How can we support designers to create theory-driven Designs for Behaviour Change?
The Persuasive by Design–Model and how it came to be.
If you are interested in design and behaviour change, you may have noticed that there aren’t many tools that do justice to the complexities of human behaviour while at the same time being useable for design practice. With our Persuasive by Design-model, we aim to make it easier for designers (and other creatives) to base their products, campaigns and services on recent and relevant behavioural science theories, in a way that fits seamlessly into designerly practice. In this text, I would like to introduce the model, offer an insight into its development process and describe how we tried to make this model a useful tool for theory-driven design.
Much of our personal health and wellbeing depends on our behaviour. Behaviour-related conditions are now the largest cause of death in most countries. Most threats to our planet’s health are similarly behaviour-related because of the way we use its natural resources. It is therefore hardly a miracle that there has been a growing interest in designing for behaviour change in recent years.
When we set about creating designs that support people in living healthier and more sustainable lives, theories and evidence from the behavioural sciences can be tremendously helpful. Without knowing it, designers of products, services or communication campaings always have implicit views on how the users of their designs are going to use them. These implicit user theories more often than not see the user as either overly rational, capable of motivating behaviour change at the presentation of a single fact, or overly naive and easy to manipulate. Based on these implicit theories, designers make choices for their designs that do not deliver the results they expected: the user does not dance along to the tune the designer is playing.
Insights from the behavioural sciences could help replace these implicit user theories by explicit ones, grounded in theory and evidence. Unfortunately, for designers, these insights are difficult to come by. Behavioural scientists publish their work in journals that charge hefty fees for non-academics, and write their papers in a language that is not only hard to read for outsiders, but also makes it difficult to evaluate the relative merit of a work. When psychological theories do make it to the creative industries, this is more often than not the result of the marketing abilities of the researcher in question, and not necessarily related to the practical value of the theory.
The results of this divide between theory and practice range from a waste of time and energy, all the way up to serious damage.
Take, for instance, health-related apps. Every day, new apps aimed at healthy behaviours hit the app store. Only a very small proportion of these apps have some sort of scientific backing, be it an underlying theory or evidence of the efficacy of its working ingredients. A range of apps would like to help you quit smoking, but the easier you can find them in the app store, the smaller the chance of their actually helping you quit. Other bad examples are these apps that pretend to help you diagnose skin discolouration, but tend to miss dangerous melanomes. Or these apps that ‘help’ you deal with pain. And that’s just considering apps. A similar phenomenon can be seen in health campaigns; there are many examples of ineffective or even counter-effective campaigns that cost the public sector a lot of money.
In the past two years we have been working at building a bridge between behavioural science theory and designerly practice, in a project called Touchpoints.
The project unites behavioural scientists and design researchers from academia on the one hand, and designers of products, services and visual communication on the other. In this project, we have been searching ways to make it easier for designers to use evidence from the behavioural sciences to inform their designs, in a way that fit seamlessly into already existing design methods — one of our assumptions was that designers do not need to become fully trained behavioural scientists to build effective behaviour change solutions.
After two years, this process has resulted in the development of a model for evidence-based (or at least theory-driven) design of behaviour change interventions, and a suite of tools to enable use of the model in design practice.
We did not set out to devise a model at all. But the thing is, designers are prone to cherry picking: they tend to select those theories they like and ignore the rest, and often use evidence as an inspirational tool that can be discarded as soon as the actual designing begins. So to minimize cherry picking behaviour and to add structure to the many insights that can be found in the behavioural sciences, we needed some overarching logic. And that logic turned out to be a model.
The Persuasive by Design–Model
Upon starting the project, our first step was to try and summarize all of the insights from the behavioural sciences that we considered essential. One such central principle is the divide between reflective / controlled versus reflexive / automatic behaviour. Such dual mode theories are very widespread in psychology, be it Kahneman’s system 1 versus system 2, the central versus peripheral route, the notion of reflective versus automatic cognition, or heuristic versus systematic information processing. In reality, of course, all these dual process models are exaggerations; automatic and reflective processes in human cognition are deeply entwined. Many behaviours can be both automatic and reflective at different times (e.g. driving a car), others are somewhere in the middle of the automatic–reflective continuum (e.g. spending every holiday on the same campsite in France), and many behaviours are of such complexity that they have both automatic and controlled components.
