(Book notes) Why Nudge: Libertarian Paternalism by Cass Sunstein

1–2 sentence(s) Summary: Basics about “behavioral economics” & why “choice architecture” is inevitable. Thus, Sunstein argues, “why not shape choice as to produce better prudential outcomes without engaging in paternalistic coercion”

People are greatly affected by default rules, which establish what happens if they do nothing. Because nothing is what many people will do, default rules can produce a lot of damage.

It is important to emphasize that free markets provide significant protection against such errors.

A great deal remains to be understood about the nature and extent of human error in various contexts. We do know that people’s behavior displays “ecological rationality,” in the sense that we tend to choose well in environments for which our rules of thumb, or heuristics, are well suited. Those environments are the norm, not the exception, and for that reason, our choices generally tend to be right for us. It makes far more sense to say that people display bounded rationality than to accuse them of “irrationality,” and for many purposes, bounded rationality is just fine, producing outcomes that are equal to or perhaps even better than what would emerge from efforts to optimize by assessing all costs and benefits.

We do know that people are much affected by choice architecture, meaning the background against which choices are made. Such architecture is both pervasive and inevitable, and it greatly influences outcomes, whether or not we are even aware of it. In fact choice architecture can be decisive. It effectively makes countless decisions for us, and it influences numerous others, by pressing us in one direction or another.

Should choice architects, including those in the public sphere, be authorized to move people’s decisions in their preferred directions? Would any such effort be unacceptably paternalistic? Who will monitor the choice architects, or create a choice architecture for them?

The various empirical findings allow us to identify a set of behavioral market failures, understood as market failures that complement the standard economic account and that stem from the human propensity to err. Is it unacceptably paternalistic to use such failures to justify regulation? Is it legitimate to use choice architecture to counteract behavioral market failures?

My basic answers are that choice architecture is inevitable and that behavioral market failures do, in fact, justify certain forms of paternalism.

When these failures occur and are significant, there are good (presumptive) reasons for a regulatory response even when no harm to others can be found.

In light of the pervasive risk of government error and the inescapable fact of human diversity, it is usually best to use the mildest and most choice-preserving forms of intervention. These forms include “nudges,” understood as initiatives that maintain freedom of choice while also steering people’s decisions in the right direction (as judged by people themselves).

Nudges include disclosure of information, warnings, and appropriate default rules, which establish what happens if people do nothing at all.

The First (and only) Law of Behaviorally Informed Regulation: In the face of behavioral market failures, nudges are usually the best response, at least when there is no harm to others.

Behavioral economists generally focus on means, not ends. Most of their key findings involve human errors with respect to means. Their goal is to create choice architecture that will make it more likely that people will promote their own ends, as they themselves understand them.

Soft paternalism is weaker and essentially libertarian, in the crucial sense that it preserves freedom of choice. In that respect, soft paternalism might be compatible with the Harm Principle, though we will see some serious complexities on that count. A jail sentence and a fine count as hard paternalism, whereas a disclosure policy, a warning, and a default rule count as soft or libertarian paternalism.

We need to add behavioral market failures to the class of justifications for government regulation, and the prospect of paternalism should not deter a careful inquiry, based on the likely consequences, into the appropriate response. As in the case of standard failures, however, the argument for a government response must be qualified by a recognition that public officials are fallible, that the cure may be worse than the disease, and that all relevant benefits and costs must be taken into account.

The upshot of these conclusions, which is that there is an emphatically moral argument for certain kinds of paternalism. Officials can make people’s lives better if they are alert to behavioral market failures and design choice architectures that give health, wealth, and well-being the benefit of the doubt.

Behavioral economists have shown that people are affected by how a problem is “framed.” Suppose that you are deciding whether to have an operation. You are far more likely to have that operation if you are told that of a hundred people who have the operation, ninety are alive after five years than if you are told that after five years, ten are dead. The purely semantic re-framing has a major effect on people’s judgments. In response to questions, people persistently show both framing effects and loss aversion. (There is a nice lesson here for policymakers. If you want to have an impact, choose effective frames and enlist loss aversion. Is it paternalistic for policymakers to heed that lesson? Before you answer “yes,” note that some kind of framing is inevitable.)

