What's in a Nudge? Part II
Diving into the behaviour change techniques in the nudge universe
I originally intended for this series to be written in 2 parts. In the course of writing this article, it’s become clear that we need a Part III, which will be published in the upcoming weeks.
Welcome back to our series on nudges!
In Part I, we examined the core ideas of Nudge Theory, briefly discussed the issues with its theoretical framework, and decided to separate nudges from their troubled theoretical framework.
We then adopted an operational definition of a nudge proposed by Carsta Simon and Marco Tagliabue:
Nudging focuses on the role of antecedents of behaviour when analysing choice architecture. Changes in choice architecture referred to as nudging, do not program changes in consequences.
Simply put, nudges influence behaviour through manipulating antecedent stimuli (stimuli that predict outcomes) but not changing the outcomes themselves.
In getting a food delivery app users to make healthier choices, ordering the options such that healthier choices appear first is a nudge. Providing a discount on healthy options or jacking up the price of unhealthy options is not. In getting employees to recycle more, increasing the number and accessibility of recycling bins in an office is a nudge. Putting up a leaderboard showing which teams recycled the most/least each week is not.
We also noted that manipulating antecedent stimuli only works well if the stimuli predict outcomes that are important to the individual. Ordering healthier choices to appear first in a food delivery app is unlikely to make a dent in the eating habits of a fast-food fiend. Increasing the accessibility of the recycling bins is unlikely to increase the recycling rates of a climate change cynic1.
Today, we will examine some behaviour change techniques (BCTs) subsumed under nudges and delve into the mechanisms by which they influence behaviours.
Nudge-Type BCTs
One barrier to understanding and using nudges effectively is that they really are a highly diverse collection of BCTs. In Chapter 5 of Nudge: Improving Decisions About Health, Wealth, and Happiness, Richard Thaler and Cass Sunstein (henceforth T&S) listed 6 types of nudges, including setting defaults, providing information, providing feedback, simplifying choices, error-proofing, making incentives salient, and provided examples of others, including framing, providing/aligning incentives, ordering choices, and recruiting social norms. All these BCTs influence behaviour through very different mechanisms.
Since then, the nudge universe has further expanded to include more than a dozen other nudge-type BCTs, including just-in-time prompts, creating friction, using commitment devices, and so forth. Without a framework, figuring out what all these BCTs are and how they operate on behaviour becomes overwhelming rather quickly.
To assist the reader, keeping in mind our definition, we propose that most nudge-type BCTs fall within 4 categories2.
Only the first 2 actually involve manipulating antecedent stimuli, thus strictly qualifying as nudges under our definition. The third involves manipulating some dimension of the target behaviour or preceding behaviours, which some may argue falls outside our criteria. The last involves directly manipulating outcomes, clearly violating the conditions of our definition. However, given that a fair number of nudge-type BCTs fall within the latter 2 categories, we would be remiss not to discuss them. Let us examine each in greater detail.
1. Establishing/Manipulating Cues
Each day, we are bombarded with a relentless cascade of stimuli. Given our limited attentional capacity, not all pass through our attention filter to influence our behaviours.
After a long day at work, you enter your home—the lights are low and warm; the couch is soft and inviting; the smell of food is irresistible; the TV beckons with a new season of The Last of Us—the perfect set up for dinner in front of the screen. You end up forgetting about that pre-dinner HIIT workout you planned last night.
Cues work by triggering specific behaviour-outcome relationships. In this case, all the wrong cues and none of the right ones were present. Just-in-time prompts establish the right cue at the right moment to increase the likelihood of the right behaviour occurring. In our example, it can take the form of setting a 6 PM phone reminder, or pre-placing a written note or workout clothes on your couch the night before.
Importantly, for maximal effectiveness, these cues must be introduced at the moment the decision is made. A phone reminder at 2 PM, hours before you were to return home would not be very effective, nor would placing your workout clothes on the bed if you will not enter the bedroom until after dinner.
Here’s an example of a cue that misses the moment. The painted-on words clearly evoke a behaviour-outcome relationship (taking the stairs = healthier you), but is only clearly visible after you enter the stairwell, presumably already with the intention of taking the stairs. Right cue, wrong moment.
Now, the same behaviour often links to several different outcomes. Using a different example, smoking produces immediate pleasure and can serve a social identity function (Big Bill and all my brodies do it). Smoking also represents a financial cost, is detriment to long-term health, can be a source of conflict with a loved one, and can produce social disapproval in public. Not all outcomes are salient at any given moment, particularly if they are distant. Out of mind, they are unlikely to influence behaviour. Increasing outcome saliency brings attention to specific behaviour-outcome relationships to influence behaviour. To reduce smoking, we might highlight its detrimental effects on your health (or worse, on your child’s health).