For practical purposes however, for instance when designing a campaign to reduce food waste, automatic versus controlled is a very useful distinction to make. It allows designers to distinguish between campaign elements aimed at automatic behaviours, such as impulses or deeply engrained habits, and reflective behaviours that are available to conscious thought. (This also provides a useful way to define ‘nudging’: anything that aims at changing automatic behaviour — or (and we’ll get to that later) takes advantage of biases in our information processing.)
A second principle is the shape that these two modes have. Automatic behaviours are very simple: there is a cue, and there is an automated response, which is often (but not always) unconscious, unintentional, and uncontrollable, and always very efficient. In a diagram, an automatic behaviour looks much like this:
Reflective, controlled processing of behaviour is reminiscent of a thermostat: We compare our goals to current behavior. Upon noting a discrepancy, given enough motivation, ability and opportunity, we change our behavior. monitor our changed behavior, compare once again our current behavior to our goals and so on, until our goal is reached. This would look something like this:
This basic framework can help designers to determine what strategy would be most promising when designing a behaviour change concept. When dealing with automatic behaviour such as habits or impulsive behaviours, there are basically two strategies: hide or replace the cue, or disrupt the cue-response chain so that the behaviour becomes available for conscious scrutiny in the self-regulatory cycle. Campaigns aimed at increasing knowledge by offering information, or aimed at attitude change through emphasizing social norms, are not likely to have any effect on behaviours that are largely automatic.
Our modelling of controlled behaviour change illustrates how this is an ongoing process that takes time and effort. Motivation, opportunity and ability play a key role. When dealing with controlled, reflective behaviour change processes, there is a plethora of possible strategies. However, these should fit the specific strenghts and weaknesses of the target group and the behaviour we’re aiming to change.
And there we arrive at the third principle: behaviour change is hard work. The cue-response-chain is simple, but it is not easy to intervene in these behaviours. The self-regulatory cycle looks simple, but there are many processes and circumstances that easily lead to disengagement. We think we know our own behaviour and its consequences, but there is a large and very entertaining list of cognitive biases on Wikipedia that proves otherwise. Besides, every behaviour has more attractive alternatives, there is always something we’d rather be doing than all that healthy, sustainable stuff. Not to mention the resistance, white lies, and reduction of cognitive dissonance: they are part of every behaviour change process.
Oh, and there’s also a ‘principle zero’: ethics. In our work, we think there is no such thing as ‘Persuasion’, at least not in a sustainable form. What we aim for here is to support people in self-persuasion, i.e. changing their own behaviour.
A short history of scientific model evolution: a case study in six iterations
So, after examining a large section of the relevant scientific literature, we had gathered the basic principles we thought necessary for effective theory-driven design for behaviour change. Our next step was to hold a series of co-design sessions with service designers, to see how we could integrate our insights into design methods.
I studied visual design at the Arnhem College of Arts, and social psychology at the University of Nijmegen. As a designer, I tend to draw and sketch to maintain an overview of what I am doing. As a social scientist, that means I try to draw crude models and graphs of what I am trying to express. At the first co-creation session, I did the same with these basic principles. The designers in the room, all experts of some level in service design, product design and the like, were very enthusiastic about the drawing, saying this was something they had immediate use for in their work. And so the Persuasive by Design–Model was born.
A good thing about having a model is that it provides an
intermediate layer between theories and insights, and design tools.
We were not planning on devising new design methods because there are already so many out there. Many design firms have based their unique selling points on design methods they developed themselves. Of course they are unwilling to discard them and adopt ours at a whim. Another problem with developing a design method are the vast differences in expertise among our audience. When you are building a method that is usable for novices, it is often way too restrictive for experts (and a bit too restrictive for intermediates), and the other way around.
The very first version of the model looked like this (click here for a pdf):
Note: although we did make the distinction between automatic and controlled behaviours, I did not yet incorporate it in my drawing. What this version does contain is an overview of intervention strategies. At the time, I assumed the best results could be expected from implementing three (related) groups of strategies for controlled behaviour: communicating (normative) information about the behaviour, providing feedback on current behaviour and using some form of goal setting or action planning to encourage new behaviour. This set of strategies would cover the entire self-regulatory cycle. This may seem restrictive, because there are so many other strategies out there, but there is indeed evidence that this combination of strategies often proves fruitful.
One design cycle later, the model looked like this (pdf here):
Note the way I try to be nuanced about the controlled / automatic — distinction. Also note how all the pitfalls on the way to change have been integrated. This would enable designers to consider these influences, because if you don’t, we thought, you are naive and your interventions will surely fail.
