Nudge nudge, say no more

Alan Finlayson puts it well when he writes of “the tiresome business of informing and persuading people replaced by psychological techniques designed to ‘nudge’ us in the right direction.”

I think that’s about right. Nudging makes sense as part of a package that already includes information and persuasion. For example, tell us that smoking causes cancer, persuade us to reduce or quit, then also provide nudges that make smoking more difficult and make non-smoking more comfortable. But to try to nudge people without informing and persuading them, that seems like a mistake.

38 thoughts on “Nudge nudge, say no more

  1. “But to try to nudge people without informing and persuading them, that seems like a mistake.”

    Basic economics and biology suggest otherwise (in many cases).

    Rational cognition is costly, both economically and biologically. Economically, it takes time, training and effort to think clearly and that time, training and effort can be put to alternate uses. Biologically, the brain is a costly organ and uses resources that could be put to other uses.

    Informing people is costly, becoming informed is costly, and persuasion is costly. They all consume valuable time and resources.

    Further, persuading people with poor cognition can be very costly indeed, and is often impossible. Sometimes it is even dangerous: mobs can be persuaded to terrible things, so the absence of persuasion can be a good thing.

    Therefore, basic economics and biology suggest it can be more efficient to use informing and persuading sparingly. Not everyone needs to be a car mechanic for everyone to enjoy the benefits of widespread car ownership.

    Nudging and a wide range of related phenomena happen all the time and are quite beneficial. We constantly look for signals from people better informed than ourselves. I don’t have the expertise to judge the smoking literature, so I look to others to do that and I go along with their nudges that help me avoid smoking. I have no idea how to test water purity, so I rely on nudges from the city engineers: they nudge me not to drink bad water by not allowing bad water into my tap water. I’m fine with that. I really, really don’t want to become informed or to be persuaded on that issue. I just want a giant “nudge” that makes it impossible for me to drink bad water.

    The “nudge” literature is, therefore, an example of taking an idea that is trivial and obvious and pretending it is novel and shocking. Nudging is just a division of labor that efficiently assigns tasks that use costly resources to those most able to perform the tasks efficiently. It is only seen as shocking because of the straw man in economics that rational choice is costless and that everyone is fully rational all the time. Of course, this is just a simplifying assumption in economics that is useful in some contexts. In reality, rational choice has costs, and information costs are often important. These ideas have been a major part of economics for decades.

    • Terry,

      While I agree in principle with what you say, we live in “interesting times” where trust is breaking down. Whether well-founded or not, more and more people believe that those who are doing the cognitive work for us and nudging us do not have the public interest at heart and are using their position to screw us over.

      It is a profound dilemma: it is not possible to become sufficiently expert in all the areas that are important in life to truly absorb enough information to make rational choices. But outsourcing this cognitive work to others requires enormous trust. Trust is evaporating before our eyes. What to do?

      I can’t speak for Andrew, but I read his message as saying that blank-check style trust is no longer possible and that those to whom we outsource our cognitive work need to show us more of the evidence behind what they do before we will look to them for signals.

      @Andrew: the link offers only an introductory paragraph, the rest of the article being behind a paywall. What shows in that introductory paragraph appears to be, at most, tangentially related to what you say. Is there an ungated link to the content you’re talking about here?

      • Clyde:

        Here’s the full paragraph:

        The transformations in our current era of the ways we produce information and knowledge have challenged our ideas of who is thought to be someone who knows and who, therefore, should rule. That challenge started inside government, in the New Labour era, in the form of opposition to established expertise: the view, for example, that there is no need for professionally trained teachers when you can run endless experiments on ‘what works’ and then measure how fully individual teachers comply with the findings. The New Labour years also saw the tiresome business of informing and persuading people replaced by psychological techniques designed to ‘nudge’ us in the right direction. The Cameron government intensified this approach in 2010 through the formation of the Behavioural Insights Team in the Cabinet Office. The application of such techniques to politics in general – to electoral and referendum campaigns but also and in particular to the kind of anti-feminist and anti-liberal protest politics associated with online trolls, ‘men’s rights’ activists and alt-right celebrities such as the now disgraced Milo Yiannopoulos – has led to the emergence and rapid growth of ways of doing politics that aren’t concerned with the formation of ‘a people’ but with the activation of momentary ‘swarms’ around assertions of identity or grievance, the demand for recognition of pain or enthusiasm, anger or vaporous hope. In the emergent system of communication and information, power doesn’t consist in the capacity to structure or direct what is thought and said, ‘hegemonising’ it and connecting it to a system of decision-making which is thereby legitimated. Instead it rests on the ability to read the ebbs and flows of mood and opinion so as to anticipate what is coming, find a wave that it is useful to amplify, and capitalise on the temporary force and intensity of numbers. It is a practice of politics analogous (not coincidentally) to high-frequency trading on financial markets or venture capital speculation. And it is the political right that has so far been best able to exploit it.

