Misbehaving: The Making of Behavioral Economics

by Richard H. Thaler

  Soon after the coalition agreement between David Cameron and Nick Clegg was sorted out, Rohan was in touch. The new government was serious about using behavioral economics, and behavioral science more generally, to make government more effective and efficient. He wanted to know if I would be willing to help. Of course I said yes. We had written Nudge in the faint hope that a few people with some influence might read it and get some useful policy ideas. Since then, Cass had gone to work for his longtime colleague and friend at the University of Chicago Law School who had become president of the United States, and now the Brits were interested as well.

  By some stroke of luck, genius, and timing, David Halpern was selected to run this as yet unnamed operation. David not only is a first-rate social scientist who taught at Cambridge University, but also served as the chief analyst in Prime Minister Tony Blair’s strategy unit. He also coauthored previous UK reports on how behavioral approaches might be used by government, including one while working for Blair. This meant two things: he possessed vast knowledge and experience about how government works, and had the kind of nonpartisan credentials that would be crucial in establishing the team as a source of impartial information. Halpern is also charming and modest. If you cannot get along with David Halpern, then there is something wrong with you.

  During this visit, the team made a quick trip to Paris, where a psychologist, Olivier Oullier, was trying to encourage the Sarkozy government to get interested in behavioral science. On the train ride over, Steve Hilton and I got into a heated debate about what the new team should be called. Steve wanted to use the term “behavior change,” which I thought had awful connotations. David Halpern and I were lobbying for Behavioural Insights Team, the name finally chosen. The argument consumed most of the trip to Paris. At some point Rohan took Steve aside and told him to give in, arguing, prophetically, that “no matter what we name it, everyone will call it the ‘nudge unit.’”

  By the time of my next trip to London, the initial team had been established and was set up in temporary facilities in an obscure corner of the Admiralty Arch, located a short walk away from 10 Downing Street and Parliament. It was winter, and London had been hit with what locals considered a massive snowstorm. Accumulation was about an inch. And it was not much warmer inside than outside the drafty building that served as the team’s first home.

  The official mission of the Behavioural Insights Team (BIT) was left broad: to achieve significant impact in at least two major areas of policy; to spread understanding of behavioral approaches across government; and to achieve at least a tenfold return on the cost of the unit. The basic idea was to use the findings of behavioral science to improve the workings of government. There was no manual for this task, so we had to figure it out on the fly. On this and subsequent visits, I would often go to meetings with some high-level government official, the minister of some department or that minister’s deputy, joined by David and another team member. We would typically begin these meetings by asking what problems the department faced and then brainstorm about what might be done to help. It was vital to the success of the project that we let the departments select the agenda, rather than lecture them on the glories of behavioral science.

  The first meeting I attended went so well that I could easily have gotten the impression that this business of employing behavioral insights to improve public policy would be easy. Nick Down, of Her Majesty’s Revenue and Customs (HMRC), the British tax collection authority, had heard about BIT and had reached out. His job was to collect tax revenues from people who owed the government money. For most British taxpayers, there is little risk of falling into this situation. Employers withhold taxes from employees’ paychecks through what is called a “pay as you earn” system. For those who earn all their income through wages and salary there is no need to file a tax return and no bill to pay. However, people who are self-employed or have other sources of income besides their regular job have to file a return and can be confronted with a sizable bill.

  For taxpayers who have to file a return, payments are required on January 31 and July 31. If the second payment is not received on time, the taxpayer is sent a reminder notice, followed by letters, phone calls, and eventually legal action. As with any creditor, the HMRC views the use of a collection agency or legal action as a last resort, since it is expensive and antagonizes the taxpayer, who is, of course, also a voter. If that first notice could be written more effectively, it could save HMRC a lot of money. That was Nick Down’s goal.

  He was already off to a good start. He had read the work of psychologist Robert Cialdini, author of the classic book Influence. Many people have called Danny Kahneman the most important living psychologist and I would hardly disagree, but I think it would be fair to say that Cialdini is the most practical psychologist alive. Beyond Cialdini’s book, Nick Down had also received some advice from a consulting firm that is affiliated with Cialdini to help him think about how he might get people to pay their taxes promptly.

  Nick’s team had already run a pilot experiment with a letter that used a standard recommendation from the Cialdini bible: if you want people to comply with some norm or rule, it is a good strategy to inform them (if true) that most other people comply.† In Nudge, we had reported on a successful use of this idea in Minnesota. In that study, overdue taxpayers were sent a variety of letters in an effort to get them to pay, with messages varying from telling them what their money would be spent on to threatening legal action, but the most effective message was simply telling people that more than 90% of Minnesota taxpayers paid their taxes on time. This latter fact was also true in Britain, and the pilot experiment used a letter with similar language. The results seemed supportive, but the pilot had not been done in a scientifically rigorous manner; it lacked a control group and several things were varied at once. Nick was keen to do more but did not have the training or staff to conduct a proper experiment, and did not have the budget to rely on outside consultants.

