A nudge is an intervention that changes the way a choice is presented or organized in order to influence behavior, without forbidding options or substantially changing economic incentives. Examples include automatically enrolling employees in a retirement plan while allowing them to opt out, simplifying an application form, and sending a reminder shortly before a deadline. The concept belongs to behavioral economics and is closely associated with choice architecture: the design of the environments in which decisions are made.
Richard H. Thaler and Cass R. Sunstein popularized the term in their 2008 book Nudge: Improving Decisions about Health, Wealth, and Happiness. Its importance lies in connecting findings about attention, judgment, and self-control with practical questions of institutional design. Governments, employers, and businesses already determine defaults, arrange information, and create procedures; nudge theory treats those arrangements as consequential rather than neutral background conditions.
Nudges vary greatly in mechanism, purpose, and effectiveness. Some reduce obstacles to carrying out an existing intention, while others draw on social influence or established patterns of judgment. Their study therefore encompasses psychological explanation, experimental evaluation, public administration, commercial design, and questions about autonomy and welfare. Evidence supports particular interventions in particular settings, but not the proposition that every behavior can be predictably or cheaply changed through a nudge.
Definition and boundaries
The influential definition offered by Thaler and Sunstein describes a nudge as an aspect of choice architecture that changes behavior predictably without forbidding options or significantly changing economic incentives. Avoiding the intervention must remain easy and inexpensive. Placing fruit at eye level can qualify; banning less nutritious foods does not. Likewise, a reminder about an unpaid bill can be a nudge, whereas imposing a large financial penalty is a change in incentives.
The distinction concerns the instrument used, not necessarily the size of its behavioral effect. A default can produce a large change in participation even when opting out requires little effort. Conversely, a small tax remains a price intervention, even if its behavioral effect is modest. Policies often combine instruments: a retirement program may include automatic enrollment, an employer contribution, and a tax preference. Only the enrollment arrangement is straightforwardly a nudge; the financial advantages operate through additional mechanisms.
Choice architecture is broader than nudging. Every application form has a sequence, every menu has an ordering, and every online service has an interface. These features constitute an architecture whether or not anyone deliberately designed them to influence behavior. A nudge is an intervention within that architecture, usually identified relative to a specified alternative. The relevant comparison might be automatic enrollment versus active enrollment, or a short reminder versus no reminder. Without that counterfactual, describing a design as a nudge says little about what it changes.
The boundaries remain contested. Clear disclosure may influence behavior through deliberation rather than through a cognitive shortcut, but it is commonly included within a broad understanding of nudging. Educational programs that develop durable decision-making skills are often distinguished as “boosts,” because they change the decision-maker’s competence rather than merely the immediate environment. A voluntary commitment arrangement can also occupy a boundary: an invitation to make a plan may be a nudge, while a binding contract with a substantial forfeiture changes the person’s future incentives or options.
Nudges are therefore not synonymous with persuasion, advertising, or all noncoercive policy. Nor does the label establish that an intervention is beneficial. A design that increases retirement saving and one that makes an unwanted subscription easier to purchase may employ similar behavioral mechanisms. Classification identifies how an intervention works; judging its desirability requires separate evidence about purpose, consequences, and the interests of those affected.
Intellectual origins and institutional development
The intellectual background of nudging predates the term. Standard models of microeconomics often represent individuals as choosing consistently among available alternatives, given their preferences, information, and constraints. Behavioral research investigated systematic departures from those assumptions. Herbert A. Simon developed the concept of bounded rationality, emphasizing limits on information processing and the use of workable rather than globally optimal decision procedures. These ideas helped establish that the organization of a decision could matter independently of its formal payoff structure.
Research by Daniel Kahneman and Amos Tversky supplied further foundations. Their work on judgment under uncertainty documented the use of heuristics, while prospect theory, introduced in 1979, described decisions relative to reference points rather than solely in terms of final wealth. Framing effects showed that equivalent descriptions could elicit different responses. Thaler connected such findings to economic behavior, including saving, consumption, and the treatment of gains and losses. Research on mental accounting and self-control further explained why people might treat economically interchangeable resources differently or postpone actions they intended to complete.
