Libertarian Paternalism: Making Rational Fools Open Access

The initial round of behavioral economics literature – the so-called “heuristics and biases” program sparked by Kahneman and Tversky’s (1979) paper on prospect theory – generated a significant amount of methodological debate. The main controversy was whether the heuristics and biases program was a better way to predict and explain individual choice behavior than rational choice theory: the homo economicus-based approach that has dominated economic theory since the end of the nineteenth century. While the heuristics and biases program did not provide a unified alternative to rational choice theory, it did produce a vast number of empirical anomalies – including status quo bias, preference reversals, hyperbolic discounting, constructed preferences, framing effects and many others – which demonstrated that actual behavior often deviated from rational choice theory in significant and systematic ways. In Richard Thaler’s words: “The approach taken by most behavioral economists has been to focus on a few important ways in which humans diverge from homo economicus” (Thaler, 2017, p. 1800).

Despite the challenges that such anomalies posed for standard economics there has not been a paradigm shift in the way economists typically model individual economic behavior. Rather than a revolution, the methodological debate has mostly subsided and there now seems to be a relatively stable equilibrium, a peaceful coexistence, where most economists view behavioral economics as a complement to, rather than a substitute for, traditional rational choice theory (Angner, 2019). However, this stable equilibrium only concerns the positive science of choice theory. In recent years another, and in many ways more controversial, debate has opened up regarding the implications of behavioral economics for normative economics: welfare economics and microeconomic policy. The relationship between behavioral and welfare economics is generally called behavioral welfare economics, and the problem of trying to reconcile the various tensions between the two subjects is called the reconciliation problem (McQuillin and Sugden, 2012). There are many different aspects to the behavioral welfare economics debate, but the topic that has dominated the conversation more than any other on the policy side is behavioral paternalism, the focus of Mario Rizzo and Glen Whitman’s Escaping Paternalism: Rationality, Behavioral Economics, and Public Policy (Rizzo and Whitman, 2020, hereafter referred to as RW).

As Rizzo and Whitman explain, the “burgeoning field of behavioral economics has produced a new set of justifications for paternalism” and their book “challenges behavioral paternalism on multiple levels from the abstract to the conceptual to the pragmatic and applied” (RW, i).¹ Their remark that they “have presented a gauntlet of challenges to behavioral paternalism” (RW, 398) is somewhat of an understatement since the book is a relentless critical engagement with behavioral paternalism that attacks it on a wide range of different fronts. Although it is a multi-pronged attack, they do a very good job being systematic about each of the criticisms they discuss. For example, in chapter four they present various concerns about the empirical research on heuristics and biases, research that is the foundation for behavioral paternalism. And yet when they move on to other criticisms in later chapters, they set aside the material in chapter four and argue that even if the empirical research was unproblematic, various other issues would still be a problem. They proceed this way on a topic-by-topic basis. The result is “a series of ‘even if’ arguments” (RW, 399) that forms a systematic multi-pronged attack coalescing into a broad-based critique of behavioral paternalism.

Although the way Rizzo and Whitman’s critical account comes together is impressive, I will draw on only certain aspects of their argument: primarily the first three chapters and chapters six and seven. There are a number of places where slightly different terminology will be used, but for the most part the paper will stick with Rizzo and Whitman’s main story line from these chapters – although toward the end of the paper, my argument will turn in a different direction. Their main point is that behavioral paternalist policies are misguided and problematic, and the majority of their arguments are very persuasive, particularly those about the normative role of homo economicus and the knowledge problem of policy makers. But as they move through the book, they increasingly emphasize ideas from public choice theory – rational ignorance, government failure, slippery slopes, and such – and it seems that in the process they start to lose sight of the target – behavioral paternalism – and begin to slide into a more broad-ranging criticism of governmental action. In section four the argument will be made that while the vast majority of their critical remarks about behavioral paternalism are correct, such criticisms should lead, not in the direction of less governmental action in toto, but rather away from behavioral paternalism and toward increased governmental attention to the things that economists have traditionally considered to be the responsibility of democratic governments: that is social policy. In other words, to spend more time and resources on serious social problems rather than trying to nudge people into being more effective rational fools.²

