely)

Lottery-based distribution (eliminates selection effect — those who get tickets aren’t inherently higher-valuing).
Sellers’ WTA: ~ $2,400; buyers’ WTP: ~$ 170.
Massive gap again — losses loom larger than gains.

Loss Aversion & Endowment Effect

Loss aversion: the pain of a loss is about twice the pleasure of an equivalent gain. Researchers estimate the loss aversion parameter (slope ratio of loss domain to gain domain) ≈ 2.

Endowment effect: people value an object more once they own it — driven by loss aversion.

Key takeaways

Losses hurt more than gains please (asymmetry).
Value function: flatter for gains, steeper for losses.
Loss aversion parameter ≈ 2 (empirical estimate).
The endowment effect (e.g., mug, tickets) is a direct consequence of loss aversion.

Exam tip: The mug experiment and Duke tickets are classic demonstrations of endowment effect. Be able to explain why lottery distribution is important (controls for selection bias).

St. Petersburg Paradox

The paradox highlights that people do not simply maximise expected value — they care about expected utility.

The Gamble

Start with $1. Flip a coin.
If heads → double the money and flip again.
If tails → game ends, take current money.
You can play until first tail.

Expected Value

$EV = 1 + \frac{1}{2}\times 2 + \frac{1}{4}\times 4 + \frac{1}{8}\times 8 + \cdots = \sum_{k=0}^{\infty} 1 = \infty$

Despite infinite expected value, people pay very little to play.

Resolution via Expected Utility

Assume utility function $U(x) = \log(x)$ . Then:

$EU = \log(1) + \frac{1}{2}\log(2) + \frac{1}{4}\log(4) + \frac{1}{8}\log(8) + \cdots$

This series converges to a finite value (≈ 0.602 utility units). People are risk-averse — concave utility reduces the appeal of the gamble.

Key takeaways

Expected value alone cannot explain risk-taking behaviour.
The St. Petersburg Paradox shows why expected utility framework is needed.
Concave utility functions (diminishing marginal utility) make infinite-expected-value gambles unattractive.

Definition

A risk-averse individual prefers a certain income over a risky lottery with the same expected value.

Graphical Explanation

Consider a gamble: 50% chance of ₹10,000, 50% chance of ₹30,000. Expected value = ₹20,000. A risk-averse person with a concave utility function will choose the certain ₹20,000.

Utility at ₹10,000 = 10, at ₹30,000 = 18. Expected utility = $0.5\times 10 + 0.5\times 18 = 14$ .
Utility of certain ₹20,000 = 16.
Since $16 > 14$ , the sure thing is preferred.

For a concave utility function: $U(EV) > E[U(x)]$ (Jensen’s inequality).

Risk Premium

Risk premium = the maximum amount an individual is willing to pay to avoid risk. It is the difference between the expected value of the gamble and the certainty equivalent (the sure amount that gives the same utility as the gamble).

From the graph: certainty equivalent ≈ ₹16,000. Risk premium = ₹20,000 – ₹16,000 = ₹4,000.

Exam tip: Risk premium = $EV - \text{certainty equivalent}$ . Insurance companies price premiums based on this concept.

flowchart LR
    A[Gamble: 50% ₹10k, 50% ₹30k] --> B{Check utility}
    B --> C[U(certain ₹20k) = 16]
    B --> D[EU(gamble) = 14]
    C --> E[Choose certain ₹20k]
    D --> E
    E --> F[Risk premium = ₹4k]

Key takeaways

Risk aversion: prefer sure thing over equally valued risky gamble.
Concave utility → $U(EV) > E[U(x)]$ .
Risk premium = amount paid to avoid risk; equals $EV - \text{certainty equivalent}$ .
Insurance exploits risk premium.

Sensitivity to Gains and Losses

When humans choose between risky gains, they behave differently than when choosing between risky losses. This pattern – risk aversion in the gain domain, risk seeking in the loss domain – is a core finding of prospect theory and cannot be explained by expected utility theory.

Reference Points and Choice Patterns

The key is that people evaluate outcomes relative to a reference point, not as final wealth. The following choice problems (from Kahneman & Tversky) demonstrate the effect.

Problem 1 (Gain frame)

Option A: Get $900 for sure.
Option B: 90% chance of $1000, 10% chance of$ 0.
Most choose A – risk averse.

Problem 2 (Loss frame)

Option A: Lose $900 for sure.
Option B: 90% chance of losing $1000, 10% chance of losing$ 0.
Most choose B – risk seeking.

The same pattern appears when an endowment is added:

Problem 3 (given $1000)	Problem 4 (given $2000)
Option 1: 50% chance win $1000, 50%$ 0	Option 1: 50% chance lose $1000, 50%$ 0
Option 2: $500 for sure	Option 2: Lose $500 for sure
84% choose sure gain (Option 2) → risk averse	69% choose gamble (Option 1) → risk seeking

Preference reversal: The utility inequalities from Problem 3 and Problem 4 are identical except for the inequality sign – a violation of expected utility.

Exam tip: The same expected value can be framed as a gain or a loss. The frame determines whether people are risk averse or risk seeking.

The Value Function (Prospect Theory)

Prospect theory replaces the utility function with a value function $v(x)$ defined over gains and losses (deviations from the reference point). Three principles define its shape:

Changes matter more than levels – The reference point is the status quo; outcomes are coded as gains or losses.
Losses matter more than gains – The value function is steeper in the loss domain; the loss aversion coefficient $\lambda \approx 2$ means the pain of losing $X$ feels about twice the pleasure of gaining $X$ .
Risk averse in gains, risk seeking in losses – $v$ is concave for gains ( $v''(x)<0$ ) and convex for losses ( $v''(x)>0$ ).

Mathematically:

v(x) = \begin{cases} x^\alpha & x \ge 0 \\ -\lambda (-x)^\beta & x < 0 \end{cases}

with $0<\alpha,\beta<1$ and $\lambda>1$ (typically $\lambda \approx 2$ ).

Framing: The Asian/American Disease Problem

Problem 1 (lives saved – gain domain)

Program A: 200 people saved for sure.
Program B: 1/3 chance 600 saved, 2/3 chance 0 saved.
Most choose A – risk averse.

Problem 2 (lives lost – loss domain)

Program C: 600 people die for sure.
Program D: 1/3 chance 0 die, 2/3 chance 600 die.
Most choose D – risk seeking.

The outcomes are identical: A saves 200 (C lets 400 die); B has expected 200 saved (D has expected 400 die). The reference point shifts: 0 lives saved in problem 1 → gain domain; 600 lives saved (i.e., 0 lost) in problem 2 → loss domain.

Reference point (gain domain): 0 saved
   v(200 saved) > expected v of gamble → choose A

Reference point (loss domain): 600 saved (= 0 lost)
   v(400 die) < expected v of gamble → choose D

The value function’s concavity (gains) and convexity (losses) directly predicts this preference reversal.

Key Takeaways (Sensitivity to Gains and Losses)

People are risk averse in the gain domain and risk seeking in the loss domain.
The reference point determines whether an outcome is coded as a gain or loss.
The value function is concave for gains, convex for losses, and steeper for losses.
Loss aversion coefficient $\lambda \approx 2$ : losses hurt roughly twice as much as gains please.
Framing (e.g., “lives saved” vs “lives lost”) reverses choices even when expected outcomes are identical.
These patterns are anomalies for expected utility theory.

Risk Perception

How people perceive probabilities is not linear. The Allais paradox and other experiments show that decision weights differ from objective probabilities.

Allais Paradox

Red problem (A vs B)	Blue problem (C vs D)
A: 0.33 × 2500, 0.66 × 2400, 0.01 × 0	C: 0.33 × 2500, 0.67 × 0
B: 2400 with certainty	D: 0.34 × 2400, 0.66 × 0
82% choose B	83% choose C

Expected utility analysis shows inconsistency:

\text{B preferred to A } \Rightarrow 0.34 \cdot u(2400) > 0.33 \cdot u(2500)

\text{C preferred to D } \Rightarrow 0.33 \cdot u(2500) > 0.34 \cdot u(2400)

These contradict each other. The Allais paradox violates the independence axiom of expected utility.

Russian Roulette Experiment (Zuckerman)

Participants were asked how much they would pay to reduce the number of bullets in a 6‑chamber gun:

From 4 to 3 bullets (probability reduction 4/6 → 3/6).
From 1 to 0 bullets (probability reduction 1/6 → 0).

People pay much more to reduce 1→0 than 4→3, even though the probability reduction is the same (1/6). This shows overweighting of low probabilities (the jump from 1/6 to 0 is near certainty of safety) and underweighting of high probabilities (4/6 → 3/6 is in the upper range).

Possibility Effect and Certainty Effect

A 5‑percentage‑point increase in the chance of winning $1 million:

Change in probability	Framing	Happiness rating (1–10)
0% → 5%	Possibility effect: creates a new hope	High
5% → 10%	Quantitative improvement	Moderate
60% → 65%	Quantitative improvement	Moderate
95% → 100%	Certainty effect: moves from uncertainty to certainty	High

Possibility effect: improbable outcomes are overweighted.
Certainty effect: near‑certain outcomes are underweighted relative to full certainty.

Key Takeaways (Risk Perception)

The Allais paradox shows that people violate the independence axiom: choices between gambles are inconsistent with expected utility.
Low probabilities are overweighted; high probabilities are underweighted.
Small changes near 0 (possibility) and near 1 (certainty) have a disproportionate psychological impact.
These effects shape how people evaluate risk, especially for rare events.

Probability Weighting Function

Prospect theory models subjective probability processing via a probability weighting function $w(p)$ that maps objective probabilities $p$ onto decision weights. The weighting function is S‑shaped:

$w(p) > p$ for small $p$ (overweighting)
$w(p) < p$ for large $p$ (underweighting)
Crosses the 45° line at one point (typically around $p \approx 1/3$ )

The function is inverse‑S in shape: steep near 0 and 1, flatter in the middle.

w(p) = \frac{p^\gamma}{(p^\gamma + (1-p)^\gamma)^{1/\gamma}}, \quad 0<\gamma<1

Exam tip: The probability weighting function explains the Allais paradox and the Russian roulette experiment. Always check whether a problem involves small or large probabilities – overweighting of small probabilities often drives risk‑seeking for long shots.

Real‑World Examples

Crack dealers (Freakonomics): Most earn below minimum wage, but a tiny chance of hitting the top is overweighted – they stay despite the risk.
Aspiring film directors and cricketers: The very low objective probability of success is over‑weighted, motivating people to take risky paths (moving to Mumbai, training for national team).

These examples are consistent with (but not uniquely explained by) probability weighting.

Key Takeaways (Probability Weighting)

Decision weights $w(p)$ replace objective probabilities $p$ in prospect theory.
The weighting function is S‑shaped: low $p$ overweighted, high $p$ underweighted.
The possibility effect (overweighting of very small probabilities) and certainty effect (underweighting of near‑certainty) are both captured by $w(p)$ .
Probability weighting explains why people gamble on long‑shot risks and why they over‑insure against low‑probability catastrophes.

Mental Accounting

Mental accounting is the tendency to treat money differently depending on its source, intended use, or the mental “account” to which it is assigned – even when the total amount is identical. This violates the classical economic principle of fungibility (the idea that money can be freely substituted across uses).

Definition (Thaler): Mental accounts are the set of cognitive operations people use to organize, evaluate, and keep track of financial activities. Individuals assign transactions to exogenously created categories and, in doing so, do not maximise total wealth; instead they minimise losses within each account.

Illustrations of mental accounting

Example	What happens	Why it’s mental accounting
Ali’s camera – Ali gets a $100/day per diem (7 days ≈ ₹50,000) vs. a ₹50,000 annual raise. Same surplus, but he is more likely to buy the camera in the per-diem case.	The per-diem is mentally “fun money”; the raise belongs to a different account (salary).	Money from different sources is kept separate, even though fungible in reality.
Thaler in Switzerland – Paid talk in Switzerland + travel; vs. paid talk in New York + travel in Switzerland. Travel expenses feel less bothersome in the first scenario.	A mental “Swiss account” is credited by the talk fee and then debited for travel; in the counterfactual no credit exists.	Outflows are coded against the same account that was credited.
Travel fund – Ali & Rani set aside ₹10,000/month in a separate account for travel, even though they could use regular savings.	They treat the travel money as exclusive to trips, ignoring fungibility.	A formal mental sub-account reduces guilt and simplifies choices.
Envelope system – Grandfather allocates pension cash to envelopes labelled “rent”, “food”, etc.	Money in one envelope cannot be used for another, even if surplus exists elsewhere.	Strict mental budgets prevent cross-category transfers.
Ali’s weekend – Spent heavily on an IMAX movie Friday; declines a free Sunday dinner because he feels “perilously poor” in the entertainment account.	The entertainment account is drained; other accounts (savings) are ignored.	Spending in one category triggers a scarcity feeling in that account only.
Rani’s cardigan – Could not find a cardigan she wanted, so bought shoes instead with the same mental “mall budget”.	Feels “triumphantly wealthy” in the shopping account because the expected spend was not used.	Unspent money in a category is treated as extra wealth within that category.

