New frontiers: The growth of bespoke nudging

There are growing new findings that our behavioural tendencies can significantly differ depending on a number of individual factors. In the last few years it’s become more evident that people’s decisions and behaviours, and the degree to which they experience cognitive bias, can vary according to their genetics, age, experience and more.

Behavioural economist Koen Smets reminds us that cognitive biases are only “broad tendencies, which are not uniformly shared by everyone. We are not all equally likely to respond to, say, social proof: some of us tend to be more conformist; others are more the rebellious kind.”^[ 1 ]

Over the years, researchers have sometimes over-generalised, assuming findings about the behavioural traits and cognitive biases from one small population can be extrapolated to others. Often they can be, but not always.  Much of the research from behavioural science and psychology, but also neuroscience and other disciplines, has largely been conducted on a narrow slice of humanity – often labelled ‘WWEIRD’ (White, western, educated, and from industrialised, rich, and democratic countries). This was an outcome of a rather different research world to the one we know today, a pre-internet age where it was much harder to gain research access to diverse populations. The most easily available sample of people available to researchers in the 1970s, 80s and 90s, and even today has tended to be the psychology undergraduate.

Analysis of top journals in six sub-disciplines of psychology between 2003-2007 found that 68% of study participants came from the US and 96% were from Western industrialised countries. This means that 96% of participant samples were drawn from just 12% of the world’s population. Additionally, most research is carried out on white participants and there are few studies carried out with Hispanics, despite them making up 18% of the US population.^[ 2 ] The behavioural traits of an Indian construction worker or a Japanese business woman might be somewhat different to a white American psychology undergraduate. As researchers Joseph Heinrich and colleagues emphasise, the existing participant base “does not reflect the full breadth of human diversity”.^[ 3 ] In fact, his research even suggests that some western behaviours fall at the more extreme end of the spectrum.[ 4 ]

This opens up a huge opportunity for researchers to explore a wider pool of humanity and better understand individual variations.

Often, a simple, general nudge for a single population can be beneficial. ‘One-size-fits-all nudging’ or ‘blanket’ nudges can generate a valuable change in the behaviour of a significant proportion of people while also being economical and quick to implement. Some behavioural interventions also require scale to achieve effectiveness – e.g. public promises or social norm interventions. However, depending on the behavioural challenge, it might be the case that – where possible – more targeted nudges, tailored to specific types of individuals, cultures or contexts, may have even more impact.

Furthermore, it’s important to consider the equity implications of ‘one-size-fits-all’ interventions that aim to target the average person. Rehka Balu at the Center for Applied Behavioral Science (CABS) says: “A behavioural intervention that reduces hassle factors on average could be experienced by some groups as an increased burden. Nudges […] might prioritise individuals who already have relatively higher resources and need only a small push to get them to the finish line. If we don’t design for the needs of specific races, ethnicities, social classes, genders, and sexual orientations, are we increasing inequality in outcomes? … At an individual level, this may involve more differentiated and personalised interventions. ”[ 5 ] So, there may be situations where there is a moral and ethical responsibility to develop tailored nudges to target those in most need of behavioural change.

Some of the most exciting and valuable research today is trying to better understand variation across people and societies. This is changing the way behavioural science research is conducted, crossing a critical threshold and moving away from small-scale lab studies with WWEIRD students, to conducting research with large samples of participants from broader sections of society across a wider number of countries. We now see a growing number of large-scale field trials, tracking actual behaviours.

For example, trials of PREDICT – a tool to help patients understand treatment options for breast cancer – found that there is no ‘best’ way of visualising the outcomes of each treatment option to patients. Individual preferences and comprehension varies greatly based on a person’s numerical, verbal, or graphical skills and understanding. One size definitely does not fit all. As a result, the research team developed five different displays for communicating the data, improving each one over time based on patient feedback.

Technology and data science have also helped to enable this shift meaning research and interventions can be run at a nationwide or even global level, providing a large and more representative sample at relatively low cost. This is definite progress and in the coming years we will see even more of these types of studies.

Behavioural science is exploring the following key variables that might lead to individual differences: 

• Genetics
• Skills and long-term contextual influences
• Culture and societal differences
• Socioeconomic factors
• Age

Genetics: Our genes are now believed to influence a significant proportion of our behaviour and choices. Here we focus on one cognitive bias – present bias – to illustrate the amount of individual variation. People certainly have varying abilities to resist the ‘temptation of the now’ for long term rewards. We all know the person who can’t resist chocolate and the other person who, 12 months on from Christmas, hasn’t even opened one selection box.