A specific example of manipulating outcome saliency is the framing of information presented to an individual as a gain or a loss, leveraging on our tendency to respond more forcefully toward negative outcomes (known as loss aversion). We discussed an example of this in Part I, where a doctor has a choice between presenting a treatment in terms of its success or failure rate (90% survival rate vs. 10% mortality rate). Making success or failure rates more salient can lead to very different patient choices.
To stress again, for antecedent stimuli to be effective, their associated outcomes must be important to the individual. Establishing/manipulating a cue works best if there is already a clear relationship between the cued behaviour and a valued outcome. However, sometimes the relationship is unclear—I know using plastics is bad for the environment but I don’t know exactly how bad. Other times, there is a fundamental misunderstanding of the relationship between behaviour and outcome—I don’t know why I should buy this washer than clearly costs more. In other words, an information gap exists that needs to be bridged.
2. Providing Information
With our capacity for culture, much of what we know results from information provided by others rather than solely through direct experience. It seems somewhat disingenuous to call providing information a BCT, a nudge, or more recently, a boost. Still, sharing the right information at the right time can influence behaviour—especially when there’s a clear information gap.
I know using plastics is bad for the environment but I don’t know exactly how bad.
Most of us agree that using plastics bears an environmental cost. Most of us do not have a clear idea of that cost. A video showing a turtle with a plastic straw stuck in its nose is a powerful (not to mention, highly salient) way to get the message across. Here’s the video—hopefully it makes a small dent in your plastic straw usage.
I don’t know why I should buy this washer than clearly costs more.
Returning to an example from Part I, buying energy-efficient electronic appliances typically costs more upfront but saves you money (and reduces your carbon footprint) in the long run. Without information about the long-term costs, we are likely to base our decision solely on the upfront cost, thus opting for cheaper, less efficient appliances. Providing information in the form of energy labels containing information about the annual expected energy consumption, cost, and an energy efficiency rating can guide individuals toward making cost-effective (and climate-friendly) choices3. These labels work by making incentives salient, particularly, by aligning behaviours with long-term incentives. For me personally, they have led to thousands of dollars in estimated savings on my large appliances over the past few years.
A final note before we move on. As all parents can attest to, people sometimes do not rarely do as they are told. To influence behaviours, the provided information needs to connect behaviours to important outcomes. At the risk of oversimplification, providing information effectively essentially comes down to 2 things:
What information we communicate. If plastics in the ocean doesn’t resonate, perhaps the messaging should focus on plastics in our brains.
How we communicate the information. There is a whole art and science to this. But briefly, consider why a video of a turtle with a plastic straw in its nose is more impactful than one with statistics about the amount of plastic waste in the oceans.
However, even the clearest message often falls flat. The main culprit? Effort—it takes effort to change behaviours.
3. Manipulating Effort
Some of the greatest successes in nudge implementations come from reducing effort (also called reducing friction) to encourage behaviours4.
While not always immediately apparent, most behaviours don’t directly lead to desired outcomes—we usually take several steps to get there.
Additionally, each behaviour can vary in its difficulty of execution—the registration process can be straightforward or highly confusing; the waiting time can be 5 min or 1 hr. Therefore, to reduce effort, we can:
Reduce the number of steps required to reach the desired outcome.
Reduce the difficulty in executing one or more behaviours in the sequence.
Tuberculosis kills more than a million people annually, disproportionally impacting individuals residing in low- and middle-income countries. Treatment typically involves taking a cocktail of oral medications up to several times a day for 3–9 months. Because of its virulent nature, patients usually have to take their medications at a treatment facility under the supervision of a healthcare professional (i.e., directly observed treatment [DOT]).
As you can imagine, the constant effort required to schedule time and make a trip down to the treatment facility impacts adherence rates. Failure to adhere to treatment has important downstream implications, including the development of multidrug-resistant tuberculosis.
As an alternative, video-observed therapy (VOT) enables patients to have their medication taking observed through live videos (synchronous), or video recordings (asynchronous), greatly reducing the effort required for treatment adherence. The result? In a randomised controlled trial in the UK, more than double of participants assigned to receive asynchronous VOT successfully completed ≥80% of treatment observations in the first 2 months following enrolment compared to participants assigned to receive DOT. Owing to the success of these trials, asynchronous VOT was subsequently adopted by the National Health Services in London.
Another way to reduce effort? Simplifying choices. Have you ever scrolled through countless dinner options on your phone and found yourself unable to make a choice?
