- Cognitive biases
- Behavioural economics and policy · coming soon
- Government failure
One form, two results
In 1998, a large company in the United States changed one detail of its retirement savings plan. Before, a new worker who wanted to join had to fill in a form. After, every new worker was enrolled automatically, and anyone who did not want to save could leave by filling in a form. Nothing else changed: the same funds, the same employer contributions, the same freedom.
The economists Brigitte Madrian and Dennis Shea compared workers hired just before and just after the change. Among those hired before, about 37% had joined the plan. Among those hired after, 86% were in it (Quarterly Journal of Economics, 2001).
Now think about what the last four lectures would predict. Their consumer compares every bundle and picks the best one she can afford. In that model, it cannot matter which box is ticked when the form arrives. A worker who wants to save fills in one form; a worker who does not fills in the other. The default should change almost nothing. Yet it more than doubled participation.
That gap is where behavioural economics begins.
What the rational model leaves out
The consumer theory you have built rests on assumptions: stable preferences, known prices and options, and the best affordable bundle chosen. These assumptions predict a great deal correctly: demand curves slope down, and taxes change what people buy.
But the model says nothing about how a choice is presented. Which option is already ticked? Which item is at eye level? In the model, these details cannot matter, because the options and prices are the same.
Experiments and field studies show that they do matter. They matter in four ways that keep returning:
- Inertia. People tend to stay with whatever happens if they do nothing, even when changing is free. Partly people put things off; partly a default looks like a recommendation. Madrian and Shea found that many workers treated the company's default as advice.
- Framing. The same facts, described differently, lead to different choices.
- Limited attention. People notice what is easy to see, and skip what takes effort to find.
- Social influence. People are guided by what others like them do.
The important word is systematic. These are not random mistakes that cancel out across a population. They push most people in the same direction. And anything that pushes in a known direction can be used on purpose. The next lecture looks at the biases themselves. This one looks at the tool built from them.
Choice architecture and the nudge
In their book Nudge (2008), the economist Richard Thaler and the legal scholar Cass Sunstein gave the tool a name. They noticed that someone always decides how a choice is presented: the order of the options, the default, the wording of the form. They called that design choice architecture, and the person who designs it a choice architect. A canteen manager is one. So is whoever designs a pension form or a checkout page.
A nudge is any part of the choice architecture that changes behaviour in a predictable direction, without removing any option or significantly changing the economic incentives.
A nudge must be easy and cheap to avoid.
Each part of the definition does work. "Without removing any option" separates a nudge from a rule: a ban removes an option, and a mandate requires one. "Without significantly changing the incentives" separates a nudge from an incentive: a tax or subsidy changes the price. Only the presentation is left, so a nudge works only through the four patterns above.
The lines between the boxes are not always sharp; Try it yourself has a borderline case. Practise the main cases first.
Put each policy in a box. Read the reason after each choice. A card can go back if you change your mind.
Nine policies, sorted. A nudge changes only how the choice is presented; an incentive changes the price or reward; a rule removes or requires an option.
- Nudge: automatic pension enrolment with a one-form opt-out; fruit at eye level; the tax letter's social norm; donor registration unless you opt out. Insurance pre-added at checkout is a nudge in form, but it serves the seller.
- Incentive: a 20% tax on sugary drinks; a $500 bonus for joining a pension.
- Rule: a smoking ban in restaurants; a compulsory national savings fund.
The last card is the warning. A design can satisfy the definition of a nudge and still work against the chooser. Pre-ticked add-ons and hard-to-find cancellation pages use inertia and limited attention too. Thaler later called this kind of design sludge. The tool is the same; the direction is different.
The tools of the choice architect
Defaults. The strongest tool, because inertia works on everyone who does nothing. That is the pension result. It also explains a famous comparison of organ donor registers. In Germany, where people must opt in, about 12% were registered as willing donors. In Austria, where people are registered unless they opt out, the figure was 99.98% (Johnson and Goldstein, Science, 2003).
Austria registers about eight times as many people as willing donors as Germany does. Does Austria have about eight times as many organ donations for each person?
Show the answer
No. A registration rate is not a donation rate. Donation needs a death in a hospital in the right circumstances, a transplant system that can act quickly, and usually the agreement of the family, which most systems still ask for. A study of 22 countries from 1993 to 2002 allowed for these other causes. It found that countries with an opt-out law had donation rates about 25–30% higher (Abadie and Gay, Journal of Health Economics, 2006). That is a real effect, but it is nowhere near eight times. Always ask what was actually measured.
