What's in a Nudge? Part III
To Nudge or Not To Nudge?
We’re back for our nudge series finale. This has been an absolute blast to write!
To recap, in Part I, we analysed the core tenets of Nudge Theory, discussed the issues with them, and decided we should discuss nudge-type behaviour change techniques (BCTs) separately from Nudge Theory. We then proposed an operational definition of a nudge. In Part II, we examined the various nudge-type BCTs, and found not all of them met the conditions of our definition. We then proposed a framework for understanding their mechanisms.
At last, we turn our attention to the most important (and clichéd) question of all—should we nudge?
There are essentially two considerations when deciding on behaviour change:
What is the nature of the problem?
Is my solution right?
As we shall see, successfully changing behaviours (through nudge-type BCTs or otherwise) hinges on a nuanced, thoughtful treatment of these two considerations.
Maladies in Gluttonsbury
First, let us make a detour into the fictional town of Gluttonsbury. The inhabitants of Gluttonsbury, or Gluts as they affectionately call themselves, take immense pride in their food culture. Lunch breaks are three hours long to allow “proper nourishment”. Dinners are often social events till the early hours of the morning at the various excellent food establishments lining the main street. The town is also home to one of the world’s biggest food processing plants, and the fabled Gilly Glinka’s Chocolate Factory. Highly affordable ultra-processed foods are staples in Glut’s diets at home, but not fruits and vegetables.
As you might imagine, obesity is a hefty issue in Gluttonsbury. Mayor Glover understands that Gluts’ propensity to overeat during meals is perpetuating this weighty problem. He commissions you, a behavioural scientist of the highest calibre, to run a campaign to discourage overeating during meals, in hopes of lowering obesity rates. You run a smart, effective campaign with the slogan “NOT TOO MUCH, NOT TOO LITTLE, JUST RIGHT!”, provide salient, targeted information on the importance of refraining from overeating, work with food establishments and school canteens to reduce plate sizes, and start a “EATING JUST RIGHT!” challenge leveraging on social norms to reduce overeating.
After 6 months, the average Glut is consuming 14% less calories on average each meal and obesity rates have started to fall, decreasing from 62% to 60%. Mayor Glover is glowing with glee. And so should you. Right?
What’s the Problem?
Sometimes, the simplest questions are the hardest ones to get right.
Often, we assume problems are as they are presented to us. Often, that assumption is wrong. Good design is problem-centric, not solution-centric, but we rarely spend the necessary time and effort to understand the true nature of problems. Instead, we often tend towards simple, reductive views of problems. This tendency presents in the form of two thinking traps.
Trap 1: Focusing on Individuals Not Systems
At the heart of the recent discourse on nudging (and the behavioural sciences more broadly) is a fundamental mismatch in how problems are scoped.
If your focus was on reducing Gluts’ meal time calorie intake, your campaign was a resounding success by all accounts, given the average nudge intervention results in far more modest changes (more on this later). If you view obesity as a function of Gluttonsbury’s food-centric urban design, industries, and imbalanced food economics, then you would view your campaign as woefully inadequate. No amount of nudging will reduce obesity to an acceptable level. In fact, we can’t quite be sure these changes will even persist a year from now. Seen in this light, the campaign only serves to deflect attention from what should really be done—redress the imbalance in food economics and restructure Gluttonsbury to facilitate healthier relationships with food.
In short, there is a tension between scoping problems at what the behavioural scientists Nick Chater and George Loewenstein term the i-frame (individual level) vs. s-frame (systemic level). In the former, problems are viewed as issues of individual responsibility—thus solutions should target individuals’ behaviours. Traditionally, this is understood to be the domain of nudge-type BCTs. In the latter, problems are viewed as systemic or structural, and behaviour change is effected through changes in the systems. Traditionally, this is understood to be the domain of policy action such as regulation and taxation.
Chater and Loewenstein argue that by focusing on nudge-type BCTs, behavioural science tinkers around the edge of fundamental problems without truly addressing them. To them, not only are nudge-type BCTs less effective, they also draw resources away from the more impactful s-frame alternatives.