These effects can coexist: a person can feel simultaneously triumphantly wealthy (in one account) and perilously poor (in another).

Consequences of mental accounting

Borrow high, save low: People take out high‑interest loans while maintaining low‑interest savings, unwilling to “mix accounts”.
Credit card debt: People often do not use savings to pay off credit card balances because the savings and debt belong to different mental accounts.

Experimental evidence (Heath & Soll, JCR)

Participants estimated weekly spending in three categories: entertainment, food, clothes. They were then asked how much they would spend after a specific expense.

Event (spend $20)	% who under‑consume in category
Dinner with friends	Entertainment: 48%; Food: 52%; Clothes: 17%
Buying gloves	Entertainment: ~0%; Food: ~0%; Clothes: 55%

Spending in one category depresses spending only within the same or closely related mental account; it does not spill over to unrelated categories.

Key takeaways – Mental Accounting

People create separate mental accounts for different income sources and spending categories, violating fungibility.
Decisions are driven by the budget constraint within each account, not total wealth.
This explains borrowing while saving, ignoring credit card payoff using savings, and feeling simultaneously wealthy and poor.
Experimental evidence shows spending in one category primarily reduces spending in the same (or related) category.

Bundling Gains and Losses

Given the value function from prospect theory:

v(x) = \begin{cases} x^\beta & \text{for } x \ge 0 \\[2pt] -\lambda (-x)^\beta & \text{for } x < 0 \end{cases}

with typical values $\beta = 0.88$ , $\lambda = 2.25$ . The function is concave for gains (diminishing sensitivity) and convex and steeper for losses (loss aversion). This curvature determines how people prefer to combine or separate multiple outcomes to maximise experienced utility.

Segregating Gains

Principle: For multiple gains, segregation (treating them as separate) yields higher total value than integration (lumping them together).

$v(25) + v(50) > v(75)$

Why: Concavity implies $v(a) + v(b) > v(a+b)$ for positive $a,b$ . The reference point resets after the first gain, so the second gain is evaluated from the new reference point.

Worked example – Tax refund ₹4800 and friend repayment ₹2700 (both gains).

Integration: $v(4800+2700) = v(7500)$
Segregation: $v(4800) + v(2700)$
Because of concave curvature, $v(4800)+v(2700) > v(7500)$ . Segregation is preferred.

Integrating Losses

Principle: For multiple losses, integration (lumping them together) yields higher (less negative) total value than segregation.

$v(-100) \; \text{(one loss)} \; > \; v(-50) + v(-50) \; \text{(two losses)}$

Why: Losses are convex in the negative domain; $v(a+b) > v(a)+v(b)$ for negative $a,b$ (more precisely, the loss function is steeper and convex, so two smaller losses cause more disutility than one larger loss).

Worked example – Two losses of ₹50 vs. one loss of ₹100 (with $\beta=0.88$ , $\lambda=2.25$ ).

\begin{aligned} v(-50) &= -2.25 \times 50^{0.88} \approx -2.25 \times 31.3 = -70.4 \\ 2 \times v(-50) &\approx -140.8 \\[4pt] v(-100) &= -2.25 \times 100^{0.88} \approx -2.25 \times 57.5 = -129.5 \end{aligned}

Result: $v(-100) > v(-50)+v(-50)$ . So integrating the loss is better (less painful).

Mixed outcomes (gains and losses)

Combination	Rule	Reason
Small loss + large gain (net positive)	Integrate	Net gain is positive; loss aversion is avoided in the integrated evaluation. Example: loss 5 + gain 12.565 → net 7.565, $v(7.565) > v(12.565) + v(-5)$ .
Large loss + small gain (net large negative)	Segregate (“silver‑lining” principle)	Keeping the small gain separate provides a small positive component that partially offsets the large loss; integration would make the entire amount a large loss amplified by loss aversion. Example: loss -6000 + gain 40 → net -5960; $v(40) + v(-6000) > v(-5960)$ .
Large loss + small gain (net small negative)	Integrate	When net loss is small, integration reduces the loss aversion penalty. Example: loss -50 + gain 40 → net -10; $v(-10) > v(40) + v(-50)$ .

Exam tip: The only ambiguous case is large loss + small gain. Check the net loss: if the net loss is large, segregate (“silver lining”); if the net loss is small, integrate. For all other combinations (gains/gains, losses/losses, small loss/large gain), the rule is unambiguous.

Managerial implications

Goal	Application	Principle
Maximise employee satisfaction from bonuses	Pay bonuses in multiple smaller instalments rather than one large lump sum.	Segregation of gains → higher total experienced utility (reset reference point).
Inflict strong disincentive from a small penalty	Impose multiple small penalties (e.g., daily fines) rather than one larger fine.	Segregation of losses → greater total disutility.
Minimise demotivation from penalties (e.g., behavioural correction)	Impose the penalty all at once.	Integration of losses → lower total disutility.

Key takeaways – Bundling Gains and Losses

Due to concave gains: segregate gains (multiple bonuses > one big bonus).
Due to convex, loss‑averse losses: integrate losses (one penalty > several small ones).
For mixed outcomes: integrate when net is positive or net loss is small; segregate (silver lining) when net loss is large.
Managers can design bonuses and penalties to exploit these principles to shape employee motivation and behaviour.

Acquisition Utility and Transaction Utility

Acquisition utility is the conventional economic utility a consumer derives from consuming a good itself — the classical “value for money” based on intrinsic benefit. Transaction utility is the psychological value from paying a price below a reference price — i.e., the thrill of a deal.

$\text{Total Utility} = \text{Acquisition Utility} + \text{Transaction Utility}$

Acquisition utility depends only on the good and its price relative to its benefit.
Transaction utility depends on the gap between actual price and a reference price (what the consumer thinks is “fair” or expects to pay).

Aspect	Econ (rational)	Human (behavioral)
Considers only	Acquisition utility	Acquisition and transaction utility
Reaction to price	Buy if benefit > cost	Buy if total utility positive, including deal quality
Example choice	Cheapest functional option	Option that feels like a bargain

Key illustrations

1. The quilt that didn’t fit
Maya Bar-Hillel wanted a double-bed quilt. All sizes were on sale at $150 (regular prices: king$ 300, queen $250, double$ 200). She bought the king size — even though it would hang over her double bed — because the transaction utility (saving $150 vs. king’s regular price) outweighed the lower acquisition utility of a poorly fitting quilt.

2. Auto vs. Uber (Christ University to Bangalore)

Auto: ₹150 (metered price ~₹72) → friend prefers it for low acquisition utility (cheaper ride).
Uber: ₹150 (fair preannounced price) → narrator prefers it for positive transaction utility (paying equal to reference price, not above).
The difference: friend cares about acquisition utility; narrator cares about transaction utility.

3. Beach beer experiment (Thaler)
On a hot beach, you have only ice water. A companion offers to buy your favourite beer from the only nearby place — either a fancy resort hotel or a small rundown grocery store. The acquisition utility (cold beer) is identical in both scenarios; only the reference price changes.
Results: median willingness-to-pay was $7.25** (resort) vs. **$ 4.10 (grocery). People form a higher reference price for the resort, so paying even $7 feels like a fair deal. The same beer, same situation, different transaction utility.

Marketing implications

MRP / Suggested retail price is used to set an artificially high reference price, creating transaction utility when the actual price is lower.
Discounts are more common for infrequently purchased goods (laptops, fashion, electronics) because consumers have no daily reference; MRP becomes the anchor. For frequent purchases (FMCG, stationery), the reference is what you paid yesterday — discounts are rare.
Macy’s (2006-2007): Used coupons extensively. In spring 2007 they cut coupons by 30% → sales plummeted. Coupons generated transaction utility by making the reference price (MRP) seem high.
J.C. Penney: CEO Ron Johnson called full prices “fake prices”, eliminated coupons and prices ending in .99. Less than 1% of revenue came from full price transactions. After the change, sales fell, stock dropped, and he was ousted. Coupons returned.
Walmart / Costco: Generate transaction utility via lowest-price guarantees — the reference price becomes the outside market price, not MRP.

Extreme case: acquisition utility = 0

When a product is not needed (acquisition utility zero) but is on sale (positive transaction utility), people buy useless things purely for the deal. Example: buying an item just because it’s discounted.

Exam tip: Transaction utility explains why “sale” signs work even when the product is not essential. The size of the reference price gap drives the purchase, not the product’s intrinsic value.

Key takeaways

Total utility = acquisition utility + transaction utility.
Transaction utility depends on the difference between actual price and a context-dependent reference price.
Reference prices are influenced by store type, past prices, MRP, coupons, and guarantees.
Frequent purchases have stable reference prices; infrequent purchases rely on suggested retail prices.
Negative transaction utility (paying above reference) can reduce overall satisfaction, even for identical goods.

Sunk Costs

Sunk costs are past, unrecoverable expenditures. Classical economics (econs) ignores them: only future costs and benefits should matter. Humans, however, often continue a course of action because of what they have already invested — the sunk cost fallacy.

Why sunk costs matter: mental accounting

When you pay for something, you open a mental account with the expectation to close it by consuming the good. If you do not consume, you close the account with a loss (the money gone with no benefit). Because losses are painful (loss aversion), you are driven to “use” the sunk cost to avoid that negative account closure.

Concert ticket example
You paid ₹2000 for a concert. It’s raining heavily. If you go: account closes with (value of concert) – ₹2000. If you skip: account closes with –₹2000 (a pure loss). Loss aversion pushes you to brave the rain.

Illustrations

Example	Scenario	Behaviour explained by sunk costs
Vince’s tennis elbow	Paid $1000 for indoor season, developed painful elbow after 2 months, continued playing in pain for 3 more months.	Did not want to “waste” the membership fee; stopped only when pain unbearable.
War (US in Vietnam, Afghanistan)	Continued investing despite poor prospects.	“Too much already invested” — sunk costs justified further commitment.
Shoes that hurt	(A) MRP ₹5000, paid ₹5000. (B) MRP ₹2500, paid ₹2500. (C) MRP ₹5000, paid ₹2500.	Most wear A longest, then C, then B. In A, larger sunk cost (₹5000) → more pain tolerance. In C, transaction utility (₹2500 saved) partly compensates for the pain, so wear longer than B (same cash outlay, no transaction utility).
Health club attendance	Membership billed twice a year. Attendance spikes right after payment, then declines until next billing cycle.	Sunk cost effect: initial payment triggers use; the effect wanes over time as the payment feels more distant (payment depreciation). Annual billing → steady decline; quarterly → zigzag.

Payment depreciation effect (Gourville & Soman)

The psychological impact of a sunk cost diminishes with time. At a health club:

Members attended more immediately after the billing date.
As time passed, attendance declined — the payment felt less “salient”.
The pattern repeated at each billing cycle.

Flip side: windfall gains

Milkman et al. found that people spend more at a grocery store after receiving a “ $10 off” certificate — about$ 2 extra out of the $10. The certificate creates a mental account surplus (positive transaction utility), leading to additional spending.

flowchart TD
    A[Pay membership fee] --> B[Open mental account]
    B --> C[Attend frequently to “use” the fee]
    C --> D[Time passes, payment effect wanes]
    D --> E[Attendance drops]
    E --> F[Next billing cycle] --> B

Exam tip: Sunk cost fallacy is distinct from loss aversion but often works through it. The key test: are you continuing solely because of past unrecoverable costs? If yes, that’s the fallacy. Econs stop when marginal cost exceeds marginal benefit, regardless of past expenditure.

Key takeaways

Econs ignore sunk costs; humans often cannot, due to loss aversion and mental accounting.
Mental account: you want to “close” a purchase by consuming, to avoid a net loss.
Larger sunk costs lead to greater persistence (e.g., more expensive shoes worn longer).
Transaction utility can offset the pain of a sunk cost (shoe example C).
Payment depreciation: the sunk cost effect fades with time, causing attendance patterns tied to billing cycles.
Windfall gains (e.g., discount certificates) can increase spending — the opposite of sunk cost.

Impatience, Self-control and Strategic Thinking

Discounting the Future

Discounting the future captures the human tendency to value immediate rewards more than delayed ones — a core idea in intertemporal choice, the trade-off between now and later.