Framing. "90% of patients survive" and "10% of patients die" state the same fact. In experiments, people choose a treatment more often under the survival description. Whoever describes the treatment must choose a frame.
Salience and simplification. Making the better option easier to see or to do: fruit at eye level, a shorter form, a colour-coded food label.
Social norms. Telling people what most others do. The firm Opower mailed home energy reports comparing each household's electricity use with that of similar neighbours. In randomised trials covering about 600 000 households in the United States, the reports cut electricity use by 2.0% on average. The heaviest users cut by 6.3%, the lightest by only 0.3% (Allcott, Journal of Public Economics, 2011). Allcott estimated that the effect matched a short-run electricity price rise of about 11–20%.
Commitment devices. A way to bind your future self. In the Save More Tomorrow plan, workers agreed in advance that part of each future pay rise would go into savings. Saving more later feels easier than saving more now, and inertia keeps the plan running. In the first company to use it, the average saving rate of those who joined rose from 3.5% to 13.6% of pay. This took 40 months (Thaler and Benartzi, Journal of Political Economy, 2004).
Why a default decides so much
Here is the hard part of the lecture. A default changes nothing for a worker who decides actively: she simply picks what she wants. It changes everything for a worker who does not act. So the power of a default depends on one number: how many people do not act. The model below lets you change that number.
An invented firm hires 1 000 workers. Attentive workers decide for themselves: 70% of them want to join the pension at 7% of pay; the rest stay out. Inattentive workers do whatever the default says. Choose the default, the default rate, and how many workers are inattentive.
Default:
Default rate:
Inattentive workers: 50%
The model's results for an invented firm of 1 000 workers. Attentive workers: 70% join at 7% of pay. Inattentive workers follow the default.
| Inattentive | Default | Participation | Average contribution |
|---|---|---|---|
| 0% | either | 700 | 4.9% |
| 50% | opt-in | 350 | 2.45% |
| 50% | opt-out at 3% | 850 | 3.95% |
| 50% | opt-out at 6% | 850 | 5.45% |
With half the workers inattentive, switching the default from opt-in to opt-out raises participation from 350 to 850, close to the real results in the figure below. Under opt-out, 500 of the 850 savers sit at the default rate, so the default rate largely decides how much the firm's workers save.
Look at three settings. With no inattentive workers, the default changes nothing: that is the rational model. With half the workers inattentive, switching to opt-out raises participation from 350 to 850, close to the real jump in the figure below. And under opt-out, most savers sit at the default rate. Raising the default from 3% to 6% raises the average saving rate from 3.95% to 5.45% of pay without anyone being asked.
That is it. That is the whole idea. A default does not change what attentive people want. It decides the outcome for everyone who does not choose. Madrian and Shea saw the same second effect in the real firm. Many automatically enrolled workers stayed at the default rate of 3% and in the default fund, a combination few earlier workers had chosen.
The model leaves things out. Real workers are not simply attentive or inattentive; some act later, after a reminder. Some attentive workers are also pulled towards the default rate, because it looks like advice. And the 70% and 7% are invented.
Is nudging right? Libertarian paternalism
Thaler and Sunstein called their position libertarian paternalism. It is paternalist because the choice architect steers people towards what she believes is better for them. It is libertarian because nobody is forced: every option stays open. Their test for a good nudge is that it makes people better off as judged by themselves.
Their main argument is that there is no neutral design. The form must have some default. The canteen must put something at eye level. So the question is not whether to influence choices, but in which direction and in whose interest.
Critics raise three points. First, who decides what is "better"? A choice architect can be wrong, or can serve her own interests, as sludge shows. Second, some nudges work best when people do not notice them, which looks like manipulation. Defenders reply that good nudges can be fully public, like a pension default announced to every worker. Third, the effects may be small. A study of 126 trials run by two government nudge units in the United States found an average effect of about 1.4 percentage points, against 8.7 percentage points in published academic papers. Most of the gap came from selective publication: small studies with large effects were the ones that got published (DellaVigna and Linos, Econometrica, 2022). Defaults remain the strongest and most reliable tool.
Worked examples
An invented firm hires 2 000 workers a year, each earning $40 000. The firm adds 50 cents for every dollar a worker saves, up to 6% of pay. Under opt-in, 37% join. Under opt-out, 86% are in. (a) How many extra savers does the default create each year? (b) What does a worker who does not join lose from the firm's contribution alone? (c) What does the rational model say about this?