This is borne out in genetic studies on the heritability of present bias and impulsivity. In a study of over 1,500 identical and fraternal twins and their parents using Sweden’s infamous Twin Registry, researchers Henrik Cronqvist and Stephan Siegel found that savings behaviour was more correlated among identical twins than fraternal and could explain around 35% of savings behaviour.^[ 6 ] This effect is present over a lifetime and does not lessen as people age.^[ 7 ] Other research has found that genetics can explain as much as 45-50% of impulsive behaviour.[ 8 ]

Skills and long-term contextual differences: Behavioural scientists are also finding that not everyone is affected by cognitive bias to the same degree due to learned skills, not necessarily solely genetic.

One example of individual differences is found in the degree to which framing – how information is presented – affects individuals. Framing often involves presenting the upside or downside of a choice in numerical terms, for example a 90% chance of mortality versus a 10% chance of survival. A considerable body of research has found that those who are more numerate are less influenced by framing effects like this. This may be because less numerate individuals focus less on numerical information which they struggle to understand or find off-putting and instead are more affected by other information such as positive and negative words that elicit the bias, other more emotional information, perhaps from a linked image, or even an individual’s mood at the time.^[ 9 ] It’s important to note that numeracy is not perfectly correlated with high levels of education or even intelligence. Yes, proficiency with numbers may initially be an individual strength, but it’s also one that needs to be built on and developed to gain high levels of numeracy. So, we can consider numeracy a skill. With the wide variation in numeracy levels in many countries – the impact of framing effects is also likely to vary considerably among individuals.

Culture and societal differences: A significant question among behavioural scientists is how generalised many of the identified cognitive biases are across cultures. For instance, descriptive social norms – the understanding that we tend to want to conform to what others are already doing – are a frequently applied insight and tool among behavioural scientists. Yet research now suggests that the extent to which people tend to conform varies considerably across different societies and cultures.

For example, a review of studies across 17 countries found that motivations to conform were weakest in Westernised societies compared to other countries, particularly those in East Asia.^[ 10 ] Other research has consistently found that on average, Americans tend to be one of the most individualistic societies in the world meaning they are less likely to conform. Significantly, behavioural interventions applying social norms to change behaviour in some way or other have also had varying success across a range of locations, from Poland to Guatemala to the US.

We need to be asking questions about other cognitive biases, too. To what degree does cultural or social background affect our propensity to experience loss aversion, affect bias, or present bias? It could be worthwhile exploring to what degree your target group experience that bias and if it might cause them to change their behaviour.

Socioeconomic factors: where and how we grow up or the characteristics of our adult lifestyle may also affect our behaviour to a significant degree. Neuroscientists have found that chronic stress can affect our ability to make good decisions. There are multiple implications on cognitive ability from these effects, such as greater likelihood of impulsive decisions and present bias – when we weigh gains in the present more than we do gains in the future. We simply don’t want to wait for a reward.

Scientists have found evidence of this impact not only in adults, but also in children. As their brain is still developing, chronic stress can have long term, permanent impacts. It tends to be found among children from low income households where stressful or chaotic lifestyles tend to be more common. Research has found that children who grew up experiencing high levels of stress – such as poverty and abuse – not only show poor development of the pre-frontal cortex as a child, but also as an adult, meaning they are more likely to be affected by present bias. As biologist Robert Sapolsky says: “Poverty makes the future a less relevant place.”^[ 11 ]

This finding has wider implications for research. Much of the previous research on cognitive bias and neuroscience has tended to be conducted on higher income, well-educated groups. The same goes for childhood studies, which have also tended to be conducted on children living nearby a university or even attending the university kindergarten. Researchers have realised that there is a need to better understand the typical behaviours and neural characteristics of lower socioeconomic groups and widen our understanding of how the context we are brought up in can determine long term behavioural traits.

We are not all the same, and recognising the influence of factors like how our background and experience has shaped us is valuable to understanding our behaviour today.

Age: We often like to believe our traits and characteristics are unchanging and stable. Today, it’s hard to imagine that in 20 years we might have different preferences and tendencies to those currently defining us. However, scientists are beginning to understand how our tendency to be affected by different cognitive biases changes throughout our life, from childhood to old age.

Not long after we reach adulthood, some functions in our brain related to what’s known as our fluid intelligence – our reasoning ability and ability to engage in logical problem solving – begin a slow decline. For the most part, these declines are offset by our ever-increasing knowledge and experience of the world – known as our crystallised intelligence. Neuroscience research suggests that our brains continue to adapt and change until their forties, and even then, the brain is always plastic – highly adaptable to our surroundings and current context. By our 60s and 70s though, an overall decline in cognitive function is unavoidable which feeds into several other changes in our thinking and decision-making.

This means that the degree to which we are affected by cognitive bias may change throughout our lifetime. For instance, teenagers tend to be more motivated by peer influences than older adults and are likely to be driven by social rewards.

At the other end of the spectrum, older people tend to be more drawn towards positive emotional experiences, and positive information as opposed to negative. Younger people – even children and infants – are more drawn to negative information and stimuli. These findings have clear implications for behavioural science – gain frames and communications with positive affect are likely to be more appealing to older people than loss frames and scare tactics.