Thaler and Sunstein developed the associated idea of libertarian paternalism in publications in 2003, before presenting it to a wider audience in Nudge. The expression combines two propositions: institutions may legitimately seek to improve people’s welfare, but should preserve freedom to choose otherwise. Their examples emphasized the impossibility of avoiding choice architecture altogether. A pension plan must specify what happens when an eligible employee does nothing, just as a cafeteria must arrange its food in some order.
Government adoption accelerated during the following decade. The United Kingdom established its Behavioural Insights Team in 2010. The team initially operated within the Cabinet Office and promoted the use of behavioral research and field experiments in public administration. In the United States, the Social and Behavioral Sciences Team was established in 2014, and a 2015 executive order encouraged federal agencies to apply behavioral science insights. Other governments and international organizations developed related units and programs.
This institutional history also broadened the field beyond nudges narrowly defined. Behavioral teams have evaluated communication, application procedures, service delivery, and organizational processes, sometimes using interventions that alter material conditions as well as presentation. The World Bank devoted its World Development Report 2015, titled Mind, Society, and Behavior, to the role of psychological and social influences in development. “Behavioral public policy” consequently describes a wider enterprise than nudging: it includes diagnosis, experimentation, institutional reform, and comparisons with conventional policy instruments.
Psychological and behavioral mechanisms
Nudges do not operate through a single psychological process. One important mechanism is limited attention. People encounter more information and demands than they can continuously process, so an intention may fail to produce action simply because it is not salient at the relevant moment. A timely reminder can bring a task back into attention, while a prominent deadline can distinguish an urgent action from background information. Such interventions need not change preferences or correct a mistaken belief; they can instead help translate an existing intention into behavior.
Defaults work through several potentially overlapping channels. Taking action requires effort, so some people remain with a preselected option because switching involves a practical or cognitive cost. Others interpret the default as an implicit recommendation from an informed institution. Still others evaluate departing from the default as giving up an existing position. Loss aversion and the endowment effect can contribute to that response, although neither is necessary for a default effect. Observing that a default changes behavior does not, by itself, identify which explanation is responsible.
Intertemporal choice provides another foundation. With present bias, an immediate inconvenience may receive disproportionate weight compared with a larger future benefit. Completing a pension form, arranging an appointment, or beginning an application can therefore be repeatedly postponed. A common representation is quasi-hyperbolic discounting:
where represents conventional discounting and represents additional discounting of outcomes beyond the present. This is a model of one possible mechanism, not a definition of nudging. Reminders, immediate opportunities to act, and advance planning can sometimes reduce the consequences of procrastination without changing the underlying preference parameters.
Social influence operates differently. A message may describe what others do, what others approve of, or both. Descriptive norms concern observed behavior; injunctive norms concern approval and disapproval. Telling people that most comparable households pay a bill on time may influence expectations about normal conduct. However, emphasizing the prevalence of undesirable behavior can inadvertently make that behavior appear common and acceptable. The credibility and relevance of the comparison group matter.
Some interventions primarily reduce transaction costs rather than exploit a bias. Shorter forms, clearer instructions, and fewer redundant steps conserve time and effort for anyone, including a fully attentive decision-maker. This distinction matters because a successful intervention does not necessarily demonstrate irrationality. Similar behavioral outcomes can arise from confusion, lack of information, distrust, limited resources, or genuine preferences. Effective analysis separates these explanations rather than treating every failure to act as a cognitive error.
Main forms of intervention and their design
Defaults specify the outcome that follows in the absence of an active choice. They may be universal, assigned according to observable characteristics, or based on a person’s previous selection. Their design includes not just the preselected option but also its visibility, the timing of disclosure, and the steps required to change it. Active choice is an alternative: rather than assigning an outcome, an institution requires or prominently invites a decision. This can be useful when preferences vary substantially and no single default is appropriate.
Simplification changes the work needed to understand or complete a task. Examples include replacing technical language, eliminating repeated requests for the same information, and combining several administrative stages into one process. Prefilled forms occupy a related category, using information already held by an institution to reduce entry burdens. However, simplification must be distinguished from omission. Removing decision-relevant qualifications may make a message shorter while making the resulting choice less informed.