        But for my purposes all that was relevant was the phrase I quoted in the above post.

        • The application of such techniques to politics in general – to electoral and referendum campaigns but also and in particular to the kind of anti-feminist and anti-liberal protest politics associated with online trolls, ‘men’s rights’ activists and alt-right celebrities such as the now disgraced Milo Yiannopoulos – has led to the emergence and rapid growth of ways of doing politics that aren’t concerned with the formation of ‘a people’ but with the activation of momentary ‘swarms’ around assertions of identity or grievance, the demand for recognition of pain or enthusiasm, anger or vaporous hope.

          What is it about politics that leads to inappropriately long sentences like this? A level up, that paragraph needs to be broken up as well. On the surface it looks like having a complete disregard for the audience, but perhaps there is some rhetorical goal I am missing. Maybe it is from reading legalese so much you begin to think that is normal communication?

        • I would be interested in people’s honest reaction to this passage.

          Is it well reasoned? Hysterical? And expression of genuine concern? Garbage? A serious attempt to understand modern political trends?

        • Do the paper provide any reference to justify the statement that the UK government(?) applied nudge to “anti-feminist and anti-liberal protest politics associated with online trolls, ‘men’s rights’ activist…”? Is that true?

      • “But outsourcing this cognitive work to others requires enormous trust. Trust is evaporating before our eyes. What to do?”

        Good point.

        When there is a crying need for something, something often arises to fill that need.

        Lack of trust is a big problem in markets and there are many ways it is addressed. (1) Reputation, (2) delegation of monitoring to experts who put their reputations on the line, (3) delegation to experts you hire to look into things for you, (4) natural human suspicion.

        All these things are, of course, imperfect and there is a constant arms race between the charlatans and the sheep.

    • Terry:

      I think your examples correspond to persuasion + nudging, not nudging alone.

      You write, “I don’t have the expertise to judge the smoking literature, so I look to others to do that and I go along with their nudges that help me avoid smoking.” But they’re not just nudging you, they’re also explicitly telling you that smoking is bad for you.

      The thing I’m reacting to is the idea that nudging should happen in the absence of any persuasion.

      Your example of the tap water is different, I’ll grant you that. I guess I don’t think of that as a nudge at all, as they’re not trying to alter your behavior, they’re just trying to alter the consequences of your behavior.

      Here’s another example that fits your story: rumble strips on the highway to alert you that you’re driving on the shoulder. There, I’ll agree with you that this is much more direct than trying to persuade people to stay alert on the roads. So I guess it depends on the context.

      • Sure. There’s evidence that telling people their neighbors have voted in the last elections increases voter turnout. You could imagine that providing information about non-voting neighbors might depress turnout. All you have to do is target your opponent’s supporters, instead of your own.

        A few years ago, I saw a pharmacy advertising the flu shot with a message along the lines of “get a flu shot, because your friends won’t!” I guess the idea was that your friends will get sick and infect you. My sense would be that this is more likely to reduce the vaccination rate, because it conveys that the social norm is not to get the flu shot. This was, of course, not their intent.

      • Window displays to entice shoppers into a store seem to be an example of nudging with intent that is good for the store but bad for the consumer. Same with lots of types of advertising.

      • The whole concept of “choice architecture” rests on the idea that you are going to make a choice about how other people make choices, so you might as well put some thought into the effect you will have. There’s a presumption that the person receiving the advice is somehow altruistic, but one can always just do the opposite.