Paternalism within economics traditionally involved the use of taxes, subsidies, and regulations to change behavior in ways that would make the relevant individual better off than they would be on their own. For example, smoking cigarettes causes cancer and yet individuals choose to purchase and smoke cigarettes. Given they would be healthier, that is, better off, if they smoked less, a tax or restriction on the sale or use of cigarettes would, ceteris paribus, cause them to smoke less and therefore make them better off. Since economists have traditionally assumed that people make choices on the basis of utility/preference maximization, smokers and others who prefer to consume harmful products were not typically viewed as irrational (non-utility maximizing) by economists, but rather, simply as individuals making rational choices given unhealthy preferences. Thus, paternalist economic policies that raise the price, or restrict the sale or use, of products like cigarettes simply incentivize people to consume less of these products even though they are products the individuals prefer. This works in the same paternalistic way as incentivizing a child to consume less candy would make them better off, even though they would prefer to eat more, not less, candy. In the case of adults and products like cigarettes there has been some debate about whether the smoker’s behavior is a result of tensions between their short run and long run preferences, or whether it was a simply a matter of maximizing preferences that are not good for them, but in either case the right (paternalistic) thing to do has traditionally been to use standard microeconomic tools to change people’s behavior in ways that was good for them in spite of their preferences.

One major impact of behavioral economics was to break the traditional direct linkage between preference and choice. Kahneman and Tversky, following an established psychological tradition in behavioral decision theory, presumed that individuals possess stable and well-ordered preferences, but the overwhelming evidence from their own, and other, psychological research indicated that real people often make cognitive mistakes and fail to act optimally on their preferences. Identifying and helping individuals correct such mistakes was a main goal of this tradition in psychological research.³ The various factors – heuristics and biases – responsible for such mistakes is reflected in the laundry list of anomalies that is now associated with behavioral economics. From this point of view, all the various anomaly-creating cognitive mistakes constitute errors in rational decision-making. As Rizzo and Whitman explain:

There are many ways that an individual can make mistakes and act in a nonpreference maximizing way, but the general class of errors that has received the most attention are those related to the context of choice: the choice environment or choice architecture. Economists have traditionally viewed the choice space over which individuals maximize utility/preference to be the space of outcomes – bundles of various commodities in riskless choice and gambles in choice under risk – and thus in most economic models, the context of choice, say the way the vegetables are arranged in the grocer’s display, will have no impact on the choices made. However, as many of the now well-known choice anomalies demonstrate, real people do often allow the choice context to affect their choices, and so alteration in the choice context or choice architecture has become the primary focus in efforts to help people correct their decision-making mistakes.

But choice context is a very general concept that applies to many variations that are quite different from the way things are arranged on a store shelf. For example, Kahneman and Tversky’s influential paper on prospect theory focused on loss aversion – where an individual weighs losses more heavily than gains – and loss aversion is a type of endowment effect: the decision-maker values a particular outcome differently depending on their ownership or endowment at the time of choice. Suppose the individual starts at a particular bundle A with the associated utility level U(A). If bundle B is preferred to bundle A, then movement from A → B will increase the individual’s utility, and the movement back from B → A will reduce their utility. But if the individual is subject to loss aversion, then the perceived gain from A → B will be less than the perceived loss from B → A. This means that the utility associated with bundle A will be lower after the path through B than it was initially. But given the definition of a function, this means that the individual’s valuations of various bundles of goods cannot be represented by a stable utility function.⁴ In this way the heuristics and biases program has led to (a) disconnecting preference from choice, and (b) opening the door to changing the choice architecture as a way of nudging individuals into making more rational, that is utility-maximizing, choices. Notice that through this lens, decision-making mistakes do not come from having irrational preferences – not transitive, not complete, etc. – but rather in failing to act optimally on one’s preferences. Each individual has an inner rational agent, it is just their outer psychological shell prevents them from acting as homo economicus, or in short as an Econ, would act given such preferences. In other words: “behavioral welfare economics models human beings as faulty Econs” (Infante et al., 2016b, p. 22)