(a) 2 000 × (0.86 − 0.37) = 2 000 × 0.49 = 980 extra savers a year.
(b) Saving 6% of $40 000 is $2 400. The firm adds half of that: $1 200 a year, which the worker gives up by not joining.
(c) In the rational model, a worker joins if the gain is bigger than the cost of acting. The cost is a short form; the gain includes $1 200 a year of free money. So the default should hardly matter. A gap of 49 percentage points cannot be explained by the cost of a form. It needs inertia: people put off the decision, or read the default as advice.
Check. 0.37 × 2 000 = 740 and 0.86 × 2 000 = 1 720; 1 720 − 740 = 980 ✓.
A household uses 30 kWh of electricity a day (invented). (a) Using Allcott's average effect of 2.0%, how much does a home energy report save it in a day and in a year? (b) What would a heavy user and a light user save, using 6.3% and 0.3%?
(a) 30 × 0.02 = 0.6 kWh a day, and 0.6 × 365 = 219 kWh a year. This is close to Allcott's own average of about 0.62 kWh a day.
(b) At 6.3%, 30 × 0.063 ≈ 1.89 kWh a day. At 0.3%, 30 × 0.003 = 0.09 kWh a day. The same letter does 21 times as much for the heavy user, because it tells her that she uses more than her neighbours.
Reading it. A 2% cut sounds small. But a letter costs very little and reaches millions of homes, so a nudge must be judged against its cost and reach.
Check. 1.89 ÷ 0.09 = 21 ✓.
Real case: automatic enrolment in the United Kingdom
The policy. From October 2012, phased in until February 2018, employers in the United Kingdom had to enrol eligible workers into a workplace pension automatically. Eligible means aged 22 up to the State Pension age and earning above a threshold (currently £10 000 a year). The employer must contribute. Each worker may opt out.
The prediction. The rational model predicts little change for workers: anyone who wanted a pension could already ask for one. The behavioural model predicts a large rise in participation, and many savers staying at the minimum contribution.
What happened. Among eligible private-sector employees in Great Britain, participation rose from 42% in 2012 to 86% in 2020 (DWP, 2021). Among all eligible employees it was 88% in 2023 (DWP). The Institute for Fiscal Studies found that among eligible 22- to 25-year-olds, participation rose from 20% to 88%. And the second prediction also held: in 2023, about three in ten private-sector savers in these pensions were saving at the automatic-enrolment minimum level (DWP analysis of 2023 data).
Verdict. As a way to make saving normal, automatic enrolment worked on a national scale. Two cautions complete the evaluation. First, it is not a pure nudge. For employers it is a rule, because they must enrol workers and contribute. Part of the rise came from employers who had offered no pension before. Second, the default anchors many savers at the minimum, which may be too little for retirement. The policy succeeded on participation and left a new question about adequacy.
A tax on sugary drinks, a subsidy for solar panels and a smoking ban are not nudges. A tax or subsidy changes the price (an incentive). A ban or a requirement changes the options (a rule). Test every example against both parts of the definition.
99.98% registered as donors in Austria does not mean 99.98% of deaths lead to donation. Separate what a nudge changes (a registration, a ticked box) from the outcome that matters (donations, money saved for retirement), and quote evidence about the outcome.
Behavioural economics finds predictable patterns, not stupidity, and following a default can be a sensible shortcut. The case for nudging is that some design is unavoidable. It supports better defaults, not taking choices away.
Work each one on paper first. Open a hint only when you are stuck.
-
Classify each policy as a nudge, an incentive or a rule, with a reason. (a) Office printers set to print on both sides of the paper unless the user changes the setting. (b) A minimum age of 18 for buying alcohol. (c) Calorie counts printed next to each item on a restaurant menu. (d) A charge for plastic bags at shop tills.
Hint 1
Ask: is any option removed or required? Is the price changed?
Hint 2
For (d), the charge is a price, however small. Say why it is a borderline case.
Full answer
(a) Nudge: a default; anyone can print on one side with one click. (b) Rule: buyers under 18 lose the option. (c) Nudge: it adds information and makes calories easy to see; nothing is removed and prices are unchanged. (d) Strictly an incentive, because the charge is a price. It is borderline: the charge is too small to matter to most budgets, and much of its effect seems to come from making people notice the bag. Check: each answer used both tests, options and prices ✓.