    • I’m not going to buy that “water treatment” is a “nudge”. Water treatment is a service, like providing you with food or clothing or gasoline. A nudge is something that alters what you choose to do. A small increase in a tax on cigarettes is a nudge. Banning cigarette sales is not a nudge.

      • (I’m no expert on nudging, so the following is just my opinion.)

        Economically a small tax is a (small) incentive. A “nudge” in the cigarettes example is the requirement to put nasty pictures on the packaging. A (abstract) rational being would be indifferent between the two option (normal packaging vs. disgusting pictures), because it knows the consequences of the decision did not change. A normal (irrational) person might be nudged towards not consuming — however, coming back to the topic of the post, I also think that this can/should only happen in conjunction with persuasion (that smoking is bad).

        My point is, when talking about nudging we should always make clear the distinction between nudging and “conventional” incentives. A lot of things people label as “nudging” are actually just plain incentives (which can be small).

        • Yes, good point. there are two issues:

          1) incentives vs nudges
          2) nudges vs shoves

          As you point out, a tax actually changes the materially important inputs to the “utility” that is, dollar costs go up. A nudge is more like engaging less tangible costs (ie. the cost of having to look at that cancerous lung picture)

          But, if you require a piercing siren on every pack of cigarettes that goes off every time you open it, this is not a nudge, it’s too “big” more like a shove, or a pounding.

  2. I agree with Terry above, but would like to also point out that we lots of nudges with no information or persuasion. Criminal law is one big nudge — do this only if you don’t mind going to jail, but we neither make an effort in most cases to tell people what is legal and what is not, nor do we attempt to persuade people in advance of making things criminal.

    We put a penalty on not having health insurance (currently at least temporarily rescinded) but without making much of an effort to convince people to get health insurance (largely because the benefits of doing so are fairly elusive for the people we are seeking to nudge with the penalty.)

    The archetypal successful nudge, default contributions to a 401(k) plan, while certainly assuming that people would be better with the default, assiduously avoids telling people that they’d be better off contributing, because if you convince people to do something and they end up losing money, you’re open to liability, even if on average it’s a good idea.

    • This is a really different meaning of the word “nudge”. If I stand next to you and gently push you towards the direction of a friend I want to introduce you to, that’s a nudge. If I walk up to you and pound you into the ground until you agree to meet my uncle, that is not a nudge. I know which “going to federal penitentiary for 10 years” is more like… why don’t economists?

        • OK… I accept that criminal changes your economic incentives (many people don’t buy that, but I do.) Thaler and Sunstein define a nudge as “any aspect of the choice architecture that alters people’s behavior in a predictable way without forbidding any options or significantly changing their economic incentives.” As Terry points out above, if you accept the costs of thinking as an important economic cost, then there probably are very few nudges, although “significantly” is sort of a weasel word there.

          So I retract the claim on the criminal code, which was a little hasty. I’m sticking with the 401(k) example, and the medical insurance example depends on your definition of “significance.”

        • reducing barriers to participation in a 401k isn’t really a nudge either is it? I mean,

          Before: you have to go to the HR department, get them to engage with you, give you the paperwork, find all the information to fill it out, choose your contribution level, etc etc

          After: the HR department does all of this automatically for you unless you specifically go and tell them to do something else.

          How is this not reducing costs to you?

          A nudge would be more something like having a really nice espresso machine outside the office of the HR people so that people tend to go down there anyway.

        • “Thaler and Sunstein define a nudge as “any aspect of the choice architecture that alters people’s behavior in a predictable way without forbidding any options or significantly changing their economic incentives.”

          I think Thaler and Sunstein have to work extra hard to make their insight sound novel.

          Once you realize that information acquisition is costly, you see that people will gladly accept guidance on many issues that other people have already figured out. For these people it takes only the slightest hint or nudge to tell them what to do. Examples are legion: (1) Should I wear a coat today? (2) If I take a right turn will I fall off a cliff? (3) Should I buy a car with wheels or without wheels?

          Sure, you can still take a right turn if you want to, but you won’t once I tell you you will fall off a cliff.

          Sure, you can buy this car without wheels if you want to, but you won’t once I tell you everyone who has ever bought a car has bought it with wheels.