Heuristics tend to have “ecological rationality,” in the sense that they make sense in the settings where they are usually applied. When people use heuristics, they avoid answering a hard question and answer a simpler one instead.

While System 2 considers the long term, System 1 is myopic, and in multiple ways, people show present bias. People may, for example, delay enrolling in a retirement plan, starting to exercise, ceasing to smoke, or using some valuable, cost-saving technology. Inertia is often exceedingly powerful, and it helps account for the power of default rules, which establish what happens if people do nothing. Often nothing is exactly what people will do, so the default rule tends to stick, whether it involves saving for retirement, personal privacy, or clean energy.

Behavioral economists have been quite interested in the gorilla experiment, because it shows that people are able to pay attention to only a limited number of things, and that when some of those things are not salient, we ignore them, sometimes to our detriment. Magicians and used-car dealers try to hide gorillas. Sometimes the same is true of those who provide credit cards, cell phone service, and mortgages, with terms that may well matter but that people do not notice.

A more general point is that many nontrivial costs (or benefits) are less salient than purchase prices. They are “shrouded attributes” to which some consumers do not pay much attention. Such “add-on” costs may matter a great deal but receive little or no consideration because they are not salient.

System 1 does not handle probability well. One problem is the availability heuristic. When people use that heuristic, they make judgments about probability by asking whether a recent event comes readily to mind. If an event is cognitively “available,” people might well overestimate the risk. If an event is not cognitively available, they might underestimate the risk.

Instead of asking hard questions about statistics, System 1 asks easy questions about what comes to mind. “Availability bias” can lead to significant mistakes about the probability of bad outcomes, taking the form of either excessive fear or unjustified complacency.

In fact it might be most sensible to understand paternalistic interventions in terms of a continuum from hardest to softest, with the points marked in accordance with the magnitude of the costs (of whatever kind) imposed on choosers by choice architects. On this view, there is no sharp distinction between hard and soft paternalism; all we have are points along a continuum. But differences of degree are important.

Under this approach, a statement that paternalism is “hard” would mean that choice architects are imposing large costs on choosers, whereas a statement that paternalism is “soft” would mean that the costs are small. All costs, material or non-material, would count, and to assess the degree of hardness, we would inquire into their magnitude.

Social meanings are a central part of society’s choice architecture, and they are often taken for granted, even though they could be otherwise.

People are often affected by the social meaning of their action. If the meaning of risk taking is bravery and leadership, people might well take risks. If the meaning of risk taking is cowardice, foolishness, and timid conformity, people will be less likely to take risks.

… importance of distinguishing between means paternalism and ends paternalism. In acting paternalistically, government might well accept people’s ends but conclude that their choices will not promote those ends. A GPS provides information about how to get from one place to another. People can ignore what the GPS says and try their own route, but if they do, they run a risk of undermining their own ends (and people know that).

The GPS is an iconic nudge. It should be seen as a form of means paternalism, and means paternalists want to build on the GPS example. They seek to give people a sense of the best route, given their own ends.

I am interested in defending paternalists who respect choosers’ own views about their ends, and who seek to increase the likelihood that their decisions will promote those ends.

The distinction between means and ends is a troubled one. Paternalists might see certain activities as mere means, but they might actually be ends. People might enjoy a fattening meal or a spending spree for its own sake, not because it is a means to something else. System 1 helps to establish a lot of our ends, and choice architects might miss this fact, wrongly judging choosers as selecting the wrong means.

The Epistemic Argument insists that so long as they are not harming others, people should be allowed to act on the basis of their own judgments, because those judgments are the best guide to what will make their lives go well. The central objection, applicable to any kind of paternalism (soft or hard, means or ends), is that errors are more likely to come from officials than from individuals. Public officials lack the information that individuals have.

It is a major problem if paternalistic approaches end up freezing the process of competition. Especially if such approaches are hard rather than soft, they may impair the operation of a competitive process that produces a mixture of diverse products that are well-suited to diverse tastes and circumstances.

