Where rational choice breaks: bounded rationality and self-control, anchoring, framing, loss aversion, present bias, defaults and social norms; the prospect theory value function; nudges, choice architecture and how they compare with taxes and regulation.
Before 2012 an employee who wanted a workplace pension had to fill in a form. Automatic enrolment reversed the default: employers must enrol eligible workers, who stay in unless they opt out. By 2025, 22.6 million eligible employees, nine in ten, were saving, up from 55% in 2012; total workplace pension saving reached £166.1bn a year. Among 22 to 29 year olds participation went from about a quarter to 85%.
Nothing in the standard economic model predicts this. The scheme was always available; the employer contribution was always on offer. People simply did not act. Opt-out rates run at 11 to 12%, and participation drops sharply below the £10,000 earnings trigger at which enrolment stops being automatic, which shows how much of the saving is the default at work rather than a change of mind.
Why do people behave like this, and what does it mean for how governments change behaviour?
Source: Department for Work and Pensions, Workplace pension participation and savings trends of employees 2009 to 2025, 30 July 2026.
The traditional model assumes consumers are rational: they maximise utility with consistent preferences, full information and the self-control to act on their plans. Behavioural economics studies how real decisions depart from that. Bounded rationality (Herbert Simon): limited information, time and processing power mean people satisfice, choosing something good enough, and rely on heuristics, rules of thumb. Bounded self-control: people know what they should do and fail to do it.
The main biases: anchoring on the first number seen; availability, judging likelihood by what comes to mind; loss aversion, losses hurting roughly twice as much as equal gains please; framing, the same facts choosing differently depending on presentation; present bias, overweighting now against later; status quo and default bias, sticking with what is set; social norms and herd behaviour, doing what others do.
Policy follows. Choice architecture is the way options are presented. A nudge changes behaviour by altering the architecture without banning options or changing prices much: default choices (auto-enrolment), framing and information (traffic-light food labels), social norms (“most people pay on time”). Stronger tools: restricted choice (limits on where and when) and mandated choice (you must decide).
In 2012, 55% of eligible employees were saving into a workplace pension. In 2025, 90% were.
Pensions did not become more generous. Contribution rates were not cut. Nobody was forced to join.
What changed?
Automatic enrolment, phased in from 2012. A rational consumer would save the same whether the default is “in” or “out”. Real people stick with the default: inertia, present bias and the status quo bias did what decades of advertising could not. That is behavioural economics.
The diagram behavioural economists use. Outcomes are measured from a reference point, usually where you are now, not from total wealth. On the gains side the curve is concave: the first £50 feels like +62, the second adds only +38. That is diminishing sensitivity, the behavioural cousin of diminishing marginal utility. Rational choice says only the total matters; people judge changes from where they started.
Check: A shopper picks the second-cheapest wine on the list without comparing the others. This is best described as:
Bounded rationality means people cannot or will not process everything, so they use heuristics. The second-cheapest rule avoids both the risk of the cheapest and the cost of comparing.
Loss aversion. The loss side is steeper: −£50 feels like −139, against +62 for a £50 gain, about 2.2 times as strong. People refuse fair bets, hold losing shares to avoid “realising” the loss, and value what they own more than what they could buy: the endowment effect. It also explains why a “£5 charge for paying by card” stings more than a “£5 discount for cash” pleases, though they are the same.
Check: Offered a coin toss that wins £150 or loses £100, most people refuse although the expected value is +£25. The best explanation is:
A loss of £100 that feels like −225 outweighs a gain of £150 that feels like about +130. Even people who do the arithmetic refuse.
Framing moves the reference point. Call a surcharge a “discount for cash” and the same £5 sits on the gentle gains slope rather than the steep loss slope. Anchoring fixes the starting point: a “was £80, now £50” label makes £50 feel like a £30 gain. Present bias is a different distortion: the value of now is inflated against later, so people under-save and over-borrow while fully intending the opposite.
Teacher tip: Ask students which they would choose: a sure £50, or a 50% chance of £100. Then: a sure loss of £50, or a 50% chance of losing £100. Most are risk-averse in gains and risk-seeking in losses, exactly as the curve predicts. It makes loss aversion felt rather than described.
