Much Ado About Nudging

shutterstock_39190519

By Richard H. Thaler – University of Chicago[1]

When the editors of this journal asked me to write a commentary on a new paper by George Loewenstein and Nick Chater I did not expect that I would have anything useful to say.  After all, I have known George for longer than we would care to admit, and we mostly share similar worldviews.  Nonetheless, after reading their paper, and after trying and failing to sort things out in person, I found myself left with enough differences of opinion that it seemed worth writing something up.  With one exception I find myself largely in agreement with Loewenstein and Chater, but I am mystified about who they think they are arguing with.

In their provocative paper, “Putting Nudges in Perspective”, George Loewenstein and Nick Chater attempt to correct what they perceive to be a serious problem in the research agenda of applying behavioral economics to public policy.  The problem, according to Loewenstein and Chater (LC), is that, “many people” think that we can solve most of the world’s problems with mere nudges.  This is an odd claim to make since, as far as I know, it is not one that would be made by any of the authors of the two papers that started this line of research, namely Colin Camerer, Sam Issacharoff, George Loewenstein, Ted O’Donoghue, and Matthew Rabin, the members of the “Asymmetric Paternalism” team (2003) and Cass Sunstein and me, the creators of the “Libertarian Paternalism” (2003) faction.  Indeed, in our book Nudge (2008), Sunstein and I go out of our way to explicitly disavow the idea.

Apparently LC believe that the aims of these original authors has been “lost in translation”, though the particular translation they are referring to appears to itself have been lost, since LC do not cite anyone, be it academic, policy maker or journalist, who holds this view.  The closest they come is a quote from a campaign speech made by David Cameron when he was running for Prime Minister, which contains what seems (especially now) to be a very mild bit of campaign hyperbole.  But, after mentioning in a footnote that Cameron himself also advocated using standard economic measures in the same speech, LC go on to say: “We stress, however, that this [the value of combining behavioral and standard economic tools] is a conclusion that is not endorsed by many advocates of applying behavioral insights to public policy.”  Although I agree with this statement I would go further. I know of no behavioral economist, policy maker, or journalist who is on the record saying that nudges are a panacea, nor the appropriate tool to address every policy problem.[2]  Which begs an answer to the following question: why did Loewenstein and Chater write a paper critiquing a view that no one appears to hold?

Part of the problem, it seems, comes with the title Sunstein and I used for our book on this subject, “Nudge”.  We have to admit that we were not clever enough to come up with this title ourselves.  We had rather liked the title “Libertarian Paternalism is not an Oxymoron”, or perhaps “The Gentle Power of Choice Architecture” but when one of the many publishers who politely declined the opportunity to publish our book suggested that the word “nudge” might be a nice title, we had the good sense to adopt it.  With one of our original titles the book would have been lucky to sell a hundred copies.  But a one-word catchy title for a book is not meant to be a complete statement about the span of a policymaker’s toolkit.  Anyone who goes beyond the title and actually reads the book will find that we make it clear that our goal (and the goal of behavioral economics in general) is to enrich the standard economic analyses of policy problems by including insights from behavioral science.  That was, is and I hope shall always be the “nudge agenda”, and as far as I can tell it is identical to the agenda that LC endorse.

Numerous times in the paper LC create straw man behavioral views that they then critique.  For example, they correctly point out that obesity is a relatively recent problem.  For most of human’s time on earth binge eating was probably a pretty good idea, and in fact, carrying a bit of extra fat to act as an emergency energy supply might have been a good insurance policy back when food depended on hunting and gathering.  As LC say, the explanation for the rise in obesity must be mostly structural (and economic).  Food has become cheaper, especially fast food, which tends to be unhealthy; work has become less strenuous (I am burning very few calories writing this article); and portions have grown larger.  These observations are completely uncontroversial.  So what is the point in stating the obvious, namely that it is unlikely that the rise in obesity “resulted from a sudden population-wide increase in present-bias or any of the other behavioral effects it has been attributed to”?  Who has ever suggested such a thing?  The phrase: “Present bias contributes to obesity” is surely true but does not imply that present bias is the single or even most important cause of obesity.  Good behavioral economics includes both structural (and other economic) factors as well as psychological phenomena, just as good choice architecture requires thoughtful design of everything that might influence decision making, for good or bad.  This is what distinguishes the field of behavioral economics from psychology: market forces are front and center.  And if that is the primary goal of Loewenstein and Chater’s paper I simply want to strongly agree to agree.