Perhaps there is no reasonable concern about efforts to ensure that people’s choices are well informed — at least if those efforts do not discourage people from learning on their own. But if people are defaulted into a particular savings or health care plan, rather than being asked to choose such a plan for themselves, learning is less likely to occur.

Hard paternalism is worse, but soft paternalism can also infantilize citizens by reducing such learning (and liberty in the process).

For those who emphasize the value of learning, it might seem best for choice architects to call for active choosing rather than default rules. In many areas, government might dispense with default rules and instead require people to make choices on their own. For savings plans, health insurance, organ donation, and much more, choice architects might insist that people have to say what they want. Perhaps this approach can itself be counted as a form of soft paternalism insofar as it steers people toward active choosing, when people have not made an active choice in favor of active choosing. Those who prefer active choosing are not avoiding nudges; they are nudging. They are not avoiding libertarian paternalism; they are engaging in it.

Mill noted that conformity to custom “does not educate or develop . . . any of the qualities which are the distinctive endowment of a human being. The human faculties of perception, judgment, discriminative feeling, mental activity, are exercised only in making a choice. . . . The mental and moral, like the muscular powers, are improved only by being used.”

For paternalists, the problem is that one size is unlikely to fit all. With respect to diet, savings, exercise, romance, credit cards, mortgages, cell phones, health care, computers, and much more, different people have divergent tastes and situations, and they balance the relevant values in different ways. In many contexts, an effort to impose a single size will reduce people’s welfare on balance.

The judgments of officials about welfare may be influenced by the interests of powerful private groups.

The second problem is that even if they are well motivated, officials are human too, and there is no reason to think that they are immune from the kinds of biases that affect ordinary people.

System 1 is hardly irrelevant in the public domain. There is no question that availability bias can play a role in public arenas. Recent unfortunate events might lead people to think that a problem is more serious than it actually is, and the absence of such events might lead people to neglect real problems because no misfortunate comes to mind. Officials are hardly immune from availability bias: If a bad outcome has occurred in the recent past, it is highly salient, and it may affect ultimate decisions. Indeed, officials are subject to a kind of anticipatory availability bias: their anticipation of a terrible outcome, and of being blamed for such an outcome, can affect their judgments.

For every bias identified for individuals, there is an accompanying bias in the public sphere. This point offers serious cautionary notes about paternalism.

Anyone saw a person attempting to cross a bridge which had been ascertained to be unsafe, and there were no time to warn him of his danger, they might seize him and turn him back without any real infringement of his liberty; for liberty consists in doing what one desires, and he does not desire to fall into the river. Let us call this the Bridge Exception.

The point here is emphatically not that System 2 should be in charge and that what appeals to System 1 does not much matter. There is no claim that life must be dry, chocolate-free, and long. The point is instead that people make mistakes about what they would enjoy — or to return to our terms, that System 1 makes mistakes about what it will find appealing.

For paternalists, the serious cases are those in which we make choices that greatly endanger our health and shorten our lives.

True, people learn, and true, pleasure is hardly the only thing that people do or should care about. We choose certain activities not because they are fun or joyful, but because they are right to choose, perhaps because they are meaningful. People want their lives to have purpose; they do not want their lives to be simply happy. People sensibly, even virtuously, choose things that they will not in any simple sense “like.” For example, they may want to help others even when it is not a lot of fun to do that. They may want to do what they are morally obliged to do, even if they do not enjoy it. An important survey suggests that people’s projected choices are generally based on what they believe would promote their subjective well-being — but that sometimes people are willing to make choices that would sacrifice their happiness in favor of an assortment of other goals,

But I am speaking of cases in which people are really focusing on what they will like, and what they experience as a result of their choices is not what they hope and expect. Contrary to the Epistemic Argument, their choices do not make them happy. Those choices might even make them seriously ill (or dead). In this sense, they select the wrong means to their own ends. The unlovely technical term here is “affective forecasting errors.”

But we have seen that in other cases, competitors do best if they exploit such failures. The broadest point is that good choice architecture, even if it is behaviorally informed, should promote rather than squelch competition, and it should keep Hayekian objections very much in mind. If choice architects provide information or warnings about risks, or use default rules, they need not reduce the number of competitors or promote cartels. On the contrary, a respect for competition is central to behaviorally informed regulation, and it helps to undergird its First Law.