In my opinion, Chater and Loewenstein’s argument is compelling, but rests upon a false dichotomy. Individuals don’t exist in vacuums, but as part of systems. Systems are not monolithic structures, but are composed of individuals and individual actions. You simply cannot change one without changing the other. Understanding problems require us to consider both and identify who (e.g., end users, policymakers, corporations) and which of their actions are at the heart of the problem1.
Trap 2: Focusing on Individual Behaviours Not Extended Patterns of Behaviours
We also tend to focus narrowly on individual problem behaviours. Nothing wrong with that sometimes. Men really ought to aim better in the toilet. More frequently, the problem behaviour in focus is but one part of what the behavioural scientist William Baum describes as temporally extended pattern of behaviours. Smokers don’t just smoke. They also tend to consume more alcohol, have a poorer diet, and exercise less. In short, they display a tendency to discount their long-term health in favour of short-term pleasure. A narrow focus on curbing smoking misses the point—it won’t do to put out the fire on a porch when the whole neighbourhood is burning down.
For Gluts, overeating during meals, while no doubt problematic, is but one piece of an extended pattern broadly encompassing their relationship with food, including the types of food that they eat, when they eat them, and their cultural attitudes toward food. To expand our conceptualisation of the problem further, Gluts socialise through food, and food-related activities dominate Gluts’ leisure time, crowding out health-promoting behaviours such as physical activity.
Real-world problems are complex, with many interrelated and interdependent components. They wouldn’t be actual problems otherwise. Resisting our tendencies toward narrow, reductive conceptualisation of problems is effortful, but those who do so will be well rewarded for their efforts.
With that, we are ready to consider our potential solutions.
Is My Solution Right?
Questions about rightness come in two forms:
Is it effective—does it achieve the intended outcome?
Is it ethical—are the means of addressing the problem acceptable?
Do Nudges Work?
Let us first examine the evidence. Recently, a research group at the University of Geneva analysed over 200 nudge-type interventions studies and concluded nudge-type BCTs work, to a small-to-medium degree. They were quickly refuted by another group of researchers who reanalysed the same data and claimed that the published studies were cherry-picked2. Separately, two researchers from the University of California, Berkeley, examined evidence from 126 real-world implementations of nudge-type BCTs and found nudge-type BCTs produced slight changes in behaviours (1.4% in absolute terms, or 8.0% in relative terms). Based on these findings, one might conclude that nudge-type BCTs are ineffective, or only incrementally effective at best. Furthermore, it remains unclear how lasting these changes are.
Before we dismiss nudge-type BCTs, it is important to note that many of them were derived from behavioural principles that were developed and validated in animal and human research over the past decades, long before the term nudge was coined and popularised by Richard Thaler and Cass Sunstein. As an example, in a classic experiment published in 1972, the behaviour scientists Howard Rachlin and Leonard Green nudged pigeons to consistently select larger-later rewards (pigeons typically prefer smaller-sooner rewards) by manipulating the duration before choices were presented. As behavioural scientist Marco Tagliabue puts it:
Nudging has a short history, but a long past.
So, what gives?
First, to restate our earlier point: good design is problem-centric, not solution-centric. The effectiveness of nudge-type BCTs depends on correctly identifying problems and fitting our solutions accordingly. The goodness of fit depends on the contexts in which they are used, and a correct understanding of their mechanisms, which as discussed in Part 2, are highly diverse.
Second, much of the evidence is gathered through experiments conducted in tightly controlled environments, for narrowly defined problems, within short time periods, and without due consideration for feasibility nor sustainability. In short, the evidence is mostly based on simplified problems, against the backdrop of a static (and supportive) context. It is naive to think that out-of-box theory-driven implementation of nudge-type BCTs in real-world contexts will work exactly as results from an academic paper would suggest. Rather, real-world problems call for tailored solutions, often a combination of nudge-type BCTs and other tools.
Finally, extending from the previous point, successful implementations need to be data-driven. Even under the best circumstances, things almost always don’t go as expected. Constantly measuring and evaluating our interventions’ impact allows us to refine and tailor our approaches, and adapt our approach as contexts or even problems themselves change (they most certainly will).