The Matrix

Along with dispelling a view that I believe no one holds, Loewenstein and Chater’s paper has two other goals.  The first is to offer a new taxonomy of policy problems and solutions, provided in the form of a matrix. The second is to express the opinion that it is often better to shove instead of nudge, that is, strong paternalism is often better than soft paternalism.  This rest of this essay addresses these two issues.

The primary substantive contribution of the paper is Table 1, the matrix of types of interventions and rationales for such interventions. I have already agreed that the research agenda of behavioral public policy should not be limited to the lower right hand corner, but I would go much further and say that the lower right hand corner should not be labeled “pure behavioral economics” since it does not contain any economics! Perhaps we should call it pure behavioral science (cognitive and social psychology plus other non-economics social sciences).  All of behavioral economics (and all of the many papers that LC cite) are hybrids of psychology plus economics.  So, yes the lower right hand corner does not provide the answers to all the world’s problems, in part because it does not exist!

However, the label used for that cell is not my primary problem with the matrix, and in fact I agree with the general spirit in which is is offered.  The issue I have is with putting it to use.  My first critique is that to be practical there would have to be many more entries on each dimension of the table: problems and solutions.  One of the other titles Sunstein and I considered for Nudge was “Everything Matters”.  The phrase was meant to capture what I now call “supposedly irrelevant factors” or SIFs (Thaler, 2015).  Economic theory makes many strong predictions about things that will not influence consumer choices.  Examples include: the framing of a choice, the order in which choices are presented, the designation of default options, sunk costs, temptation, emotional states, etc.  The standard theory says that none of these things matter, but we know that they all do, as well as hundreds of others.  Of course, what a neo-classical economist might consider a SIF, a behavioral economist eyes as a potential tool.  If defaults matter, then why not pick one that will be most helpful (as judged somehow[3]).  But this means that the number of columns in this table would need to be large, though most will fit somewhere in the category of “hybrids” since any tax, subsidy, or regulation has to be framed in some way.

A more basic problem I have with this matrix is that I do not think it is a helpful way to think about designing effective solutions to public policy issues, or any other problem for that matter.  The reason is that almost every policy problem has multiple causes (as the authors well illustrate in their discussion of smoking and other policy domains), and policy makers should consider incorporating every tool that can help.  As the authors say: “… there is no logical connection between whether the rationale for intervention is behavioural or traditional economic (or a combination) and the type of response (again economic, behavioural or a hybrid) that is most efficaciously applied.”  I agree!  But what that this means is that anyone who tries to place a particular policy problem into one of the cells of this matrix will necessarily be excluding some important aspects of the problem being solved, as well as some of the potential techniques, behavioural and otherwise, that could be incorporated in the solution.

Retirement saving is a good illustration of this point.  The behavioral economics academic literature on retirement saving often stresses self-control or present bias as the primary drivers of the problem, but in any practical setting this is too simplistic of a diagnosis because it concentrates on just one cause of under-saving.[4] The reasons for a failure to save can include procrastination, inertia, inattention, financial illiteracy, choice overload, “scarcity overload” (as defined by Mullainathan and Shafir), peer pressure to keep up with the Jones’ conspicuous consumption, and many more.  Furthermore, even if someone realizes they should save more, they might not know how to go about doing it. If there is one term that best defines the majority of unsophisticated retirement savers it is “passive”.  (See Chetty et al., 2014)