The Marshmallow Experiment: Patience Measured

Walter Mischel (Stanford) placed children in a room with one marshmallow and gave a choice: eat it now, or wait 10 minutes and get a second marshmallow. The child could ring a bell to summon the experimenter and eat the single marshmallow at any time. Key results:

Reward in sight → average wait time ≈ 1 minute
Reward out of sight → average wait time ≈ 11 minutes

Follow-up studies (~10 years later, 500 children traced, one-third re-interviewed) found that the waiting time was a significant predictor of later-life outcomes: SAT scores, career success, health, income. In essence, the experiment measured patience as a stable trait.

Intertemporal choice is the foundational idea: all everyday decisions involving time trade-offs — savings, education, health, exercise, entertainment — require weighing present and future utility.

Time Preference and the Discount Factor δ

From today’s perspective, consumption today is worth more than the same consumption tomorrow. The discount factor $\delta$ captures this devaluation:

\delta = \text{rate at which future utility is discounted}, \quad 0 < \delta < 1

$\delta$ near 1: patient (future valued almost as highly as present)
$\delta$ near 0: impatient (future nearly worthless)

Example: Total utility from dinner today ( $u_0$ ) and dinner next week ( $u_1$ ) is

U_0 = u_0 + \delta_w \, u_1

where $\delta_w$ is the weekly discount rate.

Intuitive classification

Behaviour	$\delta$	Interpretation
Save more for the future	High	Patient
Say yes to drugs	Low	Impatient
Use protection while having sex	High	Patient
Be a regular at gym	High	Patient
Encash investment to buy a convertible car	Low	Impatient
Enrol for a PhD	Very high	Very patient

Exponential Discounting: The Standard Model

The standard economic (classical) model of intertemporal choice, due to Paul Samuelson, assumes exponential discounting. The present value at time 0 of a stream of future utilities is

U_0 = u_0 + \delta u_1 + \delta^2 u_2 + \delta^3 u_3 + \cdots = \sum_{t=0}^\infty \delta^t u_t

Worked Example 1: Choosing among alternatives

Given $\delta = 0.9$ and four alternatives:

Alternative	$u_0$	$u_1$	$u_2$
A	1	0	0
B	0	3	0
C	0	0	4
D	10	0	0

Compute present values:

$U_0(A) = 1$
$U_0(B) = 0 + 0.9 \times 3 = 2.7$
$U_0(C) = 0 + 0 + 0.9^2 \times 4 = 0.81 \times 4 = 3.24$
$U_0(D) = 10$

Choice: among A,B,C → C; among all → D.

If $\delta$ drops to $0.1$ :

$U_0(A)=1$
$U_0(B)=0.1 \times 3 = 0.3$
$U_0(C)=0.01 \times 4 = 0.04$
$U_0(D)=10$

Choice: among A,B,C → A; among all → D. Lower $\delta$ moves preference toward immediate consumption.

Worked Example 2: Finding δ from indifference

Ali has utility $u(x) = \sqrt{x}$ . He is indifferent between:

A: 100 today
B: 121 one year from now

Set $U_0(A) = U_0(B)$ :

\sqrt{100} = \delta \sqrt{121} \quad\Rightarrow\quad 10 = \delta \times 11 \quad\Rightarrow\quad \delta = \frac{10}{11} \approx 0.909

(Transcript shows $\delta = 2/3$ in a different example — the worked example above follows the same logic with the transcript’s given utility function and numbers.)

Dynamic Time Consistency of Exponential Discounting

Dynamic time consistency means that choices between two options do not change simply because time passes. A person using exponential discounting (an “econ”) has future selves that agree with the current self’s plan.

Example (IPL tickets):

Option A: League match this year (utility 100)
Option B: Semi-final next year (utility 150)
Option C: Final two years from now (utility 180)
Discount rate: 10% per year → $\delta = 0.9$

At time 0:

$U_0(C) = 0.9^2 \times 180 = 0.81 \times 180 = 146$
$U_0(B) = 0.9 \times 150 = 135$
$U_0(A) = 100$
→ Ali chooses C.

At time 1 (now one year later, only B and C remain):

$U_1(C) = 0.9 \times 180 = 162$
$U_1(B) = 150$ (since it’s consumed today)
→ Ali still chooses C. The plan is time consistent.

Anomalies: Dynamic Time Inconsistency for Humans

Real humans often exhibit dynamic time inconsistency: optimal plans change as time passes, all else equal. The present self and future self disagree.

Today’s choice: wake up early tomorrow and exercise or sleep in?
Choice made today for next week: same question — but the answer often differs.

Common statements betray inconsistency:

“I will watch IPL today, but I will complete the assignment tomorrow.”
“I will go out this evening, but tomorrow I will definitely go to the gym.”
“Next month I am quitting smoking.”

Plans to do the “good thing” in the future are rarely followed when that future arrives. Humans seek instant gratification now, but plan to be patient later.

Exam tip: The key distinction is that an econ (exponential discounter) is time-consistent; a human is time-inconsistent. Any question about “why do people make plans they don’t keep?” points to this anomaly.

Key takeaways

Discount factor $\delta$ (0 < δ < 1): high = patient, low = impatient.
Exponential discounting yields $U_0 = \sum \delta^t u_t$ and ensures dynamic time consistency — choices stable over time.
Anomaly: humans show time inconsistency — early selves plan patience, later selves demand instant gratification.
Real-life intertemporal choices (savings, health, education, addiction) all hinge on δ and on whether discounting is exponential or hyperbolic (implied by inconsistency).

Present-Biased Discounting

Present-biased discounting (also called hyperbolic discounting or beta-delta discounting) captures the fact that humans are much more impatient when trade-offs involve the immediate present versus the near future than when both outcomes lie further ahead. While an econ (a rational, exponential discounter) has a constant discount rate, a human exhibits a higher discount rate in the short run and a lower one in the long run.

Formal Model: Beta-Delta Discounting

The utility function evaluated at time $0$ (today) for a stream of future utilities is:

$U_0 = u_0 + \beta\delta u_1 + \beta\delta^2 u_2 + \beta\delta^3 u_3 + \dots$

Equivalently, factoring out $\beta$ :

$U_0 = u_0 + \beta\left\{\delta u_1 + \delta^2 u_2 + \delta^3 u_3 + \dots\right\}$

$\beta$ (the “present-bias” parameter) captures the extra impatience between the present and the immediate future.
$\delta$ (the “long-run discount factor”) captures how patient the individual is when both outcomes are in the future.

Key insight: $\beta$ is typically much smaller than $1$ (e.g., $\beta = 3/5$ ), while $\delta$ is very close to $1$ (e.g., $\delta \approx 1$ ). This means the bulk of impatience is concentrated on the first step – the distance between “now” and “soon”. In long-run trade-offs (e.g., year 365 vs year 366) the individual is almost perfectly patient.

Worked Example: Rani’s IPL Tickets

Rani has three options:

Option	Utility	When available
A – League match	100	Year 0 (now)
B – Semi-final	150	Year 1
C – Final	180	Year 2

Rani is present‑biased. Her discounting is:

Immediate future (0→1): 30% discount rate → discount factor $0.7$
Next year (1→2): 10% discount rate → discount factor $0.9$
Thereafter: 0% → discount factor $1$

Planning at time 0

\begin{aligned} V_0(A) &= 100 \\ V_0(B) &= 0.7 \times 150 = 105 \\ V_0(C) &= 0.7 \times 0.9 \times 180 = 0.63 \times 180 = 113 \end{aligned}

Rani plans to watch the Final in year 2 (highest value).

At time 1 (when the semi‑final is available now)

\begin{aligned} V_1(B) &= 150 \quad\text{(no discounting – immediate)} \\ V_1(C) &= 0.7 \times 180 = 126 \end{aligned}

Now Rani chooses the Semi‑final. Her choice is time‑inconsistent: what she planned at time 0 (Final) is reversed when the moment arrives.

Empirical Evidence

McClure et al. (thirsty subjects)

Choice	% choosing immediate/sooner
“Juice now” vs “2× juice in 5 minutes”	60% chose “juice now”
“Juice in 20 minutes” vs “2× juice in 25 minutes”	30% chose “juice in 20 minutes”

Estimated discount rates: short‑run ≈ 50%, long‑run ≈ 0%.

Thaler (early present‑bias study)

Scenario	Amount today	Future amount (average)	Implied annual discount rate
“ $15 today vs$ X in 1 month”	$15	$20	345 %
“ $15 today vs$ X in 10 years”	$15	$100	19 %

The huge gap in implied rates (345 % vs 19 %) is direct evidence that short‑run impatience far exceeds long‑run impatience.

Preference Reversals

Preference reversal is the signature of present bias: people systematically change their choice when the outcome moves from “future” to “now”.

Snack choice (Read & van Leeuwen)

Plan for next week: 74% choose apple (healthy).
Choose for today: 70% choose fries (unhealthy).

Movie choice (Lowenstein et al.)

Tonight: 66% choose a low‑brow movie (e.g., Fast & Furious).
Same day next week: 37% choose low‑brow (majority choose high‑brow, e.g., The Seventh Seal).
Two weeks from now: only 29% choose low‑brow.

Applications

Credit card vs Personal loan

A consumer wants a ₹70,000 TV.

Credit card: 20 % p.a. (easily swiped now).
Personal loan: 8 % p.a. (requires going to a bank, paperwork).

Present‑biased consumers swipe the card despite the higher interest, because the immediate cost of effort looms larger than the delayed interest saving.

Exam tip: A salesperson knows this – they will try to make the purchase immediate and the payment distant, exploiting the customer’s present bias.

Vices and Virtues

Present bias explains why we systematically prepone vices and postpone virtues.

Formally, let:

$\beta = \frac12$ , $\delta = 1$ (all impatience is in the first step).
Benefit $B$ and cost $C$ occur at different times.

Vices: Immediate benefit, delayed cost

Example: an egg roll – benefit 3 (taste now), cost 4 (health later).

Decision moment	Utility calculation	Decision
Plan for tomorrow (t=0)	$U_0 = \beta(B - C) = \frac12(3 - 4) < 0$	Do not eat tomorrow.
When tomorrow arrives (t=1)	$U_1 = B - \beta C = 3 - \frac12 \times 4 = 1 > 0$	Eat it now.

Result: preponement – the vice is consumed despite the plan to abstain.

Virtues: Immediate cost, delayed benefit

Example: exercise – benefit 10 (future health), cost 8 (effort now).

Decision moment	Utility calculation	Decision
Plan for tomorrow (t=0)	$U_0 = \beta(B - C) = \frac12(10 - 8) = 1 > 0$	Plan to exercise tomorrow.
When tomorrow arrives (t=1)	$U_1 = -C + \beta B = -8 + \frac12 \times 10 = -3 < 0$	Skip the gym.

Result: postponement – the virtue is never done despite good intentions.

flowchart LR
    subgraph Vices
        A[Today: Plan to avoid] --> B[Tomorrow: Consume]
    end
    subgraph Virtues
        C[Today: Plan to do] --> D[Tomorrow: Procrastinate]
    end

Preference reversal for vices and virtues arises from the same mechanism: the immediate component (benefit for vices, cost for virtues) is not discounted, while the delayed component is discounted by $\beta$ .

Key takeaways

Present‑bias (hyperbolic/β-δ) discounting: $\beta$ short‑run impatience, $\delta$ long‑run patience; $\beta < 1$ , $\delta \approx 1$ .
Utility at time 0: $U_0 = u_0 + \beta\sum_{t\ge1}\delta^t u_t$ .
Leads to time‑inconsistent plans: what you plan today differs from what you actually choose later.
Preference reversals are pervasive: snacks, movies, credit decisions.
Vices (immediate benefit, delayed cost) are preponed; virtues (immediate cost, delayed benefit) are postponed – both explained by the same β-δ model.

Exam tip: The classic preference‑reversal problem asks: “At time 0 Rani plans the Final, at time 1 she picks the Semi‑final.” Be ready to compute the discounted values and show the reversal exactly as in the worked example.

The Self-Control Problem: A Motivation

At a dinner party, Richard Thaler removed a bowl of cashews because he could not stop eating them. His economist colleague objected: if you prefer not to eat more, why not just stop? Thaler’s reply — “if the nuts were still available, I would have eaten more” — reveals a self-control problem that standard economic theory cannot capture. The “human” experiences a conflict between present passion and future preferences; the “econ” assumes preferences are consistent across time.

Further examples:

Smokers buy single sticks to avoid chain-smoking a pack.
Dieters avoid stocking ice cream at home.
People place alarm clocks out of reach.
Academics register for conferences far in advance to force paper completion.
Buying a smaller, more expensive Coke to limit consumption.