We can now see how behavioral economics opens the door for a new behavioral paternalism that is quite different from traditional paternalism. In traditional paternalism choices are presumed to be consistent with the individual’s preferences, it is just that people often prefer things that are not very good for them. Now we have the possibility that because of various cognitive mistakes, the choices that an individual makes need not be consistent with their underlying preferences. If one assumes that individuals are generally self-interested, a common presumption in economics, then such mistakes prevent people from acting rationally in their own self-interest and are thus worse off than they would be if they behaved like Econs. Thus changes in the choice architecture that would nudge individuals away from cognitive mistakes and back into proper utility-maximizing behavior would make them better off. The cognitive mistakes that real people make can be seen as internalities – the internal cost to the agent from less than fullyrational behavior⁵ – and behavioral paternalism is designed to eliminate such internalities by nudging individuals into more rational choices and thus making them better off. Such a policy of course changes homo economicus from being simply a modelling strategy used to predict and explain individual behavior, to being a normative standard for behavioral paternalist policy: something that economic agents ought to do in order to be rational. In Rizzo and Whitman’s words:

At this point it is useful to note that the preferences serving as the normative standard for behavioral paternalism cannot be just any preferences, or even just well-behaved (complete and transitive) preferences. They must meet much stricter conditions. First, they must meet all the typical conditions that economists require on preferences; for example, in the case of riskless consumer choice, preferences need to be complete, transitive, monotonic, and convex.⁶ But they also need to be stable – remain the same for the time of the analysis, or in the case of behavioral paternalist policy, for the duration of the policy – and be context-independent. They are “the preferences that would determine choice in a sanitized or bias-free environment” (RW, 239). Finally, preferences cannot be constructed, revisable, or dynamic; in particular, they cannot be: constructed preferences (Lichtenstein and Slovic, 2006), revised by dynamic learning (e.g., Dold and Schubert, 2018), or emerge from processbased (Rizzo and Whitman, 2018) or inclusive (Rizzo and Whitman, 2020) rationality. The most common term for such preferences is true preferences although other terms are used, including: purified, latent, laundered, pruned, and spruced up. The topic has, as one might expect, generated a quite a debate.⁷

All of this brings us to the libertarian paternalism (LP) that is the topic of this paper. Although the literature on LP is extensive, the main focus will be on Sunstein and Thaler (2003) and Thaler and Sunstein (2003, 2009).⁸ I will focus specifically on LP rather than Rizzo and Whitman’s behavioral paternalism, even though LP is a particular type of behavioral paternalism. There are a number of reasons for this. One is that LP is by far the most influential brand of behavioral paternalism. Its influence can be seen in the number of citations, by the numerous examples that are discussed in the popular literature, and by Richard Thaler’s Nobel Prize in 2017. But secondly, and more importantly, it is useful to focus on LP since there has been a rapid expansion of behavioral paternalist policies in recent years and many of the defining features of Thaler and Sunstein’s original LP have been diluted; this even seems to be the case in later writings of Thaler and Sunstein, for example Sunstein (2016, 2018). Finally, I will focus on LP because even though Rizzo and Whitman use the broader term behavioral paternalism, the most obvious target for most of the chapters in Escaping Paternalism is LP.

It is best to start with Thaler and Sunstein’s own account of LP. It is an example of the correcting-mistakes-through-changes-in-the-choice-architecture theme, but with its own unique style.

The program is paternalist because it makes people better off than they would be on their own. Of course the presupposition is that being better off comes automatically when individuals behave like Econ. It is libertarian not only because it sill allows the individuals to make choices, but because the choice architecture is only re-arranged and the options stay the same. Thus the impact of the policy is always reversible.

LP can be applied to corporations and a wide range of other types of institutions, even families, and does not necessarily involve government policy. That said, I will focus exclusively on government policy since that is where the issues are most clear and where most of the critical attention has been directed.