          They are just making the point that cheap information will change people’s behavior. Duh. The default option when buying a car is to buy wheels with it. You can call that a nudge if you want, but explaining why all cars come with wheels is not a deep insight.

        • “Persuasion” in the sense that Andrew means it here is not “coercion.” It means getting you to believe that something is desirable in itself.

          That belief alone is not enough to change a habit, however. A little psychological game might be enough at that point, however.

          I do suspect that the “Nudge” movement, however, is often about an elite finding economical ways to cause other people to act like they want them to act. From that perspective, some think that a coercion that costs $1 per person and works 80% of the time is better than one that costs $5 and works 95%.

    • Other Jonathan, I don’t agree: criminal law isn’t a nudge but a set of rules you follow or not. A nudge is more like social pressure to stand in line or to pull your car over into one lane though traffic engineers tell you that’s inefficient. A nudge, as DL points out, is more like a cue that you should go talk to that girl or guy. It’s a social signal that this is ok or not ok. This is separate from economic rationality: a nudge can be toward irrational, destructive behavior.

      A penalty is a stick. A reward is a carrot. Making people contribute to retirement – social security is that – is neither: it’s a social policy goal enacted as a law and thus as a rule. Laws have effects but there’s no connection between laws on books and lawfulness or lawlessness. If the police put a car on a curve every day to keep speeds down, that’s a warning, a form of stick, not a nudge.

      A campaign to get people to hold doors open might qualify as a nudge if it’s just about instilling politeness as a way of generating civility. A campaign that says you go to jail for spitting is not a nudge but a stick.

      • “A penalty is a stick. A reward is a carrot.”

        Good point. There is a fundamental difference between coercion and providing useful information. When given useful information, people will often act on it voluntarily. In some sense, this is persuasion. Coercion, however, makes you do things that would otherwise not be in your self interest.

  3. Thinking about how to encode zip codes in machine learning methods, I suggested an algorithm that essentially uses as a features the difference in performance of the algorithm between the model using the zip code and the algorithm without using the zip code, that difference becomes a new feature vector, and this procedure can be applied sequentially comparing the performance betweeen the last model and the model without the zip code.
    In a post in HN I suggested the following:

    Self-concious Statistical Methods: Difference in Performance as a Feature
    1 point by autocon7 37 minutes ago | hide | past | web | favorite | edit | delete
    Recently someone asked in a blog how to incorporate zip code to machine learning algorithms. Thinking about this problem, I suggested the following algorithm: Apply your machine learning method to your data without using the zip code variable, then for each zip code compute the associate model using only the data with that concrete zip code. Now, for each zip code compute the difference in performance between those two models, that difference in performance becomes our new variable, so one would expect the machine learning algorithm to use this new variable to improve the prediction, and at each step we can use as a new variable the difference in performance betweeen the model with the old information and the new information provide by the difference in performance. I would expect a good algorithm to converge to a fixed point in which the new variable doesn’t change form one iteraction to the next. If the algorithm converges to a fixed point then one could call the algorithm auto-correcting or autoconscience because is using the new information to take conscience of itself (how it performs on input).

    Perhaps such kind of algorithm is long time known, but it seems a promising algorithm when coupled with an adequate model. It seems that the book of Andrew Gelman about hierarchical models can contain information about such questions. Can some expert on statistical models chime in and give some references about this algorithm?

    Edit: Such kind of algorithm is trying to supervise itself to improve. The fixed point obtained is a way to encode the information provided by the new variable to the selected model, so that fixed point is model dependent. If the new variable doesn’t provide new information then the difference in performance should be statistically zero, and so the new variable is constant equal 0, and this becomes the fixed point. So any algorithm able to compute averages shoud be able that the information of random variables is zero.

    Perhaps that kind of algorithm is well known. Can you comment about this?

  4. I think some of the authors above really stretch the definition of a nudge to the point where it’s synonymous with either regulation or public policy. In that light, sure any “insight” is trivial: we know that laws, regulations, and taxes affect behavior. That’s not what nudging is about.