Check: A yoghurt labelled “90% fat-free” sells better than the same yoghurt labelled “10% fat”. This is:
The two labels carry identical information. The first frames it as a gain relative to a fatty alternative; the second as a loss relative to fat-free. Loss aversion does the rest.
Default bias. Whatever option applies when you do nothing wins, because doing nothing is free and now, while switching costs effort and its payoff is later. Choice architecture is the design of that default and everything around it: order on a menu, pre-ticked boxes, the font on a letter. Organ donation in England moved to opt-out in 2020 for the same reason. A nudge changes the architecture; it leaves every option open.
Check: Automatic enrolment raised pension participation from 55% to 90% without compulsion or new incentives. The biases it exploits are mainly:
Rational savers would choose the same regardless of the default. The default works because changing it takes effort today for a benefit decades away.
Policy. Traditional tools change prices (taxes, subsidies) or remove options (regulation). Nudges keep both and change presentation: defaults (auto-enrolment), framing and information (traffic-light labels, cigarette packaging), social norms (HMRC letters saying most people pay on time raised payment rates), salience (smart meters). Mandated choice makes you decide; restricted choice limits where or when, like minimum unit pricing hours.
Check: Which is a nudge rather than a traditional policy instrument?
A nudge alters choice architecture without changing prices much or removing options. Taxes and subsidies change prices; bans remove options.
Where this leaves the model. The rational consumer is still the benchmark: it predicts well when stakes are high, choices repeat and feedback is quick. Behavioural economics explains the gaps: one-off, complex, delayed decisions such as pensions, insurance and health. The Behavioural Insights Team, set up in 2010, tests nudges with randomised trials, so evidence replaces assumption. Both views belong in an answer.
Check: The strongest criticism of relying on nudges to correct market failure is:
A nudge on packaging does not change the price of a cigarette; where an externality is large, the price has to. Nudges complement, rather than replace, taxes and regulation.
The loss curve mirrors the gain curve, so −£50 feels like −62, the same strength as +£50. That is the rational, symmetric model. Loss aversion means the loss curve is steeper: −£50 feels like about −139.
Evaluate the view that nudges are a more effective way of changing consumer behaviour than taxes and regulation.
A nudge changes behaviour by altering the choice architecture, the way options are presented, without removing options or changing prices significantly. Taxes and subsidies work through prices; regulation removes or restricts options. Nudges draw on behavioural economics: real consumers show bounded rationality and bounded self-control, stick with defaults, overweight the present, and feel losses about twice as strongly as gains.
The case that nudges are more effective rests on evidence. Automatic enrolment raised the share of eligible employees saving into a workplace pension from 55% in 2012 to 90% in 2025, with no compulsion and no new incentive: the default was switched, and inertia and present bias, which had kept people out, now kept them in. Letters telling late taxpayers that most people had already paid raised payment rates; England moved organ donation to opt-out in 2020. Nudges are cheap, they preserve freedom of choice, which makes them politically easier than a tax, and because they are small interventions they can be tested in randomised trials, so policy is built on what works rather than on an assumed rational consumer.
**However,** effectiveness depends on the problem. A nudge leaves the price signal untouched: a warning on a packet does not internalise the external cost of smoking the way duty does, and where an externality is large, prices and rules must do the heavy lifting. Many nudge effects are small or fade; auto-enrolment got people in but anchored them to an 8% default contribution that is widely judged too low, and opt-out rates have crept up to 11 to 12%. Nudges can also be manipulative: the same techniques are used by firms in subscription traps and gambling apps, which is why regulators now police choice architecture itself.
On balance nudges are an effective complement, not a substitute. They work best for one-off, delayed, complex decisions where inertia does the damage, such as pensions; taxes and regulation remain more effective where the harm is large, immediate and priced. The best policy usually combines them, as the sugar levy and reformulation did.
Pension figures: Department for Work and Pensions, Workplace pension participation and savings trends of employees 2009 to 2025, published 30 July 2026. The value function on the diagram follows Kahneman and Tversky’s prospect theory with a loss-aversion ratio of about 2.25; the pound values are illustrative. Diagrams drawn with Rough.js and perfect-freehand, both MIT licensed.