LP has been explained and its background and motivation have been discussed. The next section moves on to critical concerns, but before moving on there is one additional point to emphasize about LP. The point is that despite the fact that many of Thaler and Sunstein’s examples involve things like increasing organ donations and engaging various environmental policies, which seem to be inherently social, LP is in fact deeply pro-self, not pro-social, nudging. When the goal of the nudge is to make someone better off by better satisfying their own preferences, it is about individual rationality and not about social policies in the way economists have traditionally defined them. The concept of paternalism is about internalities, not externalities.

For example, suppose an individual has a preference for consuming goods that lead to a reduction in atmospheric CO₂, but because of various cognitive mistakes, ends up failing, on their own, to purchase goods that accomplish that goal. Now suppose an LP policy, designed to make this person better off by eliminating her/his cognitive mistakes, is successfully implemented. This would mean that the individual will be a more successful utility maximizer, and thus under standard assumptions, be better off, but it also means a lower level of CO₂ emissions, which is a positive social consequence. But this social benefit does not make the LP nudge into a social nudge. It is simply a pro-self LP nudge that coincidentally had a positive social impact. Social policies are aimed at, that is about, social things – social costs and social benefits⁹ – and this policy was designed to help a particular individual do a better job maximizing her/his own utility. In other words, it was a nudge to correct for internalities, not externalities.

Given the confusion on this self versus social issue, borrowing some terminology from Barton and Grüne-Yanoff (2015) seems to be useful. In their terminology, the general notion of a nudge is: “an intervention on the choice architecture that is… behaviour-steering, but preserves the choice set and… does not significantly change the economic incentives” (Barton and Grüne-Yanoff, 2015, p. 343). But nudges come in at least two different forms, pro-self and pro-social. Pro-self nudges aim at steering people’s behavior in private welfareincreasing ways, and pro-social nudges which aim at steering people’s behavior in ways that internalize externalities or increase public goods (ibid., 344). Finally, LP is a special case of a pro-self nudge: “we suggest to characterize Sunstein and Thaler’s (2003) libertarian paternalism as the advocacy of governmental use of pro-self nudges” (ibid., 344). Of course, a single nudge-based policy could have both pro-self and pro-social impacts, but the LP part would be exclusively pro-self.

I will use this terminology for the remainder of the paper. Rizzo and Whitman are close to this terminology, but they seem to be less explicit about the distinction between self- and social-nudges. For example, they note that there are social nudges, but that such policies “are not our primary target in this book” (RW, 22). This may be because some of their criticisms – in particular those I will emphasize – are most relevant to pro-self nudges, but some of their other criticisms can also be applied to pro-social nudges.

There has been an extensive discussion of the knowledge problems, associated with LP. A sample of the critical literature that emphasizes these knowledge issues, in addition to work by Rizzo and Whitman, includes: Barton and Grüne-Yanoff (2015), Berg (2018), Berg and Gigerenzer (2010), Congiu and Moscati (2020), Davis (2011), Gigerenzer (2015, 2018), Grüne-Yanoff (2012, 2016), Grüne-Yanoff and Hertwig (2016), Guala and Mittone (2015), Hands (2020), Hausman (2016), Infante et al. (2016b), Infante et al. (2016a), Rebonato (2012), Sugden (2015, 2017, 2018), and many others.

In general the knowledge problem is that the policy maker – the choicearchitect-in-chief (Rebonato, 2012) – cannot possibly know what they would need to know to do LP policy; they would need to know the individual’s true preferences. Of course, standard economic theory is applied to microeconomic policy questions all the time on the basis of welfare-gains-and-losses. There are many different versions: the compensation principle, consumer surplus, special assumptions on preferences which allow economists to use a single representative agent, etc. – but all, at least in standard economics, are based on an individual preference satisfaction definition of welfare. So if standard economics can – to an acceptable degree – measure gains and losses in individual preference satisfaction, then why should it be any more difficult for behavioral choice architects to measure true preferences for LP-based policy? There are many reasons, but most important is that all of the standard techniques use the prices that people are willing to pay, or have paid, for the good to measure the increase or decrease in individual preference satisfaction. Consumer surplus, the old standby from the late nineteenth century, for example, is measured by a certain area under the consumer’s demand curve. Such a technique is considered acceptable because economists assume that the consumer’s demand curve comes about as the result of utility-maximizing behavior, which implies that price, what the consumer is willing and able to pay for the good, directly reflects the strength of the consumer’s preference for the good. But this is not appropriate in the context of a LP policy. LP takes behavioral economics as its foundation, and the cornerstone of behavioral economics is that what people choose is frequently not what they really prefer, and this breaks the link between willingness to pay (i.e., price) and individual valuation (preference satisfaction).¹⁰

