Nudge Theory

What it is

Nudge theory begins with a contrast between two pictures of the chooser. In the economist's picture, people weigh options consistently and are unmoved by how the options are laid out. In the behavioural picture that Thaler spent his career documenting and that Nudge popularised, people are moved by the order of a menu, the box already ticked on a form and the option that requires no action. If presentation moves choices, then somebody is always shaping them, and the only open question is whether the shaping is done with care.

Thaler and Sunstein define a nudge as any aspect of the choice architecture that alters behaviour in a predictable way without forbidding any options or significantly changing economic incentives. The definition carries a condition: to count as a mere nudge, the intervention must be easy and cheap to avoid. The authors state what falls outside it, which is taxes, fines, subsidies, bans and mandates. Putting fruit at eye level is a nudge. Banning junk food is not. That boundary is what separates nudge theory from ordinary regulation, and it is also the part of the idea that has been argued over most.

Reading the book closely, the theory is better treated as four claims. The first is about people: they depart from the rational-agent model in predictable ways. The second is about design: every presentation of options steers, so no neutral arrangement exists, which is the claim called choice architecture. The third is about method: steer, but leave every option open and keep the exit cheap, a position the authors named libertarian paternalism in 2003. The fourth is about effect: small and inexpensive changes in presentation shift behaviour at scale.

The distinction matters because the four claims face different tests. Claims two and three are arguments. Declining to design a choice is itself a design, and how far a designer may go is a question of ethics and policy, so no effect size can settle either. Claim one is a body of empirical findings of uneven quality. Claim four is the one that field experiments and meta-analyses can contradict, and public discussion of nudge theory tends to treat it as settled.

The authors have revised the book since 2008. The 2021 Final Edition adds a chapter on sludge, which they treat as friction placed where it works against the person moving through the system, and it makes the cheap-exit condition a standard that a given nudge either meets or fails. A nudge whose opt-out is hard to use is reclassified as sludge and its cost counted against the policy. The toolkit changed as well: the earlier acronym was dropped, and the current list includes making things easy, defaults, anticipating error and giving feedback.

In effect

Automatic enrolment in a pension plan

The standard research case is a change from opt-in to opt-out. Brigitte Madrian and Dennis Shea, in a 2001 paper in the Quarterly Journal of Economics, studied a large company that moved its pension plan from opt-in to automatic enrolment and found that participation rose sharply. The plan on offer had not changed and employees could still leave, which is why the case is used as the clean example of a nudge. The default effect is the kind of nudge that the 2022 reanalyses of the nudge literature were most willing to except from their sceptical conclusions.

Fruit at eye level

The everyday case is a canteen counter. Place fruit where people reach first and puddings further along, and what people take changes, although nothing has been removed from the menu and nobody has been told anything. It passes the test because the dessert is still there and taking it costs nothing extra. It also shows why the theory has critics: a person who never notices the arrangement has not exercised much choice in either direction.

Defaults in products and subscriptions

The commercial versions are the pre-selected shipping option, the renewal that renews itself and the privacy setting that starts at maximum sharing. They use the same device as the pension default. Under the 2021 edition's own standard, they stop being nudges when leaving is hard, and the authors call that sludge. A default is only as defensible as its exit.

Trials run by two US nudge units

A third kind of evidence comes from organisations that run nudges at scale. Stefano DellaVigna, with a co-author, assembled 126 randomised trials covering 23 million people, every trial run by two US nudge units, and compared them with nudge trials published in academic journals. The journal trials averaged 8.7 percentage points of extra take-up. The nudge-unit trials averaged 1.4 percentage points, still statistically significant. The paper's model attributes about 70 per cent of the gap to selective publication in the academic sample.

What it does not say

It does not say that nudges are large. A 2022 meta-analysis by Stephanie Mertens and colleagues pooled more than 200 studies and reported a small to medium average effect, with moderate publication bias in the same abstract. Maximilian Maier and colleagues corrected for that bias and reported a pooled effect of 0.04. Barnabas Szaszi and colleagues applied three other correction methods and reported averages between minus 0.01 and 0.08. The disagreement is over what the published literature can support, and it is conducted as a methodological argument in print.

It does not say that a result from a journal study is a forecast for a rollout. The gap between the journal trials and the nudge-unit trials is the plain illustration. Any single nudge has to be judged in its own setting.

It does not say that nudges are enough. In the Final Edition, the authors concede that light-touch methods cannot solve climate change on their own, and that getting prices right matters more. They also report that two features of their best-known savings programme, tying savings increases to pay rises and letting people decide now to join later, were attractive but not essential.

It does not say that every default is acceptable. The authors rewrote their chapter on organ donation because many readers believed they supported presumed consent, and they state that they oppose the policy. Their reasoning applies the method's own test: an opt-out that many people never learn about is not a real opt-out.

---