  It was our good fortune to run into Nick Down at such an early stage of BIT’s development. He was already sold on the idea that behavioral science could help him do his job better, he was willing to run experiments, and the experiments were cheap. All we had to do was fiddle with the wording of a letter that would be sent to taxpayers anyway. We didn’t even have to worry about the cost of postage. Best of all, fine-tuning the letters could potentially save millions of pounds. BIT had a scheduled two-year run, after which it would be up for review. The tax experiment had the potential to provide an early win that would quiet skeptics who thought that applying behavioral science to government policy was a frivolous activity that was doomed to fail.

  Our initial meeting eventually led to three rounds of experimentation at increasing levels of sophistication. Michael Hallsworth from BIT and a team of academics conducted the most recent experiment. The sample included nearly 120,000 taxpayers who owed amounts of money that varied from £351 to £50,000. (Taxpayers who owed more were handled differently.) Everyone received a reminder letter explaining how their bill could be paid, and aside from the control condition, each letter contained a one-sentence nudge that was some variation on Cialdini’s basic theme that most people pay on time. Some examples:

  • The great majority of people in the UK pay their taxes on time.

  • The great majority of people in your local area pay their taxes on time.

  • You are currently in the very small minority of people who have not paid their taxes on time.

  If you are wondering, the phrase “the great majority” was used in place of the more precise “90% of all taxpayers” because some of the letters were customized for specific localities, and BIT was unable to confirm that the 90% number was true for every locality used. There is an important general point here. Ethical nudges must be both transparent and true. That is a rule the BIT has followed scrupulously.‡

  All the manipulations helped, but the most effective message combined two sentiments: most people pay and you are one of th
e few that hasn’t. This letter increased the number of taxpayers who made their payments within twenty-three days§ by over five percentage points. Since it does not cost anything extra to add a sentence to such letters, this is a highly cost-effective strategy. It is difficult to calculate exactly how much money was saved, since most people do pay their taxes eventually, but the experiment sped up the influx of £9 million in revenues to the government over the first twenty-three days. In fact, there is a good chance that the lessons learned from this experiment will save the UK government enough money to pay for the entire costs of the BIT for many years.

  The meeting with Nick Down was atypical. More often, the minister or some agency head needed to be sold on both the value of behavioral science and the need to experiment. In many of our meetings, I found myself repeating two things so often they became known as team mantras.

  1. If you want to encourage someone to do something, make it easy. This is a lesson I learned from Danny Kahneman, based on the work of Kurt Lewin, a prominent psychologist of the first half of the twentieth century. Lewin described the first step in getting people to change their behavior as “unfreezing.” One way to unfreeze people is to remove barriers that are preventing them from changing, however subtle those barriers might be.

  2. We can’t do evidence-based policy without evidence. Although much of the publicity about the BIT has rightly stressed its use of behavioral insights to design changes in how government operates, an equally important innovation was the insistence that all interventions be tested using, wherever possible, the gold-standard methodology of randomized control trials (RCTs)—the method often used in medical research. In an RCT, people are assigned at random to receive different treatments (such as the wording of the letters in the tax study), including a control group that receives no treatment (in this case, the original wording). Although this approach is ideal, it is not always feasible.¶ Sometimes researchers have to make compromises in order to be able to run any sort of trial. The next example illustrates the importance of both mantras, as well as the practical difficulties associated with running experiments in large organizations, both government and private.

  At one point I participated in a meeting in which BIT team members met with representatives of the Department of Energy and Climate Change. It was fitting that this meeting was during that week when everyone was struggling to stay warm, because the topic was how to get more people to insulate their attics, locally known as lofts. In a world of Econs, everyone would have already insulated their attic; the savings in energy costs can repay the costs of the insulation in as little as one year. Nevertheless, about a third of the homes in Britain still did not have sufficient insulation in their attics, and the department had launched an initiative to encourage the laggards to stop procrastinating. The initiative offered subsidies to both owners and landlords to better insulate their homes and install other energy-saving products. Not many people were taking the department up on their deal. The Behavioural Insights Team promised to think about what might be done.

  The proposed intervention embraced the “make it easy” mantra. When homeowners were interviewed and asked why they had not added insulation, even though it would save them money, many replied that it was too much trouble because they had so much clutter in their attics. The BIT proposed that the private firms that installed the insulation should package the insulation upgrade with an attic cleanup service. If a homeowner bought this package, two guys would empty the attic and then help the owners sort through which stuff to give or throw away and which to put back in the attic. Meanwhile, another crew got busy putting in the insulation. Two versions of this deal were offered, one at the installer’s cost (£190) and another at retail price (£271). This was on top of the cost of the insulation itself, which was £179.