However, there is another, perhaps more serious, set of problems. Not only can economists no longer assume that demand-prices reflect preferences, the preferences that are needed by the choice architect are true preferences: the preferences that would be reflected in the individual’s choices if they made no mistakes. The choice architect needs not only access to some rough-and-tumble preferences, but preferences that are complete, transitive, insensitive to context effects, not constructed at the point of choice, stable for the duration of the policy period, and so on. It is just not possible to obtain the kind of detailed access to individual true preferences required for LP policy and it is no longer possible to use price as a proxy for the needed information. As Rizzo and Whitman make clear:

But one would need more than true preferences to formulate an effective LP policy. One would also need detailed information about the particular causal mechanisms that are responsible for the cognitive mistakes that lead to supoptimal decisions (Grüne-Yanoff, 2016). So, at least two necessary conditions are needed to design a LP policy: (a) detailed information about the true preferences of the relevant individuals, and (b) detailed information about the causal mechanisms at work in the relevant individual’s outer psychological shell that are responsible for the cognitive errors. And the latter is just as important as the former and almost as difficult to obtain. Again Rizzo and Whitman:

But, in addition to knowledge problems about both true preferences and the details of error-causing causal mechanisms, there is still another concern. As Rizzo and Whitman and many others have pointed out, just because the information that LP policy would require is not available, does not mean that choice architects are not implementing various LP-based policies. Since they do not actually have the detailed information they would need about either true preferences or the particular causal mechanisms driving mistakes, it seems that they must be using what they believe is reasonable to assume about both of these factors in any particular case. This educated-best-guess approach is reflected in the most of the applied LP literature, which almost never starts with details about an individual’s preferences or her/his outer psychological shell, and uses this information to characterize an associated problem and propose an appropriate policy. Rather, the process typically involves more presumption and less detailed information. It begins with an ‘obvious’ concern, like eating fatty food, not saving for retirement, smoking cigarettes, making poor portfolio choices, etc. as a starting point – one that implicitly, but necessarily, makes a number of presumptions about both true preferences and the sources of particular mistakes – and then goes on to directly propose a LP strategy that might nudge the individual in the direction of making fewer of these ‘obvious’ mistakes. As Daniel Hausman explains:¹¹

Even if we grant the principle of charity and assume that no opportunism or conscious manipulation is going on – an assumption that Rizzo and Whitman seem doubtful about – such policies driven by the well-intentioned hunches of choice architects, seem to be along way from the science and evidence-based rhetoric of LP advocates.

Although these three knowledge problems constitute a powerful critique of LP, there are many other knowledge-based criticisms scattered throughout the LP literature. There is certainly no reason to try to note them all, but I will close this section by mentioning two other concerns. I am not noting them here because they play a prominent role in the literature, but rather just the opposite. I think both are important and yet they have not attracted as much attention as they should.

The first is the problem of second best. It is noted by Rizzo and Whitman (RW, 260–261), and a few others, Berg (2018) and Besharov (2004) for example, but it is not a common criticism of LP. The second best theorem came long before the rise of behavioral economics: in Richard Lipsey and Kelvin Lancaster’s 1956 paper. Lipsey and Lancaster’s second-best result demonstrated that in the context of an optimization problem with multiple constraints “it is not true that a situation in which more, but not all, of the optimum conditions are fulfilled is necessarily, or even likely to be superior to a situation in which fewer are fulfilled” (Lipsey and Lancaster, 1956, p. 12). This means that unless the nudges correct for every heuristic or bias that is preventing optimization, then changing one or some of those factors, could move the individual farther away from the optimal choice. Admittedly, the second best issue can arise in any policy context where there are multiple problematic factors, but it seems that it is particularly troublesome in LP policy because choice architects almost never have detailed information about the exact causal mechanisms responsible for the mistakes, or their relative magnitudes, or the various feedbacks they might have on each other.