Information design includes framing, ordering, and feedback. Annual rather than monthly costs may make long-term expenditure more salient; translating an energy quantity into an estimated bill can make an unfamiliar measure easier to interpret. Feedback connects behavior to consequences, such as showing recent consumption alongside a household’s own previous consumption. The effect depends on the comparison and the user’s ability to respond. Information about energy costs, for instance, has limited immediate usefulness when occupants cannot control heating systems or replace appliances.
Reminders and planning prompts address the gap between intention and execution. A reminder can identify a deadline and provide a direct route to action. A planning prompt asks a person to specify when, where, or how an intended action will occur. The practical difference is important: “remember to complete the form” and “choose a time tomorrow to complete the form” impose different cognitive tasks. Additional messages can also become intrusive or lose effectiveness through repetition, so frequency is part of the intervention rather than an incidental detail.
Design frameworks organize these techniques without constituting scientific laws. The Behavioural Insights Team’s EAST framework, published in 2014, groups interventions under “Easy, Attractive, Social, and Timely.” Such frameworks can help generate candidate designs, but they do not establish which mechanism is operating or whether an intervention will succeed. A complete specification identifies the target behavior, affected population, proposed mechanism, delivery channel, comparison condition, and outcome. “Send a behavioral message” is not a reproducible intervention unless its content, timing, and context are also defined.
Retirement saving, household finance, and consumer decisions
Retirement saving is one of the most influential applications of nudging. Under voluntary active enrollment, eligible employees must initiate participation; under automatic enrollment, contributions begin unless they decline. In a 2001 study, Brigitte Madrian and Dennis Shea examined a change to automatic enrollment at a large US employer. Participation was substantially higher among employees covered by the new arrangement, and many remained at the default contribution rate and investment allocation. The findings demonstrated that plan design affected not only whether employees saved but also how they saved.
This example reveals both the strength and the limits of defaults. A low default contribution may bring more employees into a plan while anchoring some participants below the rate they would otherwise have selected. A default investment may be suitable for many participants but not for all circumstances. Participation is therefore an intermediate outcome, not a complete measure of welfare. Relevant outcomes include total contributions, fees, investment risk, withdrawals, and the relationship between pension assets and other household liabilities.
Thaler and Shlomo Benartzi’s Save More Tomorrow program addressed a different obstacle. Participants could commit in advance to increase retirement contributions when their pay rose, making the increase less salient as a reduction in current take-home pay. Their 2004 report on an early implementation found that participants’ average saving rate rose from 3.5 percent to 13.6 percent over 40 months. The design combined advance commitment, inertia, and the timing of contribution increases. It was not simply an informational campaign telling workers that saving was desirable.
Financial decisions also illustrate interactions between nudges and broader household constraints. Increasing retirement contributions does not necessarily increase net wealth by the same amount if participants reduce other saving or accumulate additional debt. Similarly, a payment reminder may prevent a fee for someone who forgot a deadline but cannot provide funds to someone experiencing a cash shortage. Evaluations must distinguish failures of attention from affordability and other material limitations.
Commercial settings introduce conflicting objectives. A provider may design an interface to increase purchases, while a customer may prefer lower expenditure or easier comparison. Preselected add-ons, prominently displayed installment amounts, and a highlighted minimum credit-card payment can shape decisions without removing alternatives. Some arrangements qualify as nudges under a purely descriptive definition; others introduce substantial obstacles or conceal relevant information. Financial regulation therefore concerns not only whether choice remains formally available, but also disclosure, suitability, conflicts of interest, and the practical cost of choosing differently.
Public services, environmental behavior, and collective outcomes
Public agencies use nudges in settings where an individual’s intention must be translated into a completed administrative action. Examples include benefit applications, appointment attendance, license renewal, and payment of taxes. A redesigned letter may state the required action more clearly, identify the deadline, and explain the available payment routes. In UK tax experiments reported by Michael Hallsworth and colleagues in 2017, messages invoking payment norms increased payment among some recipients. The findings concerned particular messages and taxpayer populations, rather than a universal effect of mentioning social norms.