-
An invented firm has 800 workers. 40% are attentive: three quarters of them join the pension at 8% of pay, and the rest stay out. The other 60% follow the default. For opt-in, and for opt-out with a default rate of 4%, find (a) the number of savers and (b) the average contribution across all 800 workers.
Hint 1
Attentive joiners: 0.4 × 0.75 × 800. They behave the same under both defaults.
Hint 2
Under opt-out, the 60% inattentive also join, at 4%.
Full answer
Attentive joiners: 0.4 × 0.75 × 800 = 240. Opt-in: (a) 240 savers (30%); (b) average 0.3 × 8 = 2.4% of pay. Opt-out: the 0.6 × 800 = 480 inattentive workers also join, so (a) 240 + 480 = 720 savers (90%); (b) average 2.4 + 0.6 × 4 = 4.8% of pay. Two thirds of the savers (480 of 720) are at the default rate. Check: (240 × 8 + 480 × 4) ÷ 800 = (1 920 + 1 920) ÷ 800 = 4.8 ✓.
-
A city has 50 000 households, each using 25 kWh of electricity a day (invented). Monthly energy reports cut use by 2.0%, and cost $1 per report. (a) How much electricity is saved in a year? (b) What is the cost for each kWh saved? (c) Allcott found the effect matched a price rise of about 11–20%. Give one reason a government might prefer the reports to such a price rise.
Hint 1
Daily saving per household: 25 × 0.02. Then multiply by 50 000 and by 365.
Hint 2
Yearly cost: 50 000 households × 12 reports × $1.
Full answer
(a) 25 × 0.02 = 0.5 kWh per household per day; 0.5 × 50 000 × 365 = 9 125 000 kWh, about 9.1 million kWh a year. (b) Cost $600 000 a year, so $600 000 ÷ 9 125 000 ≈ $0.066 per kWh saved. (c) A price rise makes every household poorer, including poor ones who already use little, and it is politically hard to pass. The report costs little and leaves prices unchanged. (A good answer also notes the limit: the reports did little for light users.) Check: 9 125 000 × 0.066 ≈ 600 000 ✓.
-
A phone shop adds insurance at $5 a month to every sale; buyers must untick a box to remove it. Of 10 000 buyers a year, 60% keep it. When buyers had to tick the box to add it, only 10% chose it. (a) How much extra insurance revenue does the pre-ticked box bring in a year? (b) Is this a nudge by Thaler and Sunstein's definition? (c) Does it pass their test for a good nudge? What would you recommend?
Hint 1
Extra buyers: 10 000 × (0.60 − 0.10). Each pays $5 for 12 months.
Hint 2
Their test asks whether people are better off "as judged by themselves". What did buyers choose when they had to decide actively?
Full answer
(a) 10 000 × 0.5 = 5 000 extra policies; 5 000 × $5 × 12 = $300 000 a year. (b) In form, yes: no option is removed, the price is unchanged, and unticking costs one click. (c) No. When buyers chose actively, only 10% wanted the insurance, so most of the extra 50% did not want it. They are worse off as judged by themselves. This is sludge: a default that serves the seller. A recommendation could be an unticked box (opt-in) or a required active choice ("Do you want insurance? Yes / No"), which removes the inertia without steering. Check: 0.6 × 10 000 = 6 000 and 0.1 × 10 000 = 1 000, so 5 000 extra ✓.
Economics has usually changed behaviour through prices: taxes, subsidies and fines. Nudge theory adds a second lever, design, and it is a lever that is already being pulled, because every form, menu and checkout page has some design. This changes how economists think about markets and about government. A market can fail not only because of external costs or missing information, but because choices are presented in a way that works against the people making them. And a government can fail in a new way too, by nudging badly or in its own interest. The next lecture opens the box: the specific biases, such as present bias, anchoring and loss aversion, that make defaults and frames so powerful.
- Choice architecture is the way a choice is presented, and every choice has one, so the designer always influences the outcome.
- A nudge changes behaviour in a predictable direction without removing options or significantly changing incentives, and it must be easy and cheap to avoid.
- A default decides the outcome for everyone who does not act, so its power depends on how many people do not choose actively.
- Defaults, framing, salience, social norms and commitment devices work through systematic patterns that the rational consumer model leaves out.
- Libertarian paternalism defends nudges that make people better off as judged by themselves, while critics question who decides, how open nudges are, and how large their effects are.
- A nudge must be judged by its effect on the outcome that matters, such as donations or money saved, not only on the box people tick.