Adam Smith’s Theory of Moral Sentiments described this as a conflict between passion and the impartial spectator (reason). He noted that pleasures ten years hence matter little compared to today’s, and that willpower is needed to overcome myopia.

Exponential vs. Hyperbolic (Present‑Bias) Discounting

Under exponential discounting, if you do not indulge now, you will not indulge in the future — preferences are time‑consistent.
Under hyperbolic (present‑bias) discounting, if you do not indulge now, you may want to indulge later — preferences can reverse over time. This creates the distinction between naifs and sophisticates.

Type	Perceived future self-control	True short-run discount factor	Example behaviour
Naif	Believes future selves will follow today’s plan	Thinks $\hat{\beta} = 1$ (no present bias)	Plans to skip bad movie, but ends up skipping the best one
Sophisticate	Knows future selves will have present bias	Has $\hat{\beta} = \beta$ (true $\beta$ )	Pre‑commits to avoid future mistakes – makes time‑consistent choices
Partial naif	Somewhere in between	$\beta < \hat{\beta} < 1$	Underestimates but does not ignore future self-control problems

Worked Example: Movie Theatre Scheduling

You can watch a movie on three of four weekends. Each weekend offers a different film:

Weekend (t)	Movie quality	Utility
0 (this weekend)	Mediocre (A)	3
1 (next weekend)	Good (B)	5
2 (two weeks)	Great (C)	8
3 (three weeks)	Fantastic (D)	13

You must skip exactly one movie. Assume $\beta = \frac12$ , $\delta = 1$ (i.e., no long‑run discounting; all present bias is immediate).

Exponential Discounter (β irrelevant)

At time 0, evaluate utilities of skipping each movie:

Skip A: $3 + 5 + 8 + 13 - 3 = 5 + 8 + 13 = 26$
Skip B: $3 + 8 + 13 = 24$
Skip C: $3 + 5 + 13 = 21$
Skip D: $3 + 5 + 8 = 16$

Best is to skip the worst movie (A). Choice is time‑consistent.

Naive Hyperbolic Discounter

At time 0, the current period is weekend 0. Any utility received in future periods ( $t \ge 1$ ) is discounted by $\beta = \frac12$ .

Nudges and Public Policy

Discrimination

Discrimination occurs when members of a minority group (e.g., women, Blacks, Muslims, Dalits, immigrants) are treated less favourably than members of a majority group with otherwise identical characteristics and in similar circumstances.

Persistent gaps in outcomes – employment, leadership, income – across social groups cannot be fully explained by instrumental channels (education, experience, health). Discrimination contributes to the residual gap.

Constitutional & Legal Backdrop

India, Article 15: State shall not discriminate on grounds of religion, race, caste, sex, place of birth.
The First Amendment (1951) permits special provisions (affirmative action) for backward classes, SC/ST.
US, 14th Amendment: Equal protection of laws. A 1971 Supreme Court ruling under the Civil Rights Act: practices that freeze the status quo of prior discriminatory employment are invalid, even if facially neutral.

Post Hoc Justification

People choose based on hidden preferences, then rationalise the choice with a different, publicly acceptable reason.

Magazine Choice Study (Zoe & Norton)

Participants chose between two magazines:

Dimension	Magazine 1	Magazine 2
Sports coverage	9 articles	6 articles
Feature articles	12 articles	19 articles
Special issue	Swimsuit	Top 10 athletes

Result: 92% chose Magazine 1. When asked why, 64% claimed “coverage” was more important than feature articles.

Treatment: identical coverage and features; only special issue swapped (Magazine 1 → sports coverage, Magazine 2 → swimsuit). Result: only 46% chose Magazine 1; only 17% now claimed coverage was more important. The true preference was the special issue, not coverage. The earlier justification was post hoc.

Construction Hiring Study (Norton et al.)

Participants acted as hiring head for a stereotypically male construction job. Two key dimensions: industry experience vs engineering education.

Control: CVs gender‑neutral (initials). 76% chose the more educated candidate; 48% ranked education higher.
Treatment 1 (Male educated): the more educated candidate was given a male name. 75% chose the educated candidate; 50% ranked education higher.
Treatment 2 (Female educated): the more educated candidate was given a female name. 43% chose the educated candidate; only 22% ranked education higher.

Interpretation: Participants decided whom to hire (man vs woman) first, then adjusted their justification of which attribute mattered. The change in justification reveals taste‑based decision‑making.

Taste‑Based Discrimination (Gary Becker, 1957)

Taste‑based discrimination arises when an employer has a personal distaste for hiring from a particular group, even when candidates are equally productive.

The distaste acts as a psychic cost $d$ (a fraction of the wage). The effective wage for a discriminated worker becomes $w^* (1 + d)$ , where $w^*$ is the market wage. This shifts the quantity of labour demanded downward.

flowchart LR
  A[Market wage w*] --> B[Add psychic cost d]
  B --> C[Effective wage w*(1+d)]
  C --> D[Lower quantity hired Lb < L*]

$L^*$ : labour hired from one’s own group at wage $w^*$ .
$L_b$ : labour hired from the discriminated group.
Because the demand curve slopes down, $L_b < L^*$ – this is the observed discrimination.

Exam tip: Taste‑based discrimination does not require productivity differences. The key is the employer’s psychic cost, which makes hiring from the minority group effectively more expensive and reduces the quantity hired.

Key takeaways

Discrimination is differential treatment of identical individuals from different groups.
Post hoc justification: people choose first, then invent a rational reason.
The magazine and hiring studies show how hidden preferences (special issue, gender) drive choices, with justifications shifting to match the decision.
Taste‑based discrimination (Becker) models an employer’s distaste as a psychic cost added to the wage, reducing employment of the disliked group.
Gaps in outcomes (employment, leadership) persist beyond education and experience; discrimination is a residual cause.

Statistical Discrimination

Statistical discrimination explains unequal treatment as a rational response to incomplete information. When an agent lacks full information about an individual (e.g., productivity, honesty), she uses the group average of that individual’s social group as a proxy – a heuristic or rule of thumb. This arises from a signal‑extraction problem: the true quality is unobserved, so the agent infers it from the group mean. Proposed independently by Arrow and Phelps, it is also called Arrow‑Phelps statistical discrimination.

Mechanism

In the canonical model, an employer cannot directly observe a job applicant’s productivity. To decide whom to hire, the employer:

Notes the applicant’s group (e.g., race, caste).
Assigns the average productivity of that group to the applicant.
Hires the applicant with the highest inferred productivity.

If group averages differ – e.g., lower average education for Dalits or Blacks – the individual from the disadvantaged group is less likely to be hired, even if her actual productivity is high.

flowchart LR
  A[Employer sees applicant] --> B{Can productivity be observed?}
  B -- No --> C[Uses group average productivity]
  C --> D[Compare inferred productivities]
  D --> E[Hire the highest]

Examples

Context	Behaviour	Statistical interpretation
Train in Germany	White German man moves away from a South Asian passenger	Lacks info on passenger’s honesty; assigns group average of unethical behaviour from home‑country corruption indices.
Auto mechanic	Overcharges a physically differently‑abled customer	Lacks info on customer’s willingness‑to‑pay; assumes lower search/monitoring costs for a differently‑abled person → higher willingness to pay.

Both examples can also be explained by taste‑based discrimination (pure distaste). Disentangling the two is often difficult but policy‑critical.

Contrast with Taste‑Based Discrimination

Feature	Taste‑based	Statistical
Core driver	Psychic disutility / prejudice	Incomplete information + rational inference
Mechanism	Decision‑maker pays a psychic cost for interacting with the out‑group	Decision‑maker uses group average as a proxy for unobserved traits
Source of discrimination	Hostility or animus	Imperfect signals, not animus
Example	Employer refuses to hire Blacks because he dislikes them	Employer fails to hire Blacks because he infers lower productivity from average group education

Policy Implications

Why distinguishing the two matters: the optimal policy differs sharply.

flowchart TD
  P[Identify type of discrimination] --> Q{Which type?}
  Q -->|Taste‑based| R[Reduce psychic cost via contact]
  R --> S[Affirmative action policies to force intermixing]
  Q -->|Statistical| T[Eliminate information asymmetry]
  T --> U[Provide more information: productivity, education quality, soft skills]

Exam tip: A one‑size‑fits‑all anti‑discrimination policy will fail if the underlying mechanism is misdiagnosed. Affirmative action may not cure statistical discrimination; improving information may not cure taste‑based discrimination.

Empirical Evidence

Distinguishing taste‑based from statistical discrimination in real‑world data requires clever research designs.

Blind Auditions (Goldin & Rouse)

Studied US symphony orchestras. Blind auditions (judges cannot see the player’s gender) vs. non‑blind auditions.

Condition	% of hires that were women
Blind audition	35%
Non‑blind audition	10%

At first glance this seems like taste‑based discrimination (same violin quality, only gender information differs). But it could also be statistical if judges lack information on women’s willingness to travel or work weekends → they assign group average of lower travel availability.

Police Stop‑and‑Search (Knowles et al.)

Maryland police data: Black men were searched more often than whites.

Taste‑based: police have distaste for African‑Americans.
Statistical: police lack info on criminality → assign group average. If the average crime rate is higher for Blacks, searches will be more frequent, even without animus.

Correspondence Study (Bertrand & Mullainathan)

Classic field experiment to isolate discrimination on observable qualifications.

Design:

Created 5,000 fictitious CVs with identical education and experience.
Randomly assigned white‑sounding names (Emily, Greg) or black‑sounding names (Lakisha, Jamal).
Submitted to ~13,000 job ads.

Results:

Name type	Call‑back rate
White‑sounding	9.65%
Black‑sounding	6.45%

The 3+ percentage‑point gap persisted across gender, city, and job type.
Whites received call‑backs faster, and the gap widened with higher CV quality.

Because the CVs were identical in education and experience, the result is widely interpreted as taste‑based discrimination. However, the employer may still be using statistical discrimination on traits not captured in the CV – e.g., soft skills, self‑confidence, or language ability – if those traits correlate with race and are unobserved. The experiment cannot fully rule this out.

Key Takeaways

Statistical discrimination arises from incomplete information and rational use of group averages; it is not based on animus.
It relies on a signal‑extraction problem: the decision‑maker infers individual quality from the group mean.
Distinguishing it from taste‑based discrimination is crucial because the two call for different policies: contact/affirmative action for taste; information provision for statistical.
Empirical studies (Goldin & Rouse, Knowles et al., Bertrand & Mullainathan) suggest both types operate, but clean separation is difficult.
Even when qualifications are identical, statistical discrimination on unobservable soft skills may persist, making it hard to claim pure taste‑based discrimination.
Policy responses must match the root cause – a mismatch wastes resources and may fail to reduce inequality.

Stereotype Threat and Underperformance

Stereotype threat occurs when awareness of a negative stereotype about one's group impairs performance. The effect is situational: making a stereotype salient can lower test scores, while removing the threat eliminates the gap.

Study	Context	Key result
Steele & Aaronson	Test performance with/without race indicated	African‑American performance decreased when race was indicated before the test; increased when told the test did not measure ability (relative to control).
Hoff & Pandey	Dalit and upper‑caste male students in UP, maze‑solving task	Control (caste not revealed): no caste gap. Treatment 1 (caste revealed via last names): Dalit scores fell 20% vs. control. Treatment 2 (same as T1 but payoff partly determined by random chance): no gap – attributable to anticipation of being discriminated, not stereotype threat.
Aaronson, Fried & Good	Letters of encouragement to struggling junior students	T1 (intelligence is a muscle): higher enjoyment and valuing of education. T2 (intelligence is fixed): lower outcomes compared to control.
Good, Aaronson & Harder	GRE‑style math test	T1 (test measures mathematical ability – stereotype condition): women underperform. T2 (additionally told test shows no gender differences): women perform significantly better.

Exam tip: The Hoff & Pandey experiment is a clean test distinguishing stereotype threat from expectation of discrimination. The key is that adding a random‑chance component (treatment 2) eliminates the gap, which would not happen if pure stereotype threat were the mechanism.

Self‑Expectancy Effects and Self‑Fulfilling Prophecies

Two models show how initial beliefs can perpetuate skill differentials even when underlying ability is equal:

Model 1 – Worker's belief about employer updating
Minority workers believe employers are slow to update beliefs about ability → less incentive to invest in skills → skill differential emerges → employers observe the differential and act accordingly → belief becomes self‑fulfilling.

Model 2 – Employer's incorrect belief about worker
Employers incorrectly believe minority workers cannot meet the task → invest less in training those workers → skill gap emerges → original belief is confirmed → self‑fulfilling prophecy.