    But what is less trivial is the magnitude of an impact a default option can have in very high stakes decisions. Take retirement savings: changing a default savings rate from 0% (i.e. you have to sign up to join) to 6% can translate into hundreds of thousands of additional dollars saved in retirement. And that’s before considering the “free” money from employer matching.

    So could we get the same outcome just by providing information and persuading? I don’t think so. Most large employers already do heavy educational interventions and have been doing so for years. Employees are often required to attend information sessions about their retirement benefits, where they learn about what a 401(k) plan is, what matching the company provides, and what investment opportunities they have — and they get to see the nice figures explaining compound interest. If you do surveys after, everyone wants to sign up and at least get the company matching (hardly a trivial amount). Then you follow-up a few weeks later and virtually nobody did anything.

    Is that a rational choice to not save and we’re missing something? Not really: you could always withdraw the money from your 401(k) early. You pay a small penalty, but that’s more than compensated for by the company matching. So now we are left to conclude that people are willing to leave thousands of dollars on the table every year because they don’t want to spend an hour signing up for the plan and 10 minutes per year doing an early withdrawal. That this is happening and can easily be fixed by changing a default is now far from a trivial insight, especially because education is known to not work nearly as well.

    This is not just true in financial decision making, of course. Consider calorie labeling: fantastic regulation that makes available important information to the millions of people who are counting calories or otherwise dieting… zero effect on how many calories people actually consume. So either people already knew how many calories are in all their meals (pretty unlikely) or they are simply not as affected by it. For an example of a nudge, take Starbucks’ decision to serve lattes with skim milk instead of whole milk by default. How many people even know what milk they get by default and would be willing to change it with every order? Given the millions of lattes that are served, that’s a sizable impact.

    What seems to get lost in the conversation on nudges is that they are not intended to substitute for other policy. We don’t want households to choose from different water suppliers and nudge them toward the ones providing clean drinking water. Instead, we regulate water quality and mandate clean drinking water. Similarly, nudging people to save more for their own retirement cannot be a substitute for social security. But nudges can be helpful complements in many settings. We won’t solve the obesity problem by redesigning cafeterias, but if we can get students to consume 50 fewer calories at lunch by redesigning the layout, we prevented them from gaining 2-3 pounds per year at no cost. It just doesn’t imply that we now no longer need a gym class — even as many of these “nudge critique” papers seem to try and impart that motivation on those advancing the use of nudges.

    • My undergrad university dining halls used plates smaller than the usual—more like a formal luncheon plate than a dinner plate.

      I’d call this something of a nudge. Rather than taking a “large” amount of food at the first pass, you had to actually get up and walk back to the serving line. My guess is that this reduce food waste and cost without in any way limiting what the student wanted to eat.

    • “So could we get the same outcome just by providing information and persuading? I don’t think so. Most large employers already do heavy educational interventions … If you do surveys after, everyone wants to sign up and at least get the company matching (hardly a trivial amount). Then you follow-up a few weeks later and virtually nobody did anything.”

      You state the case very well.

      The nudge concept is not trivial to people enamored of the paradigm that humans are ultra rational and efficient decision makers. If people were, modest education would make almost everyone sign up for a 401k. But, as you point out, this doesn’t happen. So, you effectively demolish the ultra-rational mentality that permeates economics. (Indeed, I think Thaler was given a Nobel to balance Fama’s Nobel for the efficient market hypothesis.)

      When you talk to actual people, though, you find that many are far from ultra rational and efficient. Many are also inattentive and some are downright lazy. So, an hour’s paperwork is enough to stop many from signing up. So, I fully agree that you should set a default so inattentive people do the right thing.

      Perhaps I don’t see this as novel because I see it everywhere. Car dealers don’t require buyers to spend an hour’s paperwork to get wheels on their cars. They don’t even give them the option because some fraction will take them up on the offer for god knows what reasons.

      I suspect larger, political forces are at work. Thaler and Sunstein want to accustom people to the idea that someone (preferably government) needs to make decisions for people, and ultra rational economists want to accustom people to the idea that people should be left to make their own decisions. So Thaler and Sunstein choose the 401k example because it is a good example of a modest intervention that is clearly beneficial. They also want to present it as a new insight to imply that it isn’t already being done so we should do more of it.