Antipaternalists are correct to emphasize the importance of learning.

Paternalists should certainly respect heterogeneity. A central advantage of soft paternalism and nudges is that they do exactly that, because they allow people to go their own way. (Return to the First Law.)

We have seen that with respect to default rules, personalized paternalism may well be best, precisely because one size does not fit all. And choice architects should acknowledge that in some contexts, active choosing is the best response to heterogeneity.

No one should neglect the fallibility of public officials, and hence the real issue is insistently comparative. The public choice problem cannot safely be ignored. But those who invoke that problem should not use it as a kind of all-purpose bludgeon against initiatives that promise to do more good than harm. Interest groups can and do have harmful effects, but that reasonable point is not sufficient to justify a refusal to respond to standard market failures of the sort that support the most conventional roles of government, if we want to promote welfare on balance, we should adopt a rule or at least a presumption against paternalism, whether hard or soft, and whether focused on means or ends.

The rule-consequentialist position would be supported with the following questions: Who will monitor public officials? Who will nudge them? What about the value of private learning, to ensure that (in Mill’s words) people are “active and energetic,” rather than “inert and torpid”?

We have seen that the public choice problem should not be used as an all-purpose battering ram or trump card, and it is possible to identify cases in which people are better off if government is authorized to act paternalistically. But according to one view, the risks outweigh the potential gains. On that view, we should adopt a general rule against paternalism on rule-consequentialist grounds, not because the general rule always leads in good directions, but because it is far safer, and far better, than a case-by-case approach.

We have seen an immediate objection to the rule-consequentialist suggestion, and it cannot be repeated often enough, simply because it is so often ignored (and so please forgive the italics): Choice architecture is inevitable

The social environment influences choices, and it is not possible to dispense with a social environment. This point holds whether the social environment is a product of self-conscious designers or of some kind of invisible-hand mechanism. There can be (and often is) choice architecture without choice architects.36 Default rules are omnipresent, and they matter. Do we have an opt-in design or an opt-out design? Whenever there is an answer, there is an effect on outcomes.

People should not be regarded as children; they should be treated with respect.1 They should be seen as ends, not means. Ludwig von Mises insisted that people “are the sovereign,” and that the proof of this claim is “that they have the right to be foolish”; in his view, “freedom really means the freedom to make mistakes.” If government substitutes its own judgments for those of choosers, it violates these principles. The real problem, on this view, is that all forms of paternalism, including those that grow out of an understanding of behavioral market failures, threaten individual dignity and endanger liberty. Here too, however, we should make a distinction.

Empirical research suggests that people do indeed believe that decision rights have intrinsic value, apart from their instrumental value, and when this is so, the welfare effects of compromising those rights must be taken into account.

It is also important to see that in some contexts, people do not enjoy freedom of choice and would much prefer not to have to spend time on the question at all. Especially in complex and unfamiliar domains, active choosing can be a burden, not a benefit. Whether people feel frustrated by a denial of choice, or instead relieved and grateful, depends on the context. There are also issues about the extent to which active choosing increases or decreases satisfaction with ultimate choices.

The thick version of the autonomy position stresses not that freedom of choice is part of welfare but that it is an end in itself and thus decisive — or at least a very weighty matter, to be overridden only for the most compelling reasons.

Freedom of choice has intrinsic value not only in the sense that most of us independently value the right to choose (the thin version), but also and more fundamentally because human beings have dignity, and are entitled to do as they wish.

Experimental evidence strongly suggests that people do in fact welcome many nudges, including those that do not call for conscious deliberation, at least when they are aware that they face some kind of self-control problem or are otherwise prone to error.

Every hour of every day, choices are implicitly made for us, by both private and public institutions, and we are both better off and more autonomous as a result. If we had to make all decisions that are relevant to us, without the assistance of helpful choice architecture, we would be far less free. In a literal sense, choice architecture enables us to be free.