Access to public services shows why information alone may be insufficient. In a US experiment reported in 2012, Eric Bettinger and colleagues studied assistance with the Free Application for Federal Student Aid. Help completing the application, combined with information about aid eligibility, increased application completion and college attendance among relevant groups. Providing information without comparable application assistance did not produce the same results. The intervention demonstrated the importance of procedural burdens, but also the analytical difficulty of treating all successful assistance as a minimal nudge.
In public health, appointment reminders and streamlined booking systems can reduce missed opportunities to use services. Their immediate outcomes—opening a message, booking an appointment, or attending—must be separated from later health outcomes. A message cannot remedy a shortage of appointments, an inaccessible location, or an unaffordable service. Behavioral design can complement a functioning health system, but effects depend on service capacity and the circumstances of the population receiving it.
Household energy reports provide a well-studied environmental application. Reports comparing a household’s electricity consumption with that of similar households, often accompanied by conservation information, have been evaluated in large field experiments. Hunt Allcott’s 2011 analysis found average electricity-use reductions of about 2 percent across the programs studied. Such effects are modest at the household level but potentially consequential when delivered widely. Their persistence depends partly on continued communication, household adaptation, and changes in the underlying opportunities for conservation.
Environmental applications also highlight the difference between private and collective welfare. A reduction in energy use may lower a household’s bill and reduce an externality, but the size of the environmental benefit depends on how electricity is generated and when demand falls. Nudges differ from emissions standards, infrastructure investment, and a Pigouvian tax, which changes the price of harmful activity. Whether a behavioral intervention complements or substitutes for these instruments is an empirical and institutional question, not something established by its low delivery cost.
Evaluation, effect sizes, and the accumulation of evidence
A nudge is a causal claim: behavior would differ if the choice environment were changed. Randomized controlled trials are widely used because randomly assigning comparable individuals or groups to alternative conditions helps isolate that effect. For a binary outcome such as application completion, an experiment might estimate
where and are the outcome proportions in the treatment and control groups. An increase from 10 percent to 12 percent is a two-percentage-point effect and a 20 percent relative increase; reporting only the latter can exaggerate the practical impression.
Evaluation must specify whose behavior is measured and over what period. An intention-to-treat estimate includes everyone assigned to an intervention, even if some never read the message or encounter the redesigned interface. Restricting analysis to people who opened an email can destroy the comparability produced by randomization. When people influence one another or an institution changes its practices for an entire office, assignment may need to occur at a group level. Statistical uncertainty must then reflect that design.
The evidence base contains many effective interventions, but its average magnitude is disputed. A 2022 PNAS meta-analysis by Stephanie Mertens and colleagues reported a positive overall effect of choice-architecture interventions. Subsequent published analyses questioned how much of the estimated average reflected publication bias and other features of the available literature. Small studies with large positive results may be more likely to appear in journals than null findings, making a published sample unrepresentative of all interventions attempted.
Evidence from government implementation provides an important comparison. Stefano DellaVigna and Elizabeth Linos analyzed 126 randomized trials involving more than 23 million people from two US nudge units. Their 2022 Econometrica article reported an average increase of about 1.4 percentage points in the targeted outcomes, compared with about 8.7 percentage points in the academic comparison sample they assembled. The discrepancy did not imply that field interventions were worthless; it showed that effects observed in selected academic studies could substantially overstate typical effects in operational programs.
Generalization requires attention to baseline behavior, population, institutional trust, and delivery. A reminder may work differently when messages are rare than when recipients receive many automated notifications. Replication can test whether an effect recurs, while longer follow-up can identify fading, delayed responses, or displacement of other activity. Registration of hypotheses and analysis plans, publication of null results, and transparent outcome reporting improve the evidence base. No single pooled effect adequately describes an intervention class spanning defaults, reminders, social comparisons, and simplified administrative systems.