Since there are many causes of the problem there is no single good solution.  The package of behavioral modifications that have become the norm in a majority of large US defined contribution retirement plans are threefold: automatic enrollment, automatic escalation (or Save More Tomorrow (2004)), and well-designed default investment vehicles such as low-cost target date funds.  If automatic escalation is the default and the stopping point is high enough (at least 10%), then the results are, in my view, satisfactory, at least for workers who are not frequent job changers, and certainly a big improvement on the state of 401(k) plans a decade ago.[5]  Notice that when creating Save More Tomorrow, Benartzi and I deliberately used a combination of behavioral tools.  We invited people to join sometime in the future to accommodate present bias; once people joined they remained in the plan until they opted out or reached some maximum, to take advantage of inertia; and we tied savings increases to pay increases to mitigate loss aversion (which is experienced in nominal dollars).  And in contrast to the argument that standard economic policy tools are more powerful than nudges, it is useful to remember that these retirement plans come with an economic incentive in the form of a tax shelter.  But in their study of the effect of such plans in Denmark, Chetty et al (2014) find that these tax breaks do not have a big effect.  They conclude their paper as follows: “the findings in this study call into question whether subsidies for retirement accounts and reductions in capital income taxation are the best way to increase savings rates. Our findings strengthen recent arguments for using ‘nudges’ such as automatic payroll deductions instead of such policies.”

Behavioral approaches to more complex problems such as climate change or health care will need to incorporate an even more diverse set of interventions, of course, including standard economic approaches based on prices.  But one trap must be avoided, which is to look at what appear to be small effect sizes of nudge-style interventions and conclude that they can’t contribute to a big problem like climate change.  There is no silver bullet to a problem as complex as climate change, and small changes can add up, especially if they are cost effective. A recent study by Benartzi et al, (forthcoming) finds that small “nudge-style” interventions compare very favorably with more traditional economic solutions on the basis of cost-effectiveness.

Of course nudges should be combined with economic incentives, but even the impact of prices have to be viewed through a behavioral lens.  As Baicker et al (2015) have shown, the usual economists’ prescription to rising health care costs, which is to increase deductibles and co-pays to give patients more “skin in the game”, can easily backfire if patients are not judicious in the way they react to the higher marginal costs they face.  Their article notes that in addition to the usual moral hazard associated with zero marginal cost prices there is a “behavioral hazard” induced by positive prices when patients cut back their spending on medications, such as those for diabetes or heart disease, that demonstrably reduce future health care costs.  In these cases, the proper co-pay should theoretically be negative!

My main point is that policy problems will not fall neatly into any of the cells in Table 1.  To the extent that the table suggests otherwise, it will be counterproductive, and help create the illusion that big problems can be solved with one intervention.  My own approach to thinking about such problems is to conduct what I call a “choice architecture audit”, the goal of which is to find the most critical decisions various actors have to make, as well as the potential levers (behavioral and economic) that policy makers can use to improve outcomes.  Perhaps that is the spirit in which LC intended the matrix to be interpreted.  If so, then please consider my remarks to be just an elaboration, and we can again agree to agree.

When push comes to shove

The only portion of the paper about which I have actual substantive disagreements with the authors is their apparent preference for hard rather than soft paternalism.  Loewenstein, it seems, now has had some second thoughts or misgivings about his contribution to the Camerer et al paper that helped start the investigation of soft paternalism strategies.  He and Chater now suggest the possibility that the behavioral modifications to retirement plans discussed above may have done more harm than good “because more heavy-handed policies that remove individual choice seem to produce superior outcomes to nudge approaches that stop short of ‘forbidding any options’ and that are ‘easy and cheap to avoid’ (features of nudges described by Sunstein & Thaler, 2008).”  LC say they prefer the Australian retirement system in which employer based savings is required rather than merely nudged.  Furthermore, participants are unable to borrow or withdraw money from their retirement savings, unlike in the United States.

Of course, Loewenstein and Chater are entitled to their own political opinions, and it is certainly possible to defend a mandatory retirement savings plan such as the Social Security system in the US But it is sheer fantasy to think that in (say) 2006, in the final years of the George W. Bush administration, it would have been politically feasible to enact anything like the Australian retirement system in the United States. In fact, just a year earlier, President Bush tried—and failed—to push his plan to partially privatize the existing Social Security system, making it more like a defined-contribution plan.  That is why when some of us tried to get the Pension Protection Act, a huge bill that was passed in 2006, to include a few provisions to improve 401(k) plans, we aimed for an approach that would appeal across party lines.