Pygmalion effect (Rosenthal & Jacobson, 1968):
In a US public elementary school, teachers were told that a randomly selected 20% of their class were expected to develop much faster (as measured by IQ). By month 8, the “bloomers” gained 12 IQ points vs. 8 points for controls. Mechanisms: extra attention, encouragement, increased student motivation.

Diversity in Leadership Positions (Quotas)

Mandated reservation/quota for women on corporate boards, evaluation committees, etc., forces majority members to evaluate minority performance more often. However, results can be counter‑intuitive:
Bagues et al. (Review of Economic Studies) found that women were less likely to succeed in the Spanish judiciary entry exam when the evaluation committee had more women (and men more likely to succeed with more men). This may arise because women on committees go to great lengths to avoid appearing biased.

Role of Minority Leaders and Attitude of the Majority

Taste bias against a group (e.g., women) can feed into statistical discrimination: if no minority leaders have ever been observed, majority members lack evidence of competence → they continue not to elect/select minority leaders → the cycle continues. Forcing exposure (e.g., through quotas or reservations) can break this cycle by providing positive examples, updating beliefs, and reducing bias.

Affirmative Action and Minority Role Models

Successful role models from a minority community can change aspirations and effort:

Hoff & Pandey (revisited): Dalit children reduced effort when caste was made salient because they expected unfair evaluation. Seeing a successful person from one’s own group can counteract that belief.
Cheryan et al.: Exposure to a non‑stereotypical computer science role model (e.g., someone who plays football, swims, loves cats) increased female subjects’ belief in their own success in CS – irrespective of the role model’s gender.

Intergroup Contact (Allport’s Hypothesis)

Intergroup contact reduces prejudice only under four conditions:

Equal status between groups in the situation.
Common goal.
Intergroup cooperation.
Support of authorities/law/custom.

Setting	Satisfies conditions?	Effect on prejudice
Taxi driver (Dalit) & passenger (upper caste)	No equal status, no common goal	Contact unlikely to reduce prejudice.
Office project team (mixed caste)	Yes: equal status, common goal, cooperation, organisational “equal‑opportunity” policy	Contact should reduce prejudice.

Meta‑analytic evidence (Pettigrew & Tropp, 2006): Contact reduces prejudice in 94% of 515 studies, and generalises across racial, ethnic, age, mental‑illness, and LGBT contexts.

Random‑assignment experiments:

Study	Design	Finding
Sacerdote (2000)	Random roommate assignment in college	Peer characteristics affect GPA and social‑club membership but not college major.
Boisjoly et al. (AER)	Random assignment of white students to black roommates	White students with a black roommate more likely to endorse affirmative action.
Gautam Rao	Delhi private schools mandated to admit 25% from economically weaker sections (RTE)	Rich kids exposed to poor kids showed greater altruism (less taste bias) and were more likely to choose poor kids as teammates in a relay race.
Matt Lowe	Randomised cricket teams with mixed caste composition	Mixed caste teams playing together (common goal) reduced discriminatory attitudes and increased cooperation.

Exam tip: Allport’s four conditions are frequently tested. Remember the contrast between the taxi‑driver example (fails) and the office‑team example (passes). The meta‑analysis result (94%) is a powerful quantitative takeaway.

Key takeaways

Stereotype threat impairs performance when a negative identity is made salient; the effect can be eliminated by reframing the task or removing human evaluation.
Self‑fulfilling prophecies (Pygmalion effect) show that false beliefs about ability can become real through differential treatment and effort.
Combatting discrimination requires structural interventions: quotas, role models, intergroup contact that meets Allport’s conditions.
Contact reduces prejudice robustly across many group boundaries and generalises to broader attitudes (e.g., endorsement of affirmative action).

Gender Gaps

Gender gaps in labour-market outcomes — wages, employment, representation — persist across social groups. Even after controlling for education, experience, and other “human capital” factors, a residual unexplained gender wage gap remains (≈9% in the US). This suggests that behavioural or psychological differences between men and women may play a role.

The gender wage gap and representation

US wage gap (raw): women earned 62.1% of men’s income in 1980, rising to 79.3% in 2010.
Adjusted gap: after controlling for education, experience, age, etc., women still earn ≈9% less — the unexplained component.
Top of the distribution: the pay gap is larger and the explained portion is very large.
Representation (vertical segregation):
- IIM Bangalore: 25–30% of MBA students are women → fewer women in top corporate positions later.
- US: women hold only 2.5% of the five highest-paid executive positions per firm.
- Fortune 500: 3.6% of CEOs, 14% of executive officers, 16% of board directors.
- EU: fewer than 14% of board members are female.
- Bertrand (2017): Fortune 500 → 19.9% board members women, 5.8% CEOs; EU → 23.3% board, 5.1% CEOs.
India’s numbers are likely similar or worse (no specific data given in lecture).

What explains the gender gap? Two broad categories

Differences in traits: preferences for jobs, performance on different metrics, self-beliefs, confidence.
Differential response to labour‑market incentives: bargaining attitude (women less likely to negotiate raises), willingness to take roles that lead to promotion, competitiveness.

Because career progression is highly competitive, a key question is whether men and women differ in competitive performance or entry into competition:

Intensive margin (performance in competitive environments): do women perform differently when they are already in a competition?
Extensive margin (entry into competitive environments): are women less likely to choose to compete in the first place?

The lecture uses a lab experiment (Gneezy, Niederle & Rustichini, QJE) to answer the intensive-margin question.

Experiment: gender differences in competitive performance

Design: groups of 6 participants solve mazes for 15 minutes under different compensation schemes.

Treatments and results

Treatment	Compensation	Group composition	Result (gender gap in performance)
Piece rate	2 shekels per maze solved	Mixed gender (men + women)	Baseline gender gap (small)
Tournament (mixed)	Winner gets 12 shekels; others get 0	Mixed gender	Gender gap substantially larger than in piece rate
Random pay (mixed)	One random participant gets 12 shekels per maze solved	Mixed gender	Gender gap similar to tournament (women’s performance did not improve much)
Tournament (single‑gender)	Same winner‑takes‑all; groups are all‑male or all‑female	Single gender (men only / women only)	Gender gap shrinks to piece‑rate level; women’s performance in single‑gender tournaments is significantly higher than in mixed‑gender tournaments

Eliminating potential explanations

Four possible reasons for the larger gender gap in mixed‑gender tournament:

Women dislike effort when pay is uncertain. Tested by random‑pay treatment → women did not increase performance, so this is unlikely.
Women’s performance is maxed out — they cannot do better. Tested by single‑gender tournament → women performed much better when not competing against men; therefore not maxed out.
Women dislike competition in general. Single‑gender tournament shows women compete effectively → they do not dislike competition per se.
Women dislike competing against men specifically — likely due to a belief that they will lose.

flowchart TD
  A[Larger gender gap in mixed tournament] --> B{Possible explanations?}
  B --> C["1. Uncertainty aversion"]
  B --> D["2. Performance maxed out"]
  B --> E["3. Dislike competition generally"]
  B --> F["4. Dislike competing against men (belief)"]
  
  C --> G[Test: Random pay treatment]
  G --> G1[Gender gap persists → reject 1]
  D --> H[Test: Single-gender tournament]
  H --> H1[Women's performance rises → reject 2]
  E --> H
  H --> H2[Women compete well in single-gender → reject 3]
  F --> H
  H --> H3[Only explanation left: 4 – belief-driven discouragement]

Conclusion on intensive margin: Women perform worse in mixed‑gender competition not because of lower ability or aversion to competition, but because they believe they cannot win against men, leading to lower effort. In single‑gender settings, their performance matches the piece‑rate level.

Exam tip: The elimination strategy — using iterative treatments to rule out confounds — is a classic experimental design. Know the logic: random pay rules out uncertainty aversion; single‑gender rules out ability cap and general competition aversion, isolating the “competing against males” belief effect.

Key takeaways

Gender gaps in wages and top‑level representation are large and partially unexplained by conventional human‑capital factors.
Two behavioural questions: (1) Do women perform differently in competitive environments? (2) Do women choose to enter competitions less often?
Lab experiment (maze solving) isolates gender differences in performance under competition.
Piece‑rate baseline → small gender gap; mixed‑gender tournament → large gap; random pay → gap persists; single‑gender tournament → gap disappears.
The driving mechanism is belief: women believe they cannot beat men, so they underperform in mixed‑gender competition.
Single‑gender environments eliminate this performance gap, suggesting the difference is not about ability or general competition aversion.

Gender Differences in Competition Entry (Niederle & Vesterlund, 2007)

The classic question: do women shy away from competition and do men compete too much? The experiment tests whether gender predicts willingness to enter a competitive payment scheme.

Experiment design – groups of 4 participants solve simple two‑digit addition problems for 5 minutes. Three parts:

Piece rate – $0.50 per correct answer.
Tournament – $2 per correct answer for the group’s top performer; others earn$ 0.
Choice – participants choose which payment scheme to apply to their performance in that round.

The key outcome is whether women are less likely than men to opt into the tournament in part 3, even after controlling for ability and confidence.

Performance differences?
No – the distribution of scores in the piece‑rate and tournament parts is statistically similar for men and women. Ability on the simple math task is not the driver.

Results – tournament entry (Figure B in the paper):

Rank in group	% of men choosing tournament	% of women choosing tournament
4 (lowest)	~60%	~25%
3	~55%	~40%
1 (highest)	~80%	<40%

The gap is large and persists even after controlling for self‑confidence (participants’ predicted rank). For rank‑1 performers adjusted for confidence, 80% of men still choose tournament vs. only 50–55% of women.

Exam tip: The key finding is that women under‑enter competition relative to men, and this is not fully explained by lower confidence. The gap exists even among equally confident high‑performers.

A replication at IIM Bangalore with top students shows the same pattern: a significant gender gap in tournament entry.

Key takeaways

Men and women perform equally on the simple math task.
Women are less likely to choose tournament payment, even when they are top performers.
Lower self‑confidence among women explains part, but not all, of the gap.
The preference to avoid competing against men appears to be a distinct driver.

Gender Differences in Volunteering for Low‑Promotability Tasks (Babcock et al.)

Promotability – the degree to which a task influences evaluation and promotion. Non‑promotable tasks are necessary but do not advance a career (e.g., organizing a picnic, receptionist duties, committee work). Women are overrepresented in such roles.

Experiment design – groups of 3 members. Each can choose to invest in a 2‑minute task (volunteer) or not. Payoffs:

Scenario	Investor(s)	Non‑investor(s)
No one invests	Each gets $1	–
Exactly one invests	Investor gets $1.25	Each non‑investor gets $2
More than one invests	(Not tested in basic design)	–

This mimics a real office: everyone wants the picnic to happen, but no one wants to organize it because the payoff for volunteering is lower than free‑riding.

Result – mixed‑gender groups:
Women are significantly more likely to volunteer than men – a gender gap of about 15 percentage points.

Is it preferences or beliefs?
The experiment is repeated in single‑gender sessions (only men or only women). The gap disappears: men volunteer more (when only men are present) and women volunteer less (when only women are present). This shows the gap in mixed groups is driven by beliefs: women believe men will not volunteer, and men believe women will volunteer. It is not a pure preference for volunteering.

flowchart TD
    A[Mixed-gender group] --> B[Women volunteer more than men]
    A --> C[Gap ~15 p.p.]
    D[Single-gender group] --> E[Men volunteer more; women less]
    D --> F[Gap vanishes]
    B --> G[Driven by beliefs, not fixed preferences]

Key takeaways

Non‑promotable tasks are essential but do not lead to promotion.
Women volunteer for them more than men in mixed settings.
The gap is caused by beliefs about others’ behaviour, not by an intrinsic preference.
This contributes to the gender gap in top‑level representation and wages.

Policies to Reduce Gender Gaps (Banerjee & Mustafi)

Given the gender gap in volunteering for non‑promotable tasks, can social recognition close it? Paying more is not feasible (by definition, non‑promotable tasks cannot carry large monetary incentives).

Experimental treatments – baseline replicates Babcock et al.; then three variations:

Treatment	What is done	Effect on gender gap
Baseline	No recognition	Replicates original gap
Positive social recognition	Names of volunteers circulated with a smiley + applause	Gap shrinks – men begin to compete for non‑monetary recognition
Negative social recognition	Names of non‑volunteers circulated with a frowning face	Total volunteering rises, but gap widens (women volunteer even more)
Both positive and negative	Both types of recognition	Gap closes again

Intuition:

Positive recognition turns volunteering into a status contest that men enter.
Negative recognition punishes free‑riding, but women already volunteer heavily – it pushes them even further, enlarging the gap.
Combining both creates a balanced incentive that encourages both genders.

Exam tip: Social recognition is a low‑cost policy tool. The type matters: purely negative recognition can backfire and widen the gender gap.