The future

In the coming years, more research to understand various group-based and individual variations in behavioural tendencies will be welcomed. We have already learnt a lot, but more will undoubtedly be useful, particularly for designing more targeted behaviour change interventions.

Technology and data science are already helping practitioners to do the latter more effectively, even if the underlying mechanisms explaining the variation in people’s behaviour are not yet always fully understood or explainable. In particular, machine learning is enabling us to identify patterns and clusters in data, typically huge sets of data, crucially with no prior knowledge of what those patterns – such as behaviours or similar characteristics – might be. These are often patterns the human eye isn’t able to spot; and will enable practitioners to more specifically target behaviour change interventions according to the unique segments, clusters and patterns of identified behaviour.

For example, a recent trial to reduce fraudulent claims of unemployment benefit illustrates how machine learning can help to more accurately target behavioural change interventions to only the relevant sub-set of individuals. Behavioural scientists from Harvard University together with data scientists from Deloitte worked with the New Mexico state government to try to reduce numbers of dishonest unemployment claims. The state pays $2bn in overpayments so it is a significant problem. Once registered as unemployed, claimants are required to confirm their status each week, answering the question “Did you work during the reporting period listed above?”

By using machine learning to analyse the data of over 51,000 claimants, they were able to predict with 77% accuracy which individuals were likely to commit fraud. The algorithm drew on 300 data points about an individual – not only historical and habitual data but also real-time contextual data which individuals were revealing on the claimant platform. For example, a significant predictor of a fraudulent claim was if an individual changed the time of day or week that they usually made their weekly claim. This enabled the team to target only those most likely to commit fraud with nudge messages to deter and ultimately reduce fraudulent behaviour.^[ 12 ]

In conclusion

Better understanding of subgroup and individual variations in behavioural traits is certainly one of the next missions of behavioural science. It could also be the golden ticket to designing more tailored behavioural change interventions, arming us with the ability to design nudges that range from widespread to more targeted, depending on the behavioural challenge at hand. The future for applied behavioural science lies in having a much greater ability to assess the behavioural landscape and choose at what level we should customise interventions, ultimately designing more effective behavioural nudges. 

Obtaining a better understanding of subgroup and individual variations in behavioural traits is undoubtably one of the next missions within the behavioural science field, made possible through more representative and diverse research. As a result, the future for applied behavioural science lies in identifying the appropriate level of customisation and rise of ‘bespoke’ interventions, ultimately achieving more effective behaviour change.

By The Behavioural Architects’ Crawford Hollingworth and Liz Barker

Reference:

^[ 1 ] http://behavioralscientist.org/there-is-more-to-behavioral-science-than-biases-and-fallacies/

^[ 2 ] http://www.pewresearch.org/fact-tank/2017/09/18/how-the-u-s-hispanic-population-is-changing/

^[ 3 ] Henrich, J., Heine, S., & Norenzayan, A. ( 2010 ). The weirdest people in the world? Behavioral and Brain Sciences, 33( 2-3 ), 61-83. doi:10.1017/S0140525X0999152X

[ 4 ] https://www.youtube.com/watch?v=V5RxKitXHyc

[ 5 ] https://behavioralscientist.org/imagining-the-next-decade-future-of-behavioral-science/

^[ 6 ] Cronqvist, H., and Siegel, S. “The Origins of Saving Behavior” 2011, Working Paper Claremont McKenna College

^[ 7 ] Moffitt et al “A gradient of childhood self-control predicts health, wealth, and public safety” PNAS 2010

[ 8 ] Bevilacqua, Laura, and David Goldman. “Genetics of impulsive behaviour.” Philosophical transactions of the Royal Society of London. Series B, Biological sciences vol. 368,1615 20120380. 25 Feb. 2013, doi:10.1098/rstb.2012.0380

^[ 9 ] Peters, E., Västfjäll, D., Slovic, P., Mertz, C. K., Mazzocco, K., & Dickert, S. ( 2006 ). Numeracy and Decision Making. Psychological Science, 17( 5 ), 407-413. https://doi.org/10.1111/j.1467-9280.2006.01720.x; and Gamliel, E. and Kreiner, H. ( 2017 ), Outcome proportions, numeracy, and attribute?framing bias. Aust J Psychol, 69: 283-292. doi:10.1111/ajpy.12151

^[ 10 ] Bond and Smith, 1996

^[ 11 ] Sapolsky, R. “The Health-Wealth Gap” Scientific American, November 2018

^[ 12 ] O.  P. Hauser,  M.  Greene,  K.  DeCelles,  M.  I.  Norton,  F.  Gino.  ‘Minority  Report:  A  Modern Perspective on Reducing Unethical Behavior in Organizations’, Working paper 2017; Award winning paper, currently under review for publication.