Welfare, autonomy, and ethical disagreement
The normative argument for nudging often begins with a distinction between a person’s immediate behavior and that person’s considered interests. Someone may intend to save, apply for a benefit, or renew a document but fail to act because of inattention or procrastination. An intervention that helps carry out that intention can appear less intrusive than one that substitutes an official’s objective. Nevertheless, determining a person’s considered preference is difficult when preferences change, information is incomplete, or available options are poorly understood.
Libertarian paternalism proposes evaluating interventions partly by whether they make people better off “as judged by themselves.” This formulation aims to avoid imposing a single external conception of a good life. It does not eliminate problems of welfare economics, however. Enrollment rates, purchases, and stated satisfaction are imperfect measures of welfare. A participant may appreciate a default at one time and regret it later, while an observed opt-out may reflect either an informed preference or confusion about the program.
Autonomy concerns extend beyond the formal availability of alternatives. A person may retain a legal right to reject a default yet find the interface confusing or the consequences obscure. Critics distinguish interventions that help deliberation from those that exploit a response without inviting reflection. Supporters reply that all usable choice environments direct attention and structure decisions, so the relevant comparison is between alternative designs rather than between influence and its complete absence. Both positions place importance on the actual design, not merely the label “nudge.”
Transparency has several dimensions. People can be told who is responsible for an intervention, what outcome it seeks, why an option is preselected, and how to choose differently. Explaining every psychological mechanism is not the same as disclosing a policy’s practical operation. Research has found that some disclosed nudges remain effective, so transparency and effectiveness are not necessarily incompatible. However, whether a disclosure is noticed and understood is itself an empirical question.
Distribution also matters. An intervention can raise an average outcome while imposing burdens on particular groups. Digital reminders exclude people without reliable access; complex opt-out procedures may disproportionately affect those with less time, lower literacy, or disabilities. Public-interest claims therefore require attention to accessibility, legitimate authority, and competing interests. Where personal information is used to tailor interventions, data privacy and data governance become part of the ethical assessment rather than separate technical concerns.
Digital choice environments, sludge, and institutional limits
Digital systems allow choice environments to be altered quickly and tested at large scale. A website can change the prominence of buttons, the sequence of questions, or the timing of a notification for different users. Personalization may use prior behavior or other data to select an intervention. This creates possibilities for reducing irrelevant messages, but also makes the intervention harder to observe: different people may encounter different defaults, explanations, and exit procedures without knowing that alternatives exist.
The term sludge describes excessive or unjustified friction that obstructs action, such as repeatedly requesting information already supplied or making cancellation much harder than enrollment. It is related to nudging because both direct attention to the behavioral consequences of process design. They are not exact opposites, however. Friction can be legitimate when it verifies identity, protects against fraud, or gives time to reconsider an irreversible decision. The question is whether its burden is proportionate to its function and whose interests it serves.
Commercial “dark patterns” include misleading visual hierarchies, hidden costs, and interfaces that steer users toward choices they might reject under clearer conditions. Some operate through familiar nudge mechanisms, such as defaults or salience, but deception and difficult exit can place them outside the original requirement that avoiding a nudge be easy and inexpensive. Analysis must distinguish a prominent recommendation from a concealed charge, and a reversible default from an obstruction that effectively removes meaningful choice.
Institutional incentives can be examined through the principal–agent problem. A designer may be rewarded for conversion, enrollment, or reduced administrative workload rather than for users’ long-term welfare. Optimizing the measured target can then produce a superficially successful intervention that transfers costs elsewhere. An agency might reduce unanswered correspondence by discouraging contact, for example, without improving resolution of the underlying problem. Outcome selection is therefore part of governance, not merely a statistical decision.
Nudges also have limits as instruments of public policy. They cannot, by themselves, create housing supply, provide income, expand service capacity, or correct every form of market power. Their appropriate comparison may be a conventional policy, a combined program, or no intervention. Cost–benefit analysis must include development and administration costs, users’ time, mistakes, privacy implications, and downstream consequences—not only the low marginal cost of sending another message. Durable institutional evaluation consequently examines whether the targeted action remains beneficial, whether effects persist, and whether changes elsewhere in the system have altered the case for the intervention.