Key takeaways

Monetary incentives are not appropriate for non‑promotable tasks.
Positive social recognition reduces the gender gap by attracting men.
Negative social recognition increases overall volunteering but expands the gap.
A combination of positive and negative recognition eliminates the gap.
Policy interventions should be designed with awareness of belief‑driven gender differences.

Nudges: Definition and Key Concepts

A nudge is any aspect of the choice architecture that alters people’s behavior in a predictable way without forbidding any options or significantly changing economic incentives. To count as a nudge, an incentive must be easy and cheap to avoid.

Definition (Thaler & Sunstein): A nudge is any aspect of the choice architecture that alters behaviour predictably, while preserving freedom of choice and leaving economic incentives largely unchanged.

Libertarian Paternalism

Nudges are often called libertarian paternalism:

Libertarian because individuals retain the same set of options and can freely choose.
Paternalistic because the choice architecture is designed according to values or objectives set by someone else (e.g., policymakers).

Classic Examples

Example	Nudge	What it changes	What it does not do
Smart lunchroom	Place fruit at eye level, fries at the end; wheel salad bar into high-traffic areas	Order and visibility of food	No ban on fries; no price change
Escalator/stairs in malls	Markings: “stand on right, walk on left”	Social norm cues	No blocking of either option
Men’s urinals at Schiphol Airport	Engrave a housefly near the drain	Target for aiming	No physical barriers; no fines

Exam tip: The core of a nudge is that no option is removed and no monetary incentive is altered. If a policy bans fries or taxes sugary drinks, it is not a nudge.

Key takeaways

A nudge alters behaviour predictably while preserving free choice and not changing economic incentives.
Termed libertarian paternalism: free choice (libertarian) + design by others (paternalistic).
Examples: smart lunchrooms, escalator markings, urinal fly.
Banning or taxing is not a nudge.

1. Automatic Enrollment in Retirement Savings (Madrian & Shea)

Problem: People save too little. Traditional policy (interest rate changes) assumes rational intertemporal choice, but real behaviour is inertia-ridden.

Intervention: A large US corporation changed the default for new employees from “opt in” to “opt out” at a 3% contribution rate into a money market fund.

Cohort	Hiring period	Default	Participation rate	Most common contribution	Stock investment (approx.)
Old	4/1996 – 3/1997	No enrolment	57%	0% (?)	~75%
Window	4/1997 – 3/1998	No enrolment	49%	63% contributed 0%	~75%
New	4/1998 – 3/1999	Enrolment at 3% in money market	86%	65% contributed 3% (the default)	16%

Results:

Participation jumped from ~49-57% to 86%.
The default contribution rate (3%) became the modal choice.
Investment allocation shifted heavily toward the default (money market, not stocks).

Exam tip: The default effect is one of the most powerful nudges. People stick with the status quo. Changing the default from “opt in” to “opt out” dramatically increases participation without restricting freedom.

2. Save More Tomorrow Program (Thaler & Benartzi)

Problem: Employees agree with financial advisors that they should save more, but say they “cannot afford it now” (present bias).

Intervention: Ask employees to commit now to increase their savings rate next year (i.e., “save more tomorrow”). Nothing changes immediately; only future contributions rise.

Result: The treatment group showed a substantial increase in savings compared to the control group. The program was a great success because it leverages present bias: people are willing to commit to future sacrifice, but not immediate.

3. Behavioural Insights Team (BIT) and Tax Compliance

Context: The UK’s Behavioural Insights Team (BIT, or “Nudge Unit”) – established in 2010 under the Cameron government – ran a field experiment on tax debtors.

Treatments (randomised letters to 100,000 debtors):

Local norm: “Most people in your local area have paid on time.”
Debt norm: “Most people with a debt like yours have already paid.”
Both norms: Combined local and debt norm.

Results: Relative to a control letter (no norm information), all norm letters increased payment. The largest effect came from the combined local + debt norm letter.

Exam tip: This is a classic example of social norms nudges – extremely cheap (just extra ink) yet yields “medium-sized gains with nano-sized investment”. Nudges are not silver bullets but cost-effective.

Key takeaways

Default nudges (automatic enrolment) boost savings participation from ~50% to >80%.
Commitment devices (Save More Tomorrow) overcome present bias by shifting sacrifice to the future.
Social norm messages in tax letters significantly improve compliance at negligible cost.
Nudges are cheap and produce medium-sized effects – a favourable cost-benefit ratio.

Designing Choice Architecture: When Are Nudges Suitable?

Nudges are most effective when people face specific decision-making problems:

Problem type	Description	Example
Self-control (time inconsistency)	Immediate costs/benefits vs. delayed benefits/costs	Virtues (exercise, diet, flossing): cost now, benefit later → under-provided. Vices (smoking, alcohol, sweets): benefit now, cost later → over-consumed.
Rare decisions	Choices made infrequently, little practice	Investment portfolio, buying a house, life insurance
Difficult decisions	Complex, many parameters beyond layperson’s grasp	How much to save, which mutual fund to pick
No feedback	Delayed or absent feedback prevents learning	Unhealthy eating (health deteriorates years later), buying a house
Unknown preferences	Cannot translate expert advice into personal taste	Exotic dishes on a menu – restaurant uses markers (spicy, chef’s special) to draw attention

The Indian Context (Economic Survey 2019)

A table from the Economic Survey lists several policies (e.g., Give up LPG, Aadhaar, Jan Dhan Yojana, Swachh Bharat, ban on alcohol) as potential nudges. Critical note: Not all qualify as nudges. For example:

Aadhaar: Public services contingent on Aadhaar may involve forbidding options (if you lack Aadhaar, you lose access) – does it change economic incentives significantly? Possibly not a nudge.
Swachh Bharat: Toilets provided? Check definition.
Ban on alcohol: Forbids an option → clearly not a nudge.

Exam tip: Always test a policy against the two criteria: 1) No option forbidden, 2) No significant change in economic incentives. The Economic Survey’s classification is debatable – be ready to critique.

Key takeaways

Nudges are suited for self-control problems (virtues/vices), rare, difficult, no-feedback, and unknown-preference decisions.
Not every “behavioural” policy is a nudge – ban or large financial penalty disqualifies it.
Governments worldwide (UK, US, India, World Bank) have institutionalised nudge units.

Choice Architecture and Nudges

Choice architecture is the design of environments in which people make decisions. A well-designed choice architecture aligns visual cues with functionality, reduces friction, respects human cognitive limitations, and guides behaviour without coercion. The key is to understand how the automatic system (System 1) and reflective system (System 2) interact.

Choice Architecture Principles

Consistency between cues and function

A good choice architecture ensures that visual cues match the required action. Conflicting cues create inefficiencies, errors, and frustration.

Door example: Two rings on a door suggest pulling; if the door must be pushed, users waste time and become confused.
Stop sign example: A green background with the word “STOP” conflicts with the learned association (red = stop, green = go). This increases reaction time and risk.
Stroop-like colour-word conflict: The automatic system reads the word faster than it perceives colour. When the word “red” is printed in blue, the brain experiences a conflict between word-reading and colour-processing areas.

flowchart LR
  A[Visual cue] --> B{Matches function?}
  B -->|Yes| C[Efficient, intuitive]
  B -->|No| D[Conflict → errors, frustration]

Exam tip: The Stroop test illustrates the dominance of automatic processing. Nudges should not rely on conflicting cues—always align the automatic response with the desired behaviour.

Defaults – the path of least resistance

Defaults are options that take effect when an individual makes no active choice. They are powerful because of the status quo bias – people tend to stick with what is already set.

Domain	Default	Active choice required to change
Airport trolley	Brake engaged	Press hand rest to move
Lawn mower	Brake on	Release brake
Computer lock	Locked after inactivity	Enter password to unlock
Subscription (Netflix)	Auto-renew	Explicit cancellation request

Defaults are often inevitable in choice architecture. However, their interpretation can be controversial.

Controversy – No Child Left Behind policy:
In the US, a policy required schools to provide student contact details to military recruiters unless parents opted out. Some districts interpreted this as opt-in (parents had to actively consent), while the Department of Defense interpreted it as opt-out (parents had to actively refuse). This subjective ambiguity generated legal and ethical disputes.

Exam tip: Defaults are not neutral – they steer behaviour. Always identify the default in a policy and who benefits from it.

Active (mandated) choice

To avoid the biases of defaults, some architectures force a decision – the user cannot proceed without making a choice.

Example: School admissions requiring parents to check “yes” or “no” before continuing.
Downside: Cognitive burden – most people click “yes” to terms & conditions without reading. For complex processes, required choices can become a nuisance.

Trade-off: Defaults are effortless but can be manipulative; active choices respect autonomy but increase friction.

Post-Completion Errors

Post-completion errors occur when the main task is finished, but a subsidiary step is forgotten. The designer should force the user to resolve the subsidiary step before completing the main task.

ATM: Users withdraw cash but forget the card. Better design: card must be retrieved first, then cash is released.
Medicine packaging: Pills designed with calendar blisters reduce forgetting.
Gmail attachment reminder: If the email body contains “attached” but no attachment is added, a prompt appears.

Design principle: Anticipate common human errors and build safeguards that prevent the error from having consequences.

Mapping Choices to Welfare

Mapping refers to how well a person can connect a choice to its outcome. When mapping is difficult (e.g., medical treatments, investment options), choice architecture should simplify information.

Example: Camera companies used megapixels as a simple metric for image clarity, making the mapping easier for consumers.
Medical decisions: A doctor might present survival rates, side-effects, and quality-of-life scores in a comprehensible format.

Goal: Help people translate complex choices into personally relevant, measurable units.

Half a Dozen Nudges

Nudge	Mechanism	Is it a pure nudge?
Give More Tomorrow	Ask donors to commit to giving two months later (with opt-out). 32% increase in donations (Bremen study).	Yes – no financial incentive.
Helmet license	Offer a special license for riding without a helmet (requires extra training and health insurance).	Yes – less intrusive than a ban.
Destiny Health Plan	Earn “vitality bucks” for gym visits, normal blood pressure, children’s soccer. Bucks tradeable for airline tickets etc.	No – changes economic incentives (rewards).
Dollar a Day	Pay teenage mothers $1/day for every day not pregnant (for 2 years).	No – provides direct monetary incentive.
No-bite Nail Polish / Disulfiram	Bitter polish or anti-abuse medication makes the bad habit painful (System 2 disciplines System 1).	Yes – purely changes the experience, not incentives.
Civility Check	Gmail warning on uncivil emails; optionally forces a 24-hour delay before sending.	Yes – cheap, uses technology to cool emotions.

Exam tip: A nudge must be cheap and avoid altering economic incentives. Rewards like “vitality bucks” or cash payments shift the cost-benefit calculation and are therefore not pure nudges.

Civility check detail: The automatic system (System 1) reacts emotionally; the reflective system (System 2) later regrets the response. A 24-hour pause reduces hot-headed mistakes – analogous to cooling-off periods for crimes of passion.

Key takeaways

Design visual cues that match intended function to avoid automatic-system conflict.
Defaults are powerful because of status quo bias; they are inevitable but can be contested.
Active (mandated) choice respects autonomy but increases cognitive load.
Post-completion errors (e.g., forgetting ATM card) can be prevented by reordering steps.
Simplify mapping between choices and welfare using comprehensible metrics.
Evaluate whether a policy is a true nudge by checking if it changes economic incentives – if so, it is not a pure nudge.

The Experimental Approach

History of Randomised Experiments

Experiments are the engine of behavioral economics. Understanding how a well-designed experiment isolates cause and effect—and how poor design can mislead—requires knowing where the method came from. The history of randomized experiments spans medicine, social policy, and economics, and each domain contributed key design principles.

1. Medical Origins: From Lemons to the Polio Vaccine

The earliest controlled trial is credited to James Lind (1774). To test whether lemons cure scurvy, he recruited eight sailors and split them into two groups: four received lemons (treatment), four did not (control). Although small and non‑random in the modern sense, it established the idea of comparing treated and untreated groups.

Ronald Fisher (1920s–30s) introduced explicit randomization as the foundation for causal inference, arguing that random assignment eliminates systematic bias. His two books—Statistical Methods for Research Workers and The Design of Experiments—formalised this logic.

Why randomization matters: Without it, treatment and control groups may differ in ways that affect the outcome, making it impossible to attribute differences solely to the treatment.

1948 – Streptomycin trial (Austin B. Hill et al.): A true randomised controlled trial (RCT) to test streptomycin against tuberculosis. Limited supply and ethical concerns motivated the trial: rather than giving the drug to all patients (which would waste scarce doses if ineffective), randomisation allowed fair evaluation. This became the blueprint for modern medical RCTs.
1954 – Salk polio vaccine (Jonas Salk): A double‑blind RCT (neither patients nor doctors knew who received vaccine or placebo). Salk used Fisher’s exact test to analyse the results, which showed a clear reduction in polio risk.