The deal we proposed was that if employers adopted three features into their retirement plans (automatic enrolment, automatic escalation and a company match, each at set minimum levels) then they would be given an exemption from an administratively onerous regulatory review to assure they were in compliance with “non-discrimination rules” that limit the proportion of a company’s benefits that can go to the highest paid workers.[6]  This deal offered something to the Republicans (reduce regulatory burden) and something to the Democrats (nudge employers to make plans more generous).  It seems clear that this provision of the law did help increase savings.  Should we have failed to lobby for it because a system like Australia’s might be better?  To quote the best known alumnus of the London School of Economics, Mick Jagger, “You can’t always get what you want.”

Aside from the concern about what is politically feasible, I think one should think carefully before endorsing policies with mandates if there is an effective alternative available.  For example, the United Kingdom has recently introduced a nationwide employer-based retirement savings plan that uses automatic enrolment rather than a mandate, and the opt-out rate has been less than 10%.[7] Let’s suppose that Australia switched from a mandate to an opt-out system and got 90% of employers to participate: would that be obviously worse than their current plan, as LC seem to think?

To those who put a high value of freedom of choice, this would be an easy decision.  In fact, hard-core libertarians even object to making saving the default (often under the illusion that it is possible to avoid nudging altogether)[8].  For those willing to make trade-offs between liberty and other considerations, one useful question to ask about policy alternatives is how well does the system match decision makers with the choices they would make if they had the time, attention, expertise, and self-control to make a well-informed, wise decision.  Let’s label these the options people PREFER (with the understanding that PREFERENCE is just a theoretical concept). Any policy, whether it is a nudge or a mandate, will sort some people into the options they PREFER and others into the alternative, much like type 1 and type 2 errors in statistics. In a nudge-style system those who end up with the wrong option can be said to have been “mis-nudged”.  If the program is a mandate, then they have been “mis-shoved.”

All policies that alter behaviour have to confront the possibility that these mis-matches will occur, and ideally some kind of cost benefit analysis is done to evaluate whether the program should be modified (or eliminated).   But when an action is mandated, rather than encouraged, the social planner has to be even more confident that the choice is PREFERRED for nearly everyone, since citizens have no opportunity to opt out.  It is usually difficult to have such confidence in a world with heterogeneous tastes and circumstances.  Sunstein and I have been criticized often for ignoring the possibility that bureaucrats can also suffer from behavioural biases, in spite of the fact that we repeatedly acknowledge this obvious fact.  So let me say it here again: Officials can make big and damaging mistakes; they are humans after all, so they may get the cost-benefit analysis wrong, or skip it altogether. For all these reasons there should be what lawyers would call a “rebuttable presumption” in favor of freedom of choice when we are just trying to reduce people’s own errors (as opposed to causing harm to others).  Personally, I would prefer the 90% enrolment rate with the opportunity to opt out over a mandate.  At least some of the people who opt out might have good reasons for not saving right now such as a short life expectancy or a desire to pay down student or credit card debts before starting on retirement saving.

Conclusion

I hope this note has not left the impression that there is a deep disagreement between LC and me. We are almost entirely in agreement. Of course we should use all the tools of social science, including traditional neoclassical economics, to address societal problems.  Indeed, I know of no one who disagrees with this view.  As to whether pure nudges should be replaced by more authoritarian rules is a matter up for debate, even if they are politically feasible.  And each case needs to be handled carefully.  For example, I personally think it is crazy for anyone to engage in serious mountaineering, an activity that is clearly more dangerous than saving too little for retirement.  It also does not seem to be what anyone would call “fun”.  Indeed George Loewenstein (1999), himself an enthusiast, calls it “one unrelenting misery from beginning to end.” (p. 317) The logic of the last part of the Loewenstein and Chater paper is that such an activity should be banned, a position I know Loewenstein would oppose.  Maybe he would join me in preferring a gentle nudge, perhaps photographs at base camp of all the people who have died trying to climb this particular peak.  Similarly, some might find my love of wine to be foolish (as Sunstein does), but I think we should all have learned that prohibitions can be costly and ineffective.