We love to have choices, but not too many. Especially if there isn’t a clearly superior outcome. Being presented with too many options of similar quality overloads decision making and induces dissatisfaction, termed the paradox of choice by the psychologist Barry Schwartz. Most of us would do a lot better in arriving at a decision (and with less stress) if someone simply offered us 3 solid dining options to choose from.
Our final application is helping individuals plan for hard decisions in advance (termed implementation intentions). As previously written, we tend to discount larger-later outcomes (e.g., long-term health) in favour of smaller-sooner outcomes (e.g., the pleasure of consuming salty, fatty food).
Queuing at McDonald’s, with the smell of cooked fats cutting through your olfactory defences, choosing a corn cup over the upsized fries seems like an insurmountable task. Planning in advance, in the absence of immediate temptation, is much easier:
When I’m at McDonald’s, I will get the corn cup instead of the fries.
That brings us to our final category—BCTs that directly manipulate outcomes to influence behaviours.
4. Manipulating Outcomes
Many nudges programme some change in outcomes, either by modifying existing outcomes, or adding/removing outcomes. Let’s examine some of them.
A commitment device adds a cost to not acting in accordance with a stated behaviour or achieving a stated goal5. The cost can be financial—in their book, T&S elaborated on a website, StickK, in which users set a goal (e.g., run 20 km a week) and pledge a sum of money toward meeting that goal. Failure to do so would result in the money being forfeited. There’s even an option to forfeit the money to an antithetical cause (e.g., the Manchester United Fan Club for a Liverpool FC fan). Beyond financial incentives, the pledger can also nominate referees to keep them accountable, introducing elements of social affirmation for meeting their goals, and disapproval otherwise.
In fact, recruiting social norms features heavily in nudge-type BCTs. One prominent example is providing households information about their energy use in comparison to their neighbours’ to reduce household energy consumption. One such effort reduced energy consumption by 2% on average in 600,000 households6, with high energy use households reducing their energy consumption the most. Interestingly, the use of social norms can sometimes backfire, signalling that it’s OK for the most energy-efficient households to be a tad less so. A quick fix for this—include an affirmation for their energy efficiency (a ☺️ or a “Great!”).
There’s also a more subtle influence at work—many nudge-type BCTs change how we interpret stimuli by pairing them with emotionally powerful imagery or ideas7. It is no accident that the warnings on cigarette packages are accompanied by extremely graphic pictures such as shown above, specifically designed to induce a direct physiological aversion.
Bringing It All Together
In discussing nudge-type BCTs, we aimed for breadth, not depth. However, it should be clear by now that the term nudge is too broad, lacking sufficient precision and clarity to be useful.
BCTs subsumed under nudge are not a homogeneous collective. Even within the same category, different BCTs influence behaviours through different mechanisms. Furthermore, the boundaries between nudges and non-nudges are unclear, and the nudge universe encompasses a broader set of BCTs than what our adopted definition allows.
Behaviour change practitioners are increasingly cognizant of these issues. As the Chief Behavioural Scientist of the Behavioural Insights Team Michael Hallsworth noted recently,
[BCTs] have gone far beyond the self-imposed limits of nudges, even if that label is still used (often unhelpfully) as a blanket term.
To use nudges effectively, at the very least, we need to go beyond
I’m going to change [target behaviour] by [insert random BCT I heard about].
to
I’m going to change [target behaviour] by [insert specific BCT] as it influences behaviour through [specific mechanism of action].
Beyond that, several key questions remain to be answered:
When and on whom should we use a nudge-type BCT?
Who should implement them?
Are they actually effective?
If not a nudge-type BCT, then what?
We’ll be back with more answers (and more questions) in Part III. Till then.
Our efforts may even backfire if they throw food waste into the recycling bins and contaminating the bin contents.
Others have also tried classifying nudges in other ways—like System 1 (automatic) vs. System 2 (deliberative) nudges, or goal nudges vs. behavioural nudges. We consider these classifications unsatisfactory for reasons that require a separate article.
Better yet, we can present the total estimated cost over the product’s life cycle (say 10 years) and remove the hassle of having to manually compute the costs (error-proofing).
On the other hand, we can also increase effort to discourage behaviours.
The astute reader will also note that these commitments (sometimes called precommitments) are really an extension of implementation intentions, with the addition of programmed outcomes.
This number, while small, is highly consequential considering the reach of the intervention.
There are actually two possible mechanisms here—through classical conditioning (pairing cigarettes with graphic images) and through bringing other meanings into the equation (e.g., smoking = bad parent).