2. Social Experiments: Health Insurance and Cash Transfers

Randomised trials moved into social policy in the 1970s, testing large‑scale government programs.

RAND Health Insurance Experiment (1971–1982)

Question: Does free healthcare lead to over‑use and excessive spending?
Design: 7,700 participants (age < 65) assigned to three treatment conditions:
- Free care (status quo)
- Plans with varying degrees of cost‑sharing (copay)
- Health Maintenance Organization (HMO) plan (restricted network of doctors)
Results: Cost‑sharing reduced physician visits, dental visits, hospitalizations, and overall spending. It also reduced initiation of care (people avoided seeing a doctor). The policy implication: copay can curb demand.

Negative Income Tax Experiments (1970s, US and Canada)

Question: Does providing cash transfers to the poor reduce their willingness to work?
Design: Five separate social experiments testing a negative income tax (subsidy that tops up earnings).
Results: No effect on individual labour supply, but an unintended increase in divorce rates.
Critique (Josh Angrist): Too many treatment arms made the design underpowered to detect credible effects.

Experiment	Years	Sample	Key Finding	Design Issue
RAND Health Insurance	1971–82	7,700	Cost‑sharing reduces utilisation and spending	—
Negative Income Tax	1970s	Multiple sites	No work disincentive; divorce rate rose	Underpowered, many arms

3. Lab Experiments in Economics

While social experiments tested policy, lab experiments probed individual and strategic decision‑making. Early work focused on indifference curves, utility, and social dilemmas.

Researcher(s)	Year	Experiment / Contribution
Thurstone	1931	First lab experiment in economics: used choices between goods (hats, shoes, coats) to deduce indifference curves.
Mosteller & Nogee	1951	Lab experiment measuring the subjective value of additional money income (restricted setting).
Von Neumann & Morgenstern	1944 (book)	Proposed four uses of lab experiments: (a) let subjects take/refuse gambles with real money, (b) construct utility curves from behaviour, (c) make predictions about future choices, (d) test those predictions with more complex tasks.
Dresher & Flood (RAND)	1950	Implemented the first Prisoner’s Dilemma game in the lab. Participants played 100 repetitions. The Nash equilibrium (row2, column1) yields lower joint payoff than the social optimum (row1, column2). Empirically, subjects reached neither; their behaviour deviated from rational prediction, sparking decades of research on cooperation.
Guth et al.	1982	Introduced the Ultimatum Game: a bargaining game between a proposer (divides a sum) and a receiver (accepts or rejects). The receiver’s ability to reject unfair offers (even at own cost) violates the rational self‑interest prediction. After the dictator game, it is the most widely used experimental paradigm in economics.

Exam tip: The RAND Health Insurance experiment is a classic example of a field experiment with clear policy implications. The negative income tax experiment illustrates the dangers of too many treatment arms—a design flaw that reduces statistical power.

How the Pieces Fit Together

timeline
    title Key Milestones in Randomised Experiments
    1774 : James Lind : Lemon trial (medical)
    1920s–30s : Fisher : Randomisation for causal inference
    1948 : Hill et al. : Streptomycin RCT
    1954 : Salk : Double‑blind polio trial
    1950 : Dresher & Flood : First Prisoner’s Dilemma lab experiment
    1971–82 : RAND : Health insurance field experiment
    1970s : Negative Income Tax experiments
    1982 : Guth et al. : Ultimatum Game

Key takeaways

Randomised experiments emerged from medicine (Lind, Fisher, Hill, Salk) and were later adopted in social policy and economics.
Fisher formalised randomisation as essential for clean causal inference.
Social experiments (RAND, Negative Income Tax) tested real‑world policies; the latter showed the risk of underpowered designs and unintended consequences.
Lab experiments in economics began with Thurstone (indifference curves) and expanded to strategic games (Prisoner’s Dilemma, Ultimatum Game) that revealed systematic deviations from rational choice.
The Prisoner’s Dilemma lab results directly challenged the Nash equilibrium prediction and spurred the field of behavioural economics.

Behavioral vs. Experimental Economics

Behavioral economics is a sub-discipline that uses experiments as a method. Experimental economics is the broader field of designing and running experiments to test economic theories. Experiments are tools applicable beyond behavioral economics—to psychology, market design, and other disciplines.

What is an Experiment?

An experiment creates identical environments and changes one thing at a time to measure causal inference.

An experiment is a way through which you develop identical environments, but change one thing at a time to measure causal inference.

Advantages of Experiments

Control over data collection – Allows precise measurement of variables consistent with psychological factors that are unobservable in observational data (e.g., beliefs, preferences, norms).
Clean causal identification – The researcher manipulates the treatment, enabling clear inference of cause and effect.

Typology of Experiments

Experiments can be classified into four main types, each with trade-offs in control, generalizability, and cost.

Lab Experiments

Lab experiments take place in a controlled laboratory environment, typically with student participants.

Market experiments – Test price discovery and market mechanisms.

Example: Plott & Pogorelskiy (2017) – how quickly prices converge to fundamental values.
Auction mechanism tests (Kagel & Levin, 1993) – compare trading structures:

Auction Type	Bidding Rule	Winner Pays
First Price Sealed Bid (FPA)	Highest bid wins	Own bid
Second Price Sealed Bid (SPA)	Highest bid wins	Second-highest bid
Third Price Sealed Bid (TPA)	Highest bid wins	Third-highest bid

Outcomes measured: bid values and revenue (related to the revenue equivalence theorem).

Behavioral experiments – Test heuristics, biases, risk, time, and social preferences.

Linda problem (heuristics), Prospect Theory gambles, present bias discounting, social preferences (altruism, reciprocity, trust, deception).

Example: Corruption and Trust (Banerjee, Journal of Public Economics)
Correlational data shows a negative relationship between corruption and trust across countries. To test causality, a lab experiment with three treatments was designed:

Baseline – play the trust game only.

Treatment 1 – play a bribery game (corruption), then a trust game.

Treatment 2 – play a strategically equivalent ultimatum game (no corruption), then a trust game.
Result: corruption caused a decrease in social trust (measured by trust-game outcomes). A subsequent experiment measured social norms.

Ultimatum game – A staple lab experiment.

Proposer splits a pie (e.g., ₹10): offers $s$ (0–10).
Responder accepts (gets $s$ , proposer gets $10-s$ ) or rejects (both get 0).
Subgame perfect Nash equilibrium: proposer offers tiny amount, responder accepts any positive offer.
Lab evidence: average offer ≈ ₹4 (40%); 40–50% of proposers offer 50-50; offers below ₹2 (20%) rejected.
Why reject? Dislike of unfair outcomes – a psychological motivation not observable in field data.

Advantages of lab experiments:

Full control over data collection.
Clean identification (treatment manipulated by researcher).
Creative, small changes possible.

Disadvantages:

Subject pool: typically university students (WEIRD: Western, Educated, Industrialized, Rich, Democratic).
Low external validity / generalizability.
Incentives are small (but sufficient for students).

Lab-in-the-Field (Artefactual Field) Experiments

To address the WEIRD problem, experiments are moved to the field – a lab constructed in a real-world setting (e.g., a village panchayat building). Participants are non-standard (e.g., villagers, farmers).

Examples:

Affirmative action and competitive choice across caste categories – treatments vary cost-based affirmative action policies, comparing decision-making of participants from different caste groups.
Farmers’ behaviour before vs. after harvest.
Teacher expectations: compare teachers’ expectations about students’ ability (by caste/gender) with actual performance – requires setting up labs in schools, eliciting expectations, and linking to outcomes.

Advantages:

Greater generalizability than lab experiments.
Still maintain control relative to pure field experiments.

Disadvantages:

Less generalizable than large-scale field experiments.
More expensive than lab experiments (need to set up “paraphernalia”).

Online Experiments

Conducted on platforms such as Amazon MTurk, Prolific, Qualtrics, TGM. Participants complete tasks online.

Applications:

Expectations about the future economy.
Role of narratives.
Subjective models of the economy among people and experts.

Advantages:

Low cost – fraction of lab experiment cost.
Large-scale, participants from different parts of the country.
Can achieve quota-based representativeness (e.g., 50% men, 50% women) – not a true representative sample but captures key demographic characteristics.

Disadvantages:

Data credibility concerns – AI tools / LLMs may impersonate human participants.
Time constraints – participants unlikely to spend >20–30 minutes; limits complexity and control.

Survey Experiments

A special type of online experiment where a treatment condition is embedded within a survey. Usually unincentivized, enabling very large samples.

Example (Stefanie Stantcheva, Nathan Nunn, et al.):

Study how people perceive and form attitudes toward economic policies (e.g., gains from trade, tariffs).
Manipulate salience of gains/losses from trade to examine impact on policy support.
Recent paper on Zero-Sum Thinking used 20,000 US residents.

Advantages: Very large scale, can study attitudes and perceptions directly.

Disadvantages: Unincentivized – responses may be less reliable than incentivized choices.

Key Takeaways

Experiments are defined by constructing identical environments and varying one factor to infer causality. They offer control and clean identification.
Lab experiments provide maximum control but suffer from WEIRD subject pools and low external validity. They remain useful for testing theory (e.g., auction mechanisms, social preferences).
Lab-in-the-field experiments improve generalizability by recruiting non-student participants in natural settings, at higher cost.
Online experiments are cheap, large-scale, and can achieve quota representativeness, but face data quality risks and time constraints.
Survey experiments embed treatments in surveys, enabling massive samples to study attitudes (e.g., trade policy, zero-sum thinking) without monetary incentives.
The ultimatum game robustly demonstrates that fairness concerns (rejection of low offers) cannot be explained by standard equilibrium – a key result from lab experiments.
When choosing an experiment type, trade off control, generalizability, cost, and credibility of data.

Field Experiments and Randomized Control Trials

Field experiments and randomized controlled trials (RCTs) test hypotheses by observing outcomes in naturally occurring environments – students in schools, patients in hospitals, labourers in factories. Participants operate in their natural habitat, often unaware they are part of a randomized experiment.

Distinction: Field Experiments vs RCTs

While the terms are sometimes used interchangeably, the key difference lies in purpose:

Feature	Field Experiment	Randomized Controlled Trial (RCT)
Aim	Tests a precise theoretical prediction	Tests a hypothesis with direct policy relevance
Example	Does social pressure drive charitable giving?	Do treated bed nets reduce malaria?

Both use random assignment in the field, but RCTs are explicitly designed to inform policy.

Examples

John List et al. – altruism and social pressure
- Competing motivations: pure altruism vs. dislike of saying “no”.
- Design: control (door-to-door campaign), treatment (pre-announcement that a door-to-door campaign will occur).
- Result: people avoid being home when they know a solicitor will come → the dislike of saying no is a major driver of giving.
Bertrand & Mullainathan – racial discrimination in hiring
- Sent résumés with stereotypically white names (Emily, Greg) vs. African-American names (Lakisha, Jamal).
- Result: callback rate for interviews differed by race → evidence of discrimination.
Kremer & Miguel – treated bed nets (RCT)
- Tested effect of insecticide-treated bed nets on malaria, income, child mortality.
- Policy relevance: directly informs public health interventions.
- Michael Kremer (with Banerjee and Duflo) won the 2019 Nobel Prize for using RCTs in development economics.

Advantages

External validity – results are generalizable because they come from real-world settings.
No Hawthorne effect / social desirability bias – participants are unaware they are being studied, so behaviour is natural.

Disadvantages

Expensive and logistically challenging – difficult for a single researcher to run.
Lack of control – cannot isolate precise mechanisms as easily as in a lab.
Hard to add extra treatments – cost limits the number of conditions; often relies on survey measures to infer mechanisms.

RCT in Policy: Banerjee, Cole, Duflo & Linden – Education in India

Problem: how to raise learning levels in low-income countries.
Two programs evaluated via RCT:
- Balsakhi program: young women tutors helped lagging students.
- Computer-assisted learning: computers to improve numeracy.
Each program was compared to a control condition.
Direct policy relevance: successful programs could be scaled up by government.

Exam tip: The key distinction between a field experiment and an RCT is not always sharp; exam questions may ask you to classify an example by its purpose (theory-testing vs. policy evaluation).

Key takeaways – Field experiments & RCTs

Both involve random assignment in natural settings; RCTs have explicit policy focus.
Advantages: high external validity, natural behaviour, no Hawthorne effect.
Disadvantages: costly, less mechanistic insight, limited treatment flexibility.
Classic examples: List (social pressure), Bertrand/Mullainathan (discrimination), Kremer/Miguel (bed nets), Banerjee et al. (education programs).