Are The Means Acceptable?
From the outset, Nudge Theory attracted a great deal of ethical concerns. We will broadly summarise three of them3.
1. Autonomy
Critics argue that the use of nudge-type BCTs violate the individual’s autonomy, either by limiting choices or imposing costs on certain options. Costs can come in the form of increasing effort required to perform certain behaviours, or adding a costly outcome to a choice. To be clear, nudge-type BCTs that supposedly violate autonomy tend not to place limits on the availability of choices, but rather on how individuals exercise those choices. To use two examples discussed in Part I:
Ordering healthier options to appear first in a food app, and placing fast food options in a separate tab preserves the number of options available in principle, but makes fast food options less accessible, and thus, less likely to be selected. Further increasing the effort to access fast food options further decreases that likelihood.
Using opt-out defaults to increase organ donation consent rate can bypass the individual’s choice. Here in Singapore, all individuals are automatically enrolled as organ donors upon turning 21 years old by law, which we are presumably aware of (many are not). Individuals can opt out by filling in a form, but few are aware of this option, or know the processes involved.
To be fair, the degree to which nudge-type BCTs violate autonomy varies greatly and depends partly on their design and implementation. In implementing organ donation consent defaults, on one end, Singapore automatically enrols citizens upon crossing an arbitrary age threshold without explicit communication. At the other end, we might imagine a deliberative process, where individuals reflect on their own role, their family’s role4, and the impact of their decision on their wider community before making a choice. They then make a choice during an offical process (e.g., when renewing their passport), that clearly states enrolment by default, but with a clearly indicated option to refuse. Obviously, the latter approach preserves autonomy more than the former.
2. Welfare
For better or for worse, those who design and implement nudge-type BCTs are typically in positions of power, concerned with what they want. Getting adolescents to reduce social media usage is what a policymaker wants, regardless of whether it reflects adolescents’ wants. On the flip side, social media titans want the same adolescents to stay on their app forever5.
Let’s ignore for a moment the fact that actors seek to influence behaviours for all sorts of reasons and focus our attention on those acting in “good faith”. Justifications for nudging come under two pretexts:
It’s what they want. Arguably, Gluts really want to lead long, healthy lives (termed higher-order or true preference), not eat the sinful delicacies in front of them (lower-order preference). I find this argument uncompelling. Let’s face it—they want both. It’s their relative preference for both that matters, which shifts from moment to moment, which may differ from what the next person wants. It takes an oracle of extraordinary prescience to know true preferences, if such a thing even exists.
It’s for the greater good. Obesity-related diseases come with huge social and economic costs. Thus, Gluts really should consume less calories, whether they like it or not. The degree to which we can agree with this pretext depends on the problem at hand. Nudge all you want if it decreases infant mortality. But intrude upon Gluts’ sacred relationship with food and you might find a less than warm reception. Also, we must question whether Mayor Glover (and you by extension) has the tacit mandate to intervene in Gluts’ eating habits by virtue of being their elected representative, or if this mandate must be separately sought.
3. Manipulation and Opacity
People are smart. They know what they want and what’s good.
The above two considerations are important to deliberate upon, but are by no means unique to nudge-type BCTs. They apply to all the tools one might use to influence behaviours, including regulation and taxation.
For nudge-type BCTs in particular, one concern goes beyond violation of autonomy to violation of the individual’s identity itself. While other tools such as bans and fines are transparent (the consequences for flouting a restriction are clear), certain nudge-type BCTs can guide decisions in ways that bypasses one’s awareness. It is an uncomfortable thought, frightening even, that some external party can infiltrate our subconscious a la Inception and create a desire to spend lavishly on diamonds without our knowledge. To give two examples:
As discussed in Part I, individuals respond asymmetrically to loss frames and gain frames of mathematically equivalent choices. Knowing this, actors can influence individuals to save more, go for their annual health screening, or perhaps, buy a car we don’t need.