The second problem concerns the tradeoffs and interactions between LP policies and social policies. We are certainly well aware that individually optimal behaviors may cause externalities that spill over – positively or negatively – on to other individuals in the society. Given this, what is to prevent a pro-self LP policy nudging individuals in ways that lead to negative social impacts? It is not necessary of course that such impacts occur, or that they are negative, but they are always a possibly and generally ignored. Of course, even if such social effects were recognized, it is nearly impossible to know the direction or magnitude of such complex interactive changes. As Rebonato (2012, p. 234) notes, it may even involve LP microeconomic spillovers into macroeconomics. From a Keynesian perspective, an LP nudge that encourages people to save for tomorrow means less consumption and higher unemployment for today (and if the “paradox of thrift” is in effect, it could mean less savings in the future as well).

Although this paper has emphasized some things a bit more, and de-emphasized others, most of the previous two sections is broadly consistent with what Rizzo and Whitman say in Escaping Paternalism (at least in the chapters I noted in the introduction). It is now time for a quick review and then for a change in direction.

The bottom line of the previous section is that because of fundamental knowledge problems, a successful LP policy is simply not possible if it is designed to nudge people into better maximizing their true preferences.

And many philosophers like Hausman agree:

Of course, if how one defines LP or behavioral paternalism becomes more amorphous – downplaying the commitment to homo economicus, true preferences, and pro-self nudges; allowing conventional microeconomic tools like taxes and subsides to sneak in; or loosening it up in other ways – it may be possible to dodge at least some of these specific knowledge-based criticisms. And that seems to be what has taken place in the writings of some LP supporters in recent years. But of course what you loose is the originality and uniqueness of the LP program that made it such an important contribution to economics, such a much-discussed policy tool, and worthy of a Nobel prize. But knowledge problems are clearly not all of the concerns that have been raised about LP. There are also autonomy problems, inconsistency/incoherence problems, concerns with the rhetorical style of Thaler, Sunstein and other proponents of LP, public choice criticisms (much discussed by Rizzo and Whitman), as well as many others.¹²

So suppose one is persuaded by all of these arguments that LP is deeply problematic, then so what? What is to be done? Rizzo and Whitman do not provide very much in the way of an answer to this question, although they do suggest a few places where we might begin to look for a better approach. They mention Gerd Gigerenzer’s fast-and-frugal heuristics program as an alternative to traditional rational choice theory (Gigerenzer, 2002, 2008). They also note Gigerenzer’s ecological rationality, the normative framework associated with the fast-and-frugal heuristics program (Gigerenzer and Sturm, 2012; Gigerenzer and Todd, 2012) as a more flexible evaluative standard than homo economicus. These are certainly reasonable resources to consider since Gigerenzer and associates have repeatedly argued for the advantages of fast-and-frugal heuristics over the heuristics and biases program both in general and with respect to LP in particular (e.g., Gigerenzer, 2015, 2018) and also because Gigerenzer’s program is often involved in discussions about other non-Econ-based, yet behaviorally sensitive, policy approaches: for example, “boosting” (Grüne-Yanoff and Hertwig, 2016).¹³ It should also be noted that they discuss their own conception of “inclusive rationality” as a substitute for rational choice theory in Escaping Paternalism and in more detail in Rizzo and Whitman (2018). They certainly suggest some interesting and important research to consider, but it is discussed fairly briefly and no detailed theoretical or policy alternative to LP is offered.