Read the full article from Loewenstein and Chater Putting nudges in perspective for free in the first issue of Behavioural Public Policy here.


[1] Thanks to Cass Sunstein, Danny Kahneman, Sendhil Mullainathon and Linnea Gandhi for helpful comments and George Loewenstein for a constructive conversation. .

[2]  There is this sentence immediately following the passage referred to above: “One appeal of nudges is that they promise quick fixes for problems that often call for more fundamental and far-reaching interventions.” I cannot rule out the possibility that someone actually holds this view, but curiously the only reference given is to an op ed written by George Loewenstein and Peter Ubel (2010).

[3] If I seem to be skipping over a tricky detail here with the phrase “as judged somehow” don’t worry.  I will return to it.

[4] Similarly, when LC say that “many researchers have identified present-bias as a (or even the) cause of obesity” I do not know which researchers they are referring to.  I also do not know of any researcher who has said, as LC suggest, that the obesity epidemic was caused by a “sudden population-wide increase in present bias”. Making such assertions would damage the cause of both behavioural economics and good science.  Surely everyone would agree that obesity is a multi-faceted problem requiring a diverse set of interventions.

[5] A bigger problem is that many employers do not offer a retirement plan and Congress has refused to pass a bill creating a low-cost solution to this problem.  Some states, including Illinois, are introducing their own programs to address the problem.

[6] Since 401(k) plans have caps on contributions ($18,000 in 2016), the main obstacle to complying with the rules was to assure that low paid workers enrolled, which automatic enrollment achieves.

[7] Furthermore, we can expect enrollments to rise well above 90% if those who opt out when first asked are re-invited to join periodically.

[8] As Sunstein and I often note, the choice architect has to choose some specific design, be it opt in, opt out, active choice (in which people are required to make a choice), or something else.  The option chosen will inevitably influence the behavior of some participants.

References

Baiker, K. Mullainathan, S. and Schwartstein, J. (2015) ‘Behavioral Hazard in Health InsuranceQuarterly Journal of Economics 130(4) 1623-1667.

Benzarti, S. Beshears, J. Milkman, K. Sunstein, C.R. Thaler, R.H. Shankar, M. Tucker, W. Congdon, W. J. Galing, S. (Forthcoming) ‘Should Governments Invest More in Nudging?’ Psychological Science.

Camerer, C., Issacharoff, S., Loewenstein, G., O’Donoghue, T., & Rabin, M. (2003) ‘Regulation for Conservatives: Behavioral Economics and the Case for “Asymmetric PaternalismUniversity of Pennsylvania Law Review 151(3) 1211-1254.

Chetty, R., Friedman, J. N., Leth-Petersen, S., Nielsen, T. & Olsen, T. (2014) ‘Active vs. Passive Decisions and Crowd-out in Retirement Savings Accounts: Evidence from DenmarkQuarterly Journal of Economics 129(3) 1141-1219.

Loewenstein, G. (1999) ‘The Challenge of Mountaineering … For Utility TheoryKYKLOS 52(3) 315-344.

Loewenstein, G. and Ulbel, P. (2010) ‘Economics behaving badlyThe New York Times 14 (accessed 21 October, 2016).

Mullainathan, S. and Shafir, E. (2013) Scarcity: Why Having Too Little Means So Much New York: Henry Holt.

Thaler, R. H. & Benartzi, S. (2004) ‘Save more tomorrow™: Using behavioral economics to increase employee savingJournal of political Economy 112(S1) S164-S187.

Thaler, R. H. & Sunstein, C. R. (2003) ‘Libertarian paternalismThe American Economic Review 93(2) 175-179.

Thaler, R. H. & Sunstein, C. R. (2008) Nudge: Improving Decisions About Health, Wealth, and Happiness New Haven, CT: Yale University Press.

Thaler, R.H. (2015) Misbehaving: The Making of Behavioral Economics. New York: W.W. Norton.

 


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s