Time is limited, and some issues are complex, boring, or both. If we did not benefit from an explicit or implicit delegation of choice-making authority to others, we would be far worse off, and in an important sense less autonomous, because we would have far less time to chart our own course. Autonomy itself depends on a social background, many of whose basic ingredients we need to be able to take for granted. Without that background, and if active choosing were required for everything, our autonomy would quickly evaporate.

The thick version of the autonomy argument does not turn on empirical questions, and it is, in a sense, a show-stopper. If people have to be treated as ends rather than as mere means, and if this principle requires government not to influence private choices on paternalistic grounds, there is not a lot of room for further discussion. We might be forced to acknowledge that if we accept a certain view of autonomy, actions that fall in the category of hard paternalism are presumptively out of bounds. In such cases, those who believe in autonomy will insist that government needs an exceptionally strong reason to interfere with private choices. But what about soft paternalism and nudges? Is there anything insulting or demeaning about automatic enrollment in savings and health care plans, accompanied by unconstrained opt-out rights? Which nudges, and which forms of libertarian paternalism, interfere with autonomy, rightly understood? Surely disclosure policies are not an objectionable interference with autonomy; a GPS does not compromise liberty. Does government really treat people as mere means when it uses default rules to promote better choices?

Let me venture a stronger, more direct, and perhaps reckless response to those who invoke autonomy. On one view, what really does and should matter is welfare, for which claims about autonomy are best understood as a heuristic (at least in the areas under discussion here). More precisely, autonomy is what matters to System 1, but on reflection, the real concern, vindicated by System 2, is welfare. People speak in terms of autonomy, but what they are doing is making a rapid, intuitive judgment about welfare. Often they are moved by the Epistemic Argument, but they invoke autonomy instead, even though it is not their real concern. On this view, objections from autonomy are far from pointless, and for one reason: When we vindicate autonomy, we generally promote welfare. But on this view, it is much better, and much less crude, to focus directly on welfare.

It bears repeating that in ordinary life, choice architecture ensures that we do not have to make countless imaginable decisions. Of course we are allowed to participate in markets and to vote, and in these ways, we are able to influence choice architecture of multiple kinds. In many contexts, we can opt out. But our autonomy is promoted, not undermined, by the existence of helpful choice architecture, ensuring that if we do not make particular decisions, we will be just fine. If we had to make far more decisions, our autonomy would be badly compromised, because we would be unable to focus on what concerns us. This is not a point about heuristics. It is a point about the limitations of time, interest, and attention.

So long as freedom of choice is maintained, and government does not impose significant costs on those who seek to go their own way, autonomy is not undermined.

Suppose that we stipulate that governments should not treat people merely as means. Nudges, including information disclosure, do not run afoul of that prescription. On the contrary, disclosure itself promotes autonomy, by allowing individuals to make informed decisions about their own ends.

Default rules. In the absence of a system of active choosing, some rule has to specify what happens if people do nothing. If, for example, it is presumed that people are not enrolled in savings plans or in health insurance programs, it is because a particular default rule has been chosen, not because God or nature has so decreed. Perhaps an opt-in program is better, for various reasons, than an opt-out one; but in either case, people are being defaulted into one set of outcomes rather than another. As we have seen, active choosing is a possible way out of the occasional difficulty of choosing the right default rule, and in some cases, active choosing is best. But as we have also seen, some people would prefer not to choose, and active choosing has difficulties of its own, especially in complex or novel circumstances.

I have argued that the best responses to behavioral market failures usually involve nudges, which preserve freedom of choice. But I have also suggested that the master concept is social welfare, and that when the benefits justify the costs, harder forms of paternalism are not off-limits.

If paternalistic approaches impose small costs, or no material costs, on those who seek to go their own way, then such approaches are far less vulnerable to the objections I have discussed here. The central reason is that they preserve freedom of choice.

Soft paternalism and nudging do run into three potentially special concerns, and they should be addressed independently. The first involves transparency. The second involves the risk of manipulation. The third (applicable to both soft and hard paternalism) involves the legitimate claims of System 1. There are also questions about the relationship between illicit motivations and soft paternalism.