Toward the end of the book, instead of a theoretical or policy program, one begins to get the idea that for Rizzo and Whitman behavioral paternalism is just one specific, currently fashionable, example of government ineffectiveness and overreach. This is a suggestion that makes appearances throughout the book, but clearly picks up steam in later chapters (particularly chapters eight and nine). The analysis seems to switch from criticisms unique to behavioral paternalism – such as concerns about the empirical foundations of behavioral economics and the inadequacies of homo economicus as a normative benchmark – to more generic criticisms of almost any type of governmental policy. The ideas that percolate through the discussion include public sector spillovers, the rational ignorance of voters, government failure, the biases of governmental policymakers, the influence of special interests, slippery slope arguments about the growth of government, and so on. In most cases these are arguments that can be applied to behavioral paternalist polices, but many of these arguments are, unlike the discussion of the knowledge problem, not unique to, or even generally associated with, either paternalism or behavioral economics. Many of them seem to be fairly general boilerplate indictments of government action:

Although I would say that much of negative indictment of government is overstated, I also realize discussions about such matters run up against individual values – particularly regarding the trade-offs between welfare, justice, and freedom – and are seldom very productive. So instead of arguing against Rizzo and Whitman’s position in the closing chapters, I want to go back to a point of common ground, but then take the discussion in a different direction.

The common ground is Mill On Liberty (1961 [1859]). Most of the text of the Mill quote in the opening epigraph to this paper also appears on p. 437 of Escaping Paternalism. As Rizzo and Whitman say, the quote is associated with what came to be called the Harm Principle: “the idea that we are justified in coercing people only for the purpose of preventing harm to others” (RW, 437). This principle was colorfully captured by Robert Nozick’s remark that a person can leave their knife wherever they want “but not in your chest” (Nozick, 1974, p. 171). The problem is that from the two sentences Rizzo and Whitman cite from Mill, one would not necessarily get the importance of the “harm” part of the Harm Principle. Their quote only talks about what others, including government or society, cannot compel an individual to do. There is nothing in their selection of Mill’s words about what government can rightfully do and that is to coerce individuals “to prevent harm to others.” But since it is called the Harm Principle, it seems that the legitimate tasks of governments should be just as important as what it, or society in general, cannot do. Based on the version of the Mill quote in the epigraph, my own response would be to quit trying to do paternalist policy that equates paternalism with successful utility maximization and instead have the government spend its resources on the kind of activities that Mill argued they should be doing: preventing harm to others. In modern economics these things – at least in microeconomics – boil down to the traditional subjects for governmental action: positive and negative externalities and public goods.

Perhaps paternalist policies should be abandoned altogether, but if there is broad public support for such policies, make them traditional paternalism: using the best scientific evidence about what harms people the most and encourage them to do less of it. This could be done by traditional taxes or subsidies, direct regulation, social nudges, or some combination of these tools. If paternalism is going to be done at all it should be directed towards making people healthier, live longer lives, and so forth, rather than trying to uncover true preferences and the particular heuristics and biases that are preventing people from making optimal choices on those preferences. This is particularly the case because it would only achieve paternalist goals through the dubious causal chain of utility maximization → people get what they want → what people want is what is really best for them → therefore people get what is really best for them. First, it seems pretty clear that such a strict LP policy is essentially impossible to do, and second, following Mill, it is not something the government should be doing anyway. As the world grows more connected and densely populated, where truly social issues like climate change and pandemics become increasingly important, why should helping people maximize utility be a goal of government? Allowing people to get what they personally want seems to be what markets do best, and governments generally do not do very well. On the other hand, helping control things with massive negative externalities – again, the damage from climate change and pandemics comes readily to mind – is traditionally what governments can do and are supposed to do. My suggestion is thus to follow Mill by: (a) abandoning strict pro-self LP entirely, (b) keeping paternalism at a bare minimum, and when used, do it by targeting improvements in human physiology, not by trying to increase individual preference satisfaction, and (c) reaffirm the commitment to genuinely social concerns which could be approached using traditional microeconomic tools, or pro-social nudges, or both.¹⁴ And remember all of these recommendations are not only based on Mill’s vision but also what is and is not possible, or at least practical.

In closing I would like to briefly discuss two of the many recent proposals for improving behavioral economics-inspired policy and see how they compare to what I have suggested. The first approach begins where so many others begin, with the core knowledge problem behind pro-self LP. The second approach does not criticize LP but does suggest broadening and moving beyond narrowness of LP to a more integrated approach to microeconomic policy.