  An experiment was conducted to test this idea, and the results suggest that it might be a winner. I say “might” because the data are so sparse that caution is necessary. In the interest of saving money, the only way the deal was made known to people was by mailing flyers to homes in three distinct but similar neighborhoods, picked because they were thought to have homes that were likely to be eligible for the deal. All the homeowners in a given neighborhood received the same letter,# offering the discounted cleanup, the retail cleanup, or simply the standard green deal (this latter group was the control group). Nearly 24,000 fliers were distributed to each of the three neighborhoods.

  Unfortunately, the primary finding from this experiment is that very few people were willing to insulate their attics. Whether this was because they did not open their mail, did not find the deals attractive, or rather enjoyed a cold breeze wafting down from their ceilings, take-up was tiny. In total, only twenty-eight attics had insulation installed. However, there is at least a strong hint in the data that the attic cleanup offer was a good idea. Although the sample sizes were all roughly equal, only three families accepted the straight insulation deal, whereas sixteen did with the cheap cleanup condition and nine did with the more expensive version. So nearly everyone who agreed to insulate their attics did it when they were offered some help in getting ready. However, the numbers are small enough that the experiment would need to be replicated to make one confident that the effect was real. For now, I think of this example as something between a scientific finding and a nifty anecdote.

  Much as members of the team would love to run a replication, the generally low take-up rates discouraged the department from repeating the experiment. So why include this example out of the many in the BIT portfolio? I have two reasons. First, I have never come across a better example of the Lewin principle of removing barriers. In this case, the removal is quite literal. Whether or not this specific implementation will ever be adopted on a large scale, remembering this example may provide someone with an inspiration for a powerful nudge in another situation.

  Second, the example illustrates potential pitfalls of randomized controlled trials in field settings. Such experiments are expensive, and lots of stuff can go wrong. When a lab experiment gets fouled up, which happens all too often in labs run by Humans, a relatively small amount of money paid to subjects has been lost, but the experimenter can usually try again. Furthermore, smart experimenters run a cheap pilot first to detect any bugs in the setup. All of this is hard in large-scale field experiments, and to make matters worse, it is often not possible for the experimenters to be present, on site, at every step along the way. Of course, scientists skilled at running RCTs can reduce the risks of errors and screw-ups, but these risks will never disappear.

  Frustrations aside, we must continue to run trials, and continue to test ideas, because there is no other way to learn what works. Indeed, the most important legacy of the Behavioural Insights Team may be to help nudge governments to test ideas before they are implemented. In 2013 the U.K. government established a What Works Network to encourage the testing of ways to improve government effectiveness in every domain, from health to crime to education. Every government, indeed every large organization, should have similar teams conducting tests of new ideas. But we need to be realistic about the outcomes of these tests. Not every idea will work; any scientist can attest to this fact of life.

  It is also crucial to understand that many improvements may superficially appear to be quite small: a 1 or 2% change in some outcome. That should not be a reason to scoff, especially if the intervention is essentially costless. Indeed, there is a danger of falling into a trap similar to the “big peanuts” fallacy exhibited by the game show contestants. A 2% increase in the effectiveness of some program may not sound like a big deal, but when the stakes are in billions of dollars, small percentage changes add up. As one United States senator famously remarked, “A billion here, a billion there, pretty soon you’re talking about real money.”

  Tempering expectations about the magnitude of the sizes of effects that will be obtained is important because the success of automatic enrollment and Save More Tomorrow can create the false impression that
it is easy to design small changes that will have big impacts. It is not. These savings interventions combined three important ingredients that greatly increase the chance that a program will achieve its stated goal. First, the program designers have a good reason to believe that a portion of the population will benefit by making some change in their behavior. In this case, with many people saving little or nothing for retirement, that was an easy call. Second, the target population must agree that a change is desirable. Here, surveys indicated that a majority of employees thought they should be saving more. Third, it is possible to make the change with one nearly costless action (or in the case of automatic enrollment, no action at all). I call such policies “one-click” interventions. Simply by ticking a box, someone who signs up for Save More Tomorrow sets himself on a course that will increase his saving rate over time, with no need to do anything else.

  Alas, for many problems, even when the first two conditions are met, there will not be any one-click solution. For example, it is a good bet that someone who weighs 100 pounds more than their recommended body weight would benefit from shedding some pounds, and most people in that situation would agree with that assessment. But short of surgery, there is no easy answer. I have not been able to devise an Eat Less Tomorrow program that works for me or anyone else, and we know that most diet plans fail over the long run. There is no one-click diet. Nevertheless, although we cannot solve every problem with a one-click solution, there surely are some cases where such policies can be devised, and those interested in implementing new behavioral policy changes would be well advised to search for such ideas. They are the low-hanging fruit in the public policy world.