      • I think your intuition about Thaler and Sunstein is quite the opposite of what they would argue for. They are both concerned with paternalistic regulation and tend to oppose mandates (their original paper is titled “Libertarian Paternalism”). This is because policymakers, too, act based on limited information and individuals may have very good reasons to deviate. Being forced to save for retirement via a pension when you pay 20% interest on credit card debt, for example, is not exactly good. Similarly, you shouldn’t save for retirement while in grad school, whereas someone with the same income in a dead-end job should be saving. With a 401(k) plan, one can adjust based on information unavailable to a policymaker.

        Although I agree with you that this seems like common sense, it’s worth noting that policymakers have not been considering such interventions in the past. Someone had to popularize the concept and bring in lawmakers — almost none of whom, by the way, are economists. There was also a need for new regulation. For example, employers had to be protected from lawsuits of employees who were defaulted into a (conservative) plan and still lost money.

        Unrelated, but Fama shared the Nobel with Bob Shiller, who argued against the rationality of markets. One of the interesting things about the economics Nobel is that, at times, it is shared by people who reached opposite conclusions. I think we can probably all agree that there are indeed many cases where markets are incredibly efficient and some cases where they went awry. What happens often is that we forget about the many instances where standard economic models work really well — despite relying on very few assumptions and often having been designed to answer different question (and by a different researcher). That’s hard to do in the social sciences and a stark contrast to, say, psychology where the joke goes that theories are like toothbrushes: no respectable person would use someone else’s. Thaler, by the way, also got the Nobel not just for nudges, but for much of his foundational work in behavioral economics.

  5. Andrew,

    Providing information is nudging:

    “Disclosure of relevant information (for example, about the risks of smoking or the costs of borrowing) counts as a nudge” (Sunstein 2016: 124).

    “If serious risks are involved, the best nudge might be a private or public warning” (Sunstein 2014: 586).

    In fact, Sunstein (2016) finds that information nudges are often people’s preferred type of nudge.

  6. Some of the discussion above appears to be just fine distinctions between the size of nudges and which word is most appropriate for each.

    But, the fundamental point remains that much of human decision making is delegated to specialists who (hopeully) are more efficient at gathering and analyzing data and making decisions and these decisions are transmitted to other humans by various means. Sometimes people look to others for guidance with minimal internalization of the reasons, so social cues are enough and an overt “nudge” is not necessary. Sometimes we coerce people things. Sometimes this coercion is mild and sometimes it is very strong and the word nudge no longer seems appropriate.

    Fundamentally, though, it is the same type of delegation, and we are just differing in the words we use to describe the transmission. The best system will depend on the costs of information acquisition, the difficulty of decision making, and the most efficient way to transmit the results to other people via nudges/coercion/persuasion/education. This will vary from case to case, and so blanket statements, at best, apply only to a particular set of circumstances.

    Some examples:

    Criminal law is a very strong nudge to the point where the word “nudge” doesn’t seem appropriate. Safe drinking water is another example. It is so obviously a good thing and so cheap that we just do it for everyone. “Nudge” doesn’t sound appropriate because no one really wants to resist. But, in theory, I COULD drink puddle water from the street, so this is an example of a really strong nudge.

    A 401k nudge is fairly mild and people sometimes override it. We do this because sometimes people have a reason for opting out and we want to allow that because opting out isn’t such a bad thing. But, most people go along because they want someone more informed to make the decision for them.

    Sometimes a “nudge” includes “persuasion” which often amounts to just internalizing social pressure. People have internalized that smoking is “bad”, so social pressure against smoking has had the desired effect. But this isn’t intellectual persuasion because few people have the ability to evaluate the evidence, so they are only “persuaded” in the sense that they have internalized the social pressure – which can be a very good thing: societies transmit much good knowledge this way.

    Sometimes actual intellectual “persuasion” occurs as people are taught to see why an action is good. This is worth the effort when the lesson can be extended to other situations. Being shown how germs cause sickness is excellent persuasion because people will know how to avoid many different diseases. Here, the benefits of education are so high that it is worth the cost of education.

    Bottom line: same basic idea in each of these instances, but the implementation and nomenclature depends on circumstances.

Leave a Reply

Your email address will not be published. Required fields are marked *