There are several authors who start with the knowledge problem but end up with somewhat different policy proposals. One such case is Francesco Guala and Luigi Mittone’s “anti-welfarist” approach (Guala and Mittone, 2015). They interpret LP as welfarist rather than paternalist, because LP policies are evaluated on the basis of individual well-being which is equated, as in standard economics, with individual preference satisfaction. As they explain, LP is concerned with agents who “fail to behave according to their ‘true’ preferences” and LP interventions “are aimed at improving subjective well-being by replacing irrational agents with well-behaved ones,” which makes LP-nudging “an ally of the neoclassical (welfarist) approach” (Guala and Mittone, 2015, pp. 387–388). The problem is, as clear from the above discussion, that such a welfarist approach, runs directly into various knowledge problems. Guala and Mittone’s solution is to abandon LP-nudging with its single-minded normative commitment to homo economicus, but continue nudging policies with different goals. On one hand, make the paternalism more traditional (drop the goal of making people better subjective utility maximizers and focus on better objective outcomes), and on the other hand, focus on the externalities (note, externalities, not internalities) caused by systematic cognitive mistakes: “prevention of externalities: nudges are a cheap way of preventing later interventions that would be costly for the community” (ibid., 394). As they explain:

The second set of recommendations I will note comes from George Loewenstein and co-authors, particularly Bhargava and Loewenstein (2015) and Loewenstein and Chater (2017). Loewenstein was an early and influential contributor to the behavioral paternalist literature – including Herrnstein et al. (1993) which introduced the idea of an internality and the asymmetric paternalism version of LP: Camerer et al. (2003). His recent contributions to the policy debate, unlike almost every author discussed so far, do not start from the position that LP has been a failure in any way. The application of LP “has enjoyed significant success” (Loewenstein and Chater, 2017, p. 27), it is just time to move forward and improve. These authors argue that behavioral economics-based policy “should not limit itself to proposing nano-size interventions that may not significantly address the more basic causes of the magnitude of contemporary policy problems” (ibid.), in particular, the focus on pro-self LP¹⁵ policies has “overshadowed alternative ways in which policy can and should be informed by behavioral economics” (ibid.). In order to bring about this broader application behavioral economic ideas it is necessary to move beyond the narrow pro-self focus of LP and integrate traditional economic tools with pro-social nudging. The bottom line on the argument is thus:

Comparing these two sets of policy perspectives to my suggestions, it seems that Guala and Mittone’s strategy is very close to what I have proposed. They are opposed to LP as policy-assisted individual preference satisfaction, and argue for a greater social focus. Perhaps getting enough information to target mistakes that have social externalities may run into some of the same knowledge problems as targeting individual preference satisfaction, but their approach seems to be in the spirit of what I proposed. The proposals from Loewenstein and co-authors also seem to be a movement in the right direction: moving away from pro-self nudging and focusing on social policy informed by a wide range of behavioral tools. I would exclude the congratulatory remarks about LP, because it seems like the policies that have been successful are more examples of what these authors are arguing should be done, rather than applications of pure proself LP nudging, but again, the approach appears to be in the same spirit as my suggestions. The bottom line seems to be that there is (rightfully) less interest in policy designed to promote homo economicus – thus making more rational fools – and more interest in bringing a wider range of tools to bear on social concerns. Homo economicus is a very good tool for certain tasks in positive economic science, but it is a very poor goal for normative economic policy.

This paper has criticized LP theory and policy in general, but in particular, by focusing on knowledge problems: questions about whether a choice architect would, or even could, have sufficient knowledge to implement LP policy. These problems are discussed by Rizzo and Whitman and the paper builds on their discussion of these issues. The paper also draws on a fairly broad base of critical LP literature and also examines the historical origins of many of the relevant ideas. Although there is agreement with Rizzo and Whitman about their primary criticisms of LP, it draws less pessimistic conclusions about the role of ideas from behavioral economics in economic policy, particularly with respect to what economists have traditionally considered the reasons for social policy. The paper is less critical of behavioral economics and the role of government in general, but equally, or perhaps more critical, when it comes to the core idea that public policy should be aimed at nudging people into behaving like homo economicus and thus like Sen’s rational fools.