Non-Randomised Experiments

Experiments can be used without randomization – researchers carefully measure outcomes in situations where assignment is not under their control. These rely on natural or quasi-experimental variation.

Examples

Blouin & Mukand (2019) – Rwandan nation-building radio
- After the genocide, the government broadcast radio programs to promote inter-trust and harmony.
- Terrain determined which villages received the signal → non-random exposure.
- Researchers visited exposed and unexposed villages and measured social trust using a trust game.
Mani et al. – financial scarcity and cognitive bandwidth
- Studied farmers before harvest (poor) and after harvest (rich).
- No randomization – same farmers at two time points.
- Result: scarcity reduces cognitive performance.
Gneezy, List et al. – gender differences in competition
- Compared behaviour (competitiveness, self-confidence, risk) across matrilineal and patriarchal societies; these societies are similar in wealth but differ in women’s roles.
- Found that stereotypical gender differences reversed in matrilineal societies → evidence that nurture, not nature, drives these gaps.
- No randomization – natural variation across societies.
Babcock et al. (2017) – gender and low-promotability tasks
- Measured likelihood of volunteering for tasks with low career payoff.
- Non-randomized but carefully calibrated measurement.
Gautam Rao – intermixing in Delhi private schools
- A court mandate forced schools to admit students from lower socioeconomic status backgrounds.
- Compared students exposed to lower-SES peers vs. those not exposed.
- Used dictator games and other field experiments to measure effects on discriminatory preferences.
- Illustrates how experimental data can uncover mechanisms behind social change.

Purpose of Non-Randomised Experiments

Allow careful measurement of otherwise unobservable outcomes (trust, cognitive load, discrimination).
Complement other data sources to understand mechanisms behind societal phenomena.
They do not allow the same causal inference as randomized experiments, but still provide valuable evidence.

Exam tip: Non-randomised experiments are often used when randomization is impossible or unethical. The key is to identify the source of variation (geography, timing, policy) and argue why it is plausible as a natural experiment.

Key takeaways – Non-randomised experiments

Randomization is not a requirement for an experiment to be useful.
Examples leverage natural variation (terrain, season, culture, court ruling).
Weakness: weaker causal identification; strength: feasible in settings where random assignment is not possible.
Valuable for uncovering mechanisms (e.g., Rao on social exposure, Gneezy on culture vs. gender).

Ethics in Research – Historical Context

Why ethics matter in experimental economics: experiments use human beings as subjects. Historical violations of basic human rights led to strict modern protocols. The core ethical principles are codified in the Belmont Report (1979), resting on three pillars: respect for persons, beneficence, and justice.

Historical Violations (Four Key Cases)

Experiment	When	What happened	Why it was unethical
Nazi twin experiments (Josef Mengele)	1943–45	~1500 sets of twins (children) subjected to eye-colour changes, amputations, deliberate infections under pretext of genetic research.	No consent; horrific physical and psychological harm; murder.
Stanford Prison Experiment (Zimbardo)	1971	24 male students randomly assigned as prisoners or guards in a mock prison; stopped after 6 days because guards became sadistic, prisoners severely stressed.	Lack of protection from harm; no provision for withdrawal; profound psychological distress.
Tuskegee Syphilis Study	1932–72	600 African American men in rural Alabama; 399 had syphilis. Penicillin (effective treatment available from 1947) was deliberately withheld.	Deception; exploitation of a vulnerable population; denial of life-saving treatment.
Milgram Obedience Experiment	1961–62	Participants were ordered to administer electric shocks (up to 440V) to "learners" for wrong answers. Most obeyed despite hearing screams.	Severe psychological stress; deception about the nature of the shocks; potential long-term harm.

Exam tip: The Milgram study is often used to illustrate the tension between scientific knowledge and participant welfare – it produced valuable insights about authority but inflicted distress. Modern IRBs would likely reject it.

Modern Requirements (Derived from Past Violations)

Informed consent – participants must formally agree after being told risks, procedures, and that participation is voluntary.
Mandatory ethics training for all experimenters (e.g. CITI Program).
Institutional Review Board (IRB) approval – the entire experimental design must be reviewed.

Core Ethical Principles – The Belmont Report (1979)

Respect for persons
- Treat participants as autonomous agents; they have the right to say yes or no.
- Protect those with diminished autonomy (children, cognitively impaired, prisoners).
- Informed decision-making must be documented in a consent form.
Beneficence
- Do no harm to participants.
- Maximise benefits while minimising risks.
- Benefit-risk assessment must be independently evaluated by the IRB.
- Researchers benefit (publications, career); participants bear the risk.
Justice
- Fair distribution of benefits and burdens.
- Equal treatment of all participants.
- No exploitation of vulnerable populations.

Elements of a Proper Informed Consent Form

Purpose of the research
Duration of the study
Procedures involved
Risks and discomforts
Potential benefits
Confidentiality measures – how data is anonymised and stored
Compensation – essential in economic experiments
Voluntary participation and right to withdraw at any time
Contact information for questions

Special Considerations

Minors – require both parental consent and child assent.
Vulnerable populations (prisoners, pregnant women, cognitively impaired, economically/educationally disadvantaged) – extra layers of protection must be specified in the protocol.
Deception in economics – generally not allowed. Giving objectively false information is prohibited by the discipline's dominant norm. If deception is unavoidable (rare), participants must be debriefed and the IRB informed. Deception reduces credibility in economics.
Digital consent – online experiments use digital consent forms; consent is ongoing throughout the experiment.

Institutional Review Board (IRB) – Purpose & Process

Composition: Minimum 5 members – scientists, non-scientists, and community members – who evaluate protocols independently.

Review categories:

flowchart TD
  A[Study submitted] --> B{IRB assesses risk level}
  B -->|Minimal risk<br>e.g. surveys, educational| C[Exempt – no full review]
  B -->|No more than minimal risk| D[Expedited review – quick approval/modifications]
  B -->|Greater than minimal risk| E[Full board review – thorough evaluation]

Most economics experiments fall into exempt or expedited; researchers should be prepared for a full review if the board deems necessary.

Submission requirements to an IRB:

Detailed IRB form (methodology, research protocol)
Consent form and recruitment materials
Risk-benefit analysis (the researcher's own assessment)
Data protection plan – who has access, cloud security, anonymisation/de-identification
Clear policies on data retention and destruction after publication

Data Management

Anonymise or de-identify data unless the research question requires identifiable data (justification needed).
Securely store and password-protect data.
Specify retention period (often a minimum of 3 years after study completion).

Ongoing Ethics Responsibility

Report any adverse events immediately to the IRB.
Annual review for studies lasting more than one year.
Any protocol changes must be communicated to the board via a corrigendum.
Researchers must maintain accurate records for at least 3 years and publish results responsibly.

Exam tip: The shift from historical violations to the Belmont principles is a classic exam question. Be able to list the three principles and how each maps to a modern requirement (e.g., respect for persons → informed consent; beneficence → risk-benefit analysis; justice → fair selection of participants).

Key takeaways

Four landmark historical violations (Nazi twins, Stanford Prison, Tuskegee, Milgram) drove creation of modern ethical standards.
The Belmont Report's three pillars: respect for persons, beneficence, justice.
Informed consent must include purpose, duration, procedures, risks, benefits, confidentiality, compensation, and right to withdraw.
IRB reviews studies in three categories: exempt, expedited, or full review.
Deception is largely prohibited in economics; if used, debriefing is mandatory.
Ethics is an ongoing responsibility – adverse event reporting, annual reviews, and careful data management are required.

The Project: Applying Behavioral Economics through a Randomized Experiment

The final deliverable of the course is a project that applies behavioral economics principles to understand the behavior of any economic agent – a firm, consumer, citizen, etc. The output is a slide deck (8–9 min presentation) with a detailed appendix (up to 10 slides). Start early.

Purpose

Apply BE concepts to a real-world decision.
Demonstrate the ability to design and execute a randomized experiment.
Collect primary data via Qualtrics and analyse results.

Key Components of the Project

1. Research Question & Predictions

Clearly define a research question about human behaviour (e.g., “Why do people litter?”).
Lay out predictions about how people will behave.

2. Experimental Design

Design a randomized experiment with at least one treatment and one control condition.
No simple survey without randomisation.
Use Qualtrics (built-in randomizer) to assign participants.
Collect data and test whether predictions are supported.

Exam tip: The randomization is mandatory. A survey without random assignment will not answer causal questions – your grade depends on it.

3. Slide Deck Structure (Template)

The following is the recommended outline. The example uses the COVID‑19 vaccine trial; your project must replace it with your own behavioural question.

Motivation / Context
- Describe the real-world setting.
- E.g., “The COVID‑19 pandemic caused widespread misery.”
Problem Statement
- What exactly are you testing?
- E.g., “Test whether the vaccine works (efficacy ≥ 80%).”
Why It Matters
- Even if the impact is small, explain why answering the question is useful.
Research Design
- Sample size: e.g., 30,000 participants (15,000 treatment, 15,000 control for Moderna).
- Treatment: receive the vaccine.
- Control: receive placebo.
- Explain why this design answers the problem statement.
- Main outcome variable: e.g., infection rate.
- Recommendation: one treatment, one control, and at most two outcomes → test at most two hypotheses.
Hypotheses
- Written as null hypotheses (negation), to be rejected at standard α = 0.05.
- Example: H₁: The vaccine does not decrease the rate of coronavirus infection.
- Each hypothesis requires one treatment, one control, and one outcome.
Results – Balance Table
- List covariates collected after the experiment (age, gender, religion, standard 10th/12th scores, income).
- Report means in treatment and control, the difference, and a p‑value from the appropriate test:
  - Continuous covariates (e.g., age) → t‑test
  - Binary covariates (e.g., female) → Chi‑square test
- Determine if the difference is significant (Y/N). We want no significant differences – this confirms randomization worked.
Results – Main Table
- Compare the outcome variable between treatment and control.
- Report means, difference, appropriate test, p‑value.
- Expect a significant difference if the treatment worked.
- Example: Moderna trial – 5 infections in vaccine group vs. 90 in placebo group → large, significant difference.
Conclusion
- Summarise results and offer key policy insights.
- Discuss limitations (no study is perfect).
Appendix (not presented; up to 10 slides)
- Details: sample location (e.g., Bangalore, Kolkata), recruitment source (friends, family, networks), ethical compliance (no IRB needed for this exercise, but consistent with course ethics), software used (Excel, Stata, Python), and any additional interesting results.
- Make it unambiguous.

Worked Example: Moderna Vaccine Trial

Aspect	Detail
Context	COVID‑19 pandemic
Problem	Test whether vaccine reduces infection rate
Design	30,000 participants; 15,000 vaccine (treatment), 15,000 placebo (control)
Null hypothesis	Vaccine does not decrease infection rate
Outcome	Infection rate
Results (CNN)	Vaccine: 5 infected; Placebo: 90 infected
Conclusion	Large, statistically significant difference → reject null. Vaccine works.

This result was published and widely covered; it illustrates a successful causal experiment.

Visual: Process Flow

flowchart TD
  A[Define research question & predictions] --> B[Design randomized experiment: treatment + control]
  B --> C[Implement in Qualtrics with randomizer]
  C --> D[Collect data]
  D --> E[Create balance table: test covariate similarity]
  E --> F{Any significant difference?}
  F -->|Yes| G[Randomization may be compromised – check design]
  F -->|No| H[Create main results table: test outcome difference]
  H --> I[Conclude: reject null if significant, else fail to reject]
  I --> J[Write policy insights & limitations]
  J --> K[Prepare slide deck + appendix]

Exam Tips & Common Pitfalls

Hypothesis must be written as a null (negation) – this is a standard requirement for hypothesis testing.
One treatment, one control, one outcome per hypothesis. You can have up to two outcomes (e.g., infection rate + fever) and test two hypotheses, but no more is advisable.
Balance table is critical – if a covariate shows a significant difference, the credibility of your causal claim weakens.
Use the correct statistical test for each variable type (t‑test for continuous, chi‑square for binary).
Appendix should be dense – include everything needed to replicate your study (sample source, ethics, software, extra results). This is where detail goes, not in the main deck.

Key Takeaways

The project requires a clear behavioural research question and a randomized experiment with one treatment and one control.
Hypotheses are null (negation) to be rejected at α = 0.05.
Balance table checks that covariates are similar across groups; main results table tests the outcome.
Use t‑test for continuous variables, chi‑square for binary.
The slide deck must be 8–9 min with a detailed appendix (up to 10 slides) covering sample, ethics, software, and extra results.
The Moderna trial (5 vs. 90 infections) exemplifies a successful causal test published in mainstream media.