Hard paternalism generally involves measurable instruments. The public can observe the size of sin taxes and voters can tell that certain activities have been outlawed. Rules can be set in advance about how far governments can go in pursuing their policies of hard paternalism. Effective soft paternalism must be situation-specific and creative in the language of its message. This fact makes soft paternalism intrinsically difficult to control and means that it is, at least on these grounds, more subject to abuse than hard paternalism.

The underlying concern must be taken seriously, and the best response is simple. For Glaeser’s reasons, nothing should be hidden, and everything should be transparent. Soft paternalism, nudges, and any other behaviorally informed approaches, no less than hard paternalism, should be visible, scrutinized, and monitored.

Perhaps we can answer by offering a behavioral twist on Glaeser’s argument. Perhaps the problem is not so much a lack of transparency as a lack of salience. Perhaps the objection is not that the government keeps behaviorally informed initiatives secret, but that those initiatives do not attract the kinds of attention that are typically triggered by mandates and bans.

In imposing very low costs, or in failing to impose material costs, on choices, soft paternalism differs from mandates and bans. Because of the absence of such costs, soft paternalism appears to be easily reversible.

Does this mean that so long as soft paternalism or a nudge is involved, no one should worry about paternalism, or indeed about any abuse of authority or power? The answer is no. It is easy to identify an important problem with the idea of easy reversibility: the very biases and decisional inadequacies that I have traced here suggest that even when reversibility is easy in theory, it may prove difficult in practice. In part because of the power of System 1, soft paternalism may turn out to be decisive.

In view of the fact that many people do not opt out even when it is simple to do so, a self-interested or malevolent government could easily use default rules to move people in its preferred directions. If we accept very strong assumptions about the likelihood of government mistake and about the virtues of private choice (uninfluenced by government), we might regard opt-out as an illusory safeguard and for that reason reject some forms of soft paternalism.

When people really dislike a default rule, they might well reject it. For that reason, liberty of choice is a real safeguard.

The freedom to opt out is no panacea. But it is exceedingly important.

Here is a way to understand one of the central claims made here. Often because of System 1, people err. We need to strengthen the hand of System 2 by promoting self-control, reducing unrealistic optimism, unshrouding attributes, counteracting biases, and eliminating an undue focus on the short term. Some forms of paternalism move people in the directions that they would go if they were fully rational. Paternalism, whether hard or soft, creates “as if” rationality. Indeed, that is a central point of good choice architecture.

It would be possible to object that if this approach is understood in a certain way, it ignores the legitimate claims of System 1.More bluntly, it disregards what is most important in human life. People like to fall in love, even when it is pretty risky to do that. Many of our favorite foods are fattening. To be sure, most people care about their health, but unless they are fanatical, their health is not the only thing they care about. Many people like to drink and to smoke. Many of us care more about current consumption than consumption twenty years from now. When people enjoy their lives, it is because of System 1. The future matters, but the present matters too, and people reasonably and legitimately strike their own balance. As Mill wrote, “desires and impulses are as much a part of a perfect human being, as beliefs and restraints; and strong impulses are only perilous when not properly balanced.”

If properly confined, paternalistic measures, designed to promote people’s welfare (as they understand it), embody appealing moral principles.

I have urged that accumulating evidence suggests, more concretely than ever before, that in identifiable cases, people’s mistaken choices can produce serious harm, even when there is no risk to others. The result is a series of behavioral market failures that provide a plausible basis for some kind of official remedy. Behaviorally informed responses need not, and generally should not, attempt to revisit people’s ends. They are focused on correcting mistakes that people make in choosing the means to satisfy their own ends.

I have also emphasized the First (and only) Law of Behaviorally Informed Regulation, which is that the appropriate responses to behavioral market failures generally consist of nudges, typically in the form of disclosure, warnings, and default rules. But social welfare is the master concept, and in some cases, a stronger response may be justified after careful consideration of benefits and costs.

Many of the most plausible arguments against paternalism depend on empirical claims that may not be true. Everything depends on the context. We do best to avoid chest-thumping abstractions and general propositions that are appealing but sometimes false. The central question involves the likely effects of particular approaches, whether paternalistic or not. Better understandings of behavioral market failures, and of choice architecture, are uncovering many opportunities for increasing people’s welfare without compromising the legitimate claims of freedom of choice.