Our decisions tend to be influenced by reference points (termed anchoring). This makes evolutionary sense—we typically use what we already know or gather information in our environments to make decisions. But sometimes, our references can be tethered to arbitrary values. In negotiations, price setters tend to set arbitrarily high prices, knowing that the quoted price is likely to make buyers pay more than they would have otherwise.
We are constantly subject to such subtle manipulations. Marketing companies today employ many of the same tactics detailed more than four decades ago in psychologist Robert Cialdini’s classic work: Influence: The Psychology of Persuasion. Social media titans influence behaviours through sophisticated algorithms that are opaque even to the engineers that develop and tune them.
As a presumed “good faith” actor, you will have to decide (ideally with the recipient’s input) if the ends justify the use of non-transparent nudge-type BCTs for the problem you are trying to solve. Different folks will have different takes on this.
Summary
In this series, we’ve dissected nudges, explored their mechanisms, and considered how and when we should use them. Where does that leave us?
I hope to impart to the reader three key takeaways:
First, identifying problems and their boundaries should be central in any design process. Having a clear view of problems allow us to identify correctly the target of our behavioural change intervention, including the who and which behaviours. That in turn guides the tools we employ.
Second, nudges encompass a diverse group of BCTs with diverse mechanisms of action. We can do a lot with a Swiss army knife, but only if we are familiar with what its tools are for, and how they work. Sometimes, our Swiss army knife does not have the right tools, or needs to be used in combination with other tools.
Finally, questions of ethics and morality are valid and important, but tend to get sidetracked as adoption takes priority. We do not yet know the long-term impact of the use (or misuse) of nudge-type BCTs on the relationship between the implementers and end users. If the responses to behaviourally-informed COVID-19 policies in the UK are anything to go by, we still have some ways to go in figuring out how to implement such solutions right.
In the end, the question isn't just whether nudges work—but whether we are designing the right nudges, for the right problems, in the right way.
Epilogue
I would like to leave with you one of my favourite examples. Here is a solution to reduce consumption of unhealthy beverages in Singapore, which combines nudge-type BCTs and other policy tools.
Since 2022, all beverages sold in Singapore must include a nutrition labelling, called Nutri-Grade. Beverages are assigned a grade from A to D, corresponding with their sugar and saturated fat levels. Nutri-Grade labels also display the percentage of sugar in the beverage and advertisements were also banned for D grade beverages. As a result, sugar consumption through beverages has decreased, mirroring outcomes of similar initiatives from other countries.
Why did this work?
First, Nutri-Grade was effectively designed. It provided salient information to consumers who would not otherwise be aware of the sugar content of beverages. The information was clear and easily interpretable through colour codes, and provided in a timely manner, directly at the point where the choice is made. Consumers also value reducing their sugar intake from beverages. Taken together, Nutri-Grade shifted consumers’ preference toward healthier options.
Second, and more importantly, we also witness marked shifts in the behaviours of producers, including reducing sugar content in their products and reworking their marketing strategies to feature low- or no-sugar products. This meant that consumers who did not shift their choices toward healthier options were still consuming less sugar by default.
To be clear, we cannot claim Nutri-Grade alone caused these shifts in producer behaviours. In fact, beverage producers were already pledging to reduce sugar content of their beverages before Nutri-Grade was even announced. Nutri-Grade is but one of many solutions being rolled out in Singapore, and in other countries with much larger consumer markets. Insofar as Nutri-Grade might further move the needle, considerations might boil down to whether they sufficiently shift consumers’ preference, and how that enters a complex interaction with brand reputation, demand elasticity, first movers advantage, and so forth to influence producers’ behaviours.
I don’t claim to have the right answers, but I hope I’ve given you the right questions to think about.
By virtue of engaging in behavioural design, you yourself are introduced into the systems built on the behaviours you are trying to change.
In organ donation decisions, the family typically are the decision makers on the deceased’s behalf.
Mark Zuckerberg would argue that adolesents usage of Instagram clearly represents their wants. Others argue that social media usage is creating a profoundly anxious and unhappy generation.
Fantastic series! Thanks for the thoughtful commentary and great examples