Guess who else was naive? Well, as it turned out it was me. Or so I found out in our next co-design-session. This time we were joined by a group of design researchers and they were unanimous in their critique: it’s all social, stupid!
Behaviour takes place in a social practice.
Now which box of that model of yours has that?
There are two levels to this critique. One level directly reflects upon the model: our behaviour is not performed in splendid isolation, but all of our behaviours have a social component. No behaviour change is possible without taking social aspects in consideration. Peer pressure, social norms, social comparison, cooperation, social support: every step of the self-regulatory cycle is influenced by our social environment. These influences needed a place in our model so I came up with another layer. Accordingly, the next version of the model had a red ring with ‘internal’ influences, and a green ring with ‘external’ social influences on behaviour change (pdf here).
In a deeper level, the design researchers’ critique was more profound: behaviour takes place in a social practice. This means that when aiming for behaviour change, it is never enough to concentrate on the target behaviour. We must always consider the system in which the behaviour is performed. Take for instance whistleblowing behaviour. Management at a construction site might want to persuade workers to report possibly unsafe situations. However, reporting a potential hazard often means delays, extra costs, and overtime. Management is trying to push a behaviour even though the system the whistleblowing is performed in will respond with annoyance or even punishment.
So whatever incentive the managers are going to try, their interventions are bound to fail because they do not take the social practice into account.
This is a fundamental problem of looking at an undesired situation through a behavioural lense. There is no way to encorporate this into the model. The system in which a behaviour is performed needs to be something we’re aware of when we use this model.
Around this time I also noticed a disconnect between the importance of the theoretical divide between automatic and controlled behaviours, and the way this was represented in the model. So I fixed that, making this divide more visible and adding more potential interventions in automatic behaviours (pdf here).
As you can see, things are getting seriously out of hand here. What started out as a relatively simple drawing is getting more and more complex. Is this model even usable for designers? Time to take it to the test. We did that in two series of sessions: the first series were aimed at improving existing concepts for promoting alternatives to car travel during rush hours; the second series were aimed at concept creation using the model.
The sessions consisted of an introduction to the model which took an hour and a half, and a practical exercise in which the participants used a question set to access the model and use its insights to inform their concepts. Participants to all sessions found the model useful and helpful, but also very hard to use, even with the question sets. Without the introduction, they thought, the model would be too complicated to be of any use.
We presented this first official version of the model and the tests we did to assess its efficacy at the CHI Sparks conference in The Hague in april 2014. You can read the conference paper here. A PDF version of this iteration is here.
In this version, we started to simplify things a bit. The part concerning automatic behaviour moved to the bottom, which has a distinct advantage: it makes the model easier to read. Unfortunately, there is also a clear disadvantage: the automatic behaviour insights are easier to overlook.
Our next step was to use the model to inform the design of complex behaviour change interventions. To do so, we used the model in a large project aimed at safety motivation at work. We also monitored the use of the model in another large project on streamlining grant application services, commissioned by the Dutch government.
Using the model had clear benefits in both case studies: it had an impact on the designs (especially in drawing the attention of the designers to automatic components of the target behaviours), it gave structure to the design process and made it easier to take insights from informative phases
of the design process into later stages.
Some issues also transpired. The model was use predominantly in the early phases of the design process and a lot less in later phases, and once again we noticed how zooming in on individual behaviour makes it harder to consider systemic factors influencing those behaviours. And again: the model by itself is too hard to use without lots of explanation.
We presented our study of the use of the model in the two case studies at the 11th conference of the European Academy of Design in Paris on april 23rd. You can read the conference paper here.
In all our tests of the model, it became apparent that the model itself is too complicated to be useful. Two possible solutions present themselves: simplifying the model, and developing tools to use it. In a final series of co-design sessions, we set out to develop a suite of tools to use the model in different stages of the design process, and to incorporate it in existing design methods. In a constant dialogue about what the model tries to convey, we managed to seriously simplify the model without losing too much of its informational value. We used different colours that could be used in the development of tools, so different tools could refer to different parts of the model. The wording used in the model is also more accessible: labels such as ‘knowing and feeling’, ‘seeing and realising’, ‘wanting and being able to’ makes it easier to comprehend.
This redesign (pdf here) also allowed us to emphasize what we think is important in the model. For instance, the box ‘wanting and being able to’, abouth ability, motivation, and opportunity, finally gets the attention value it deserves. On the other hand, in previous versions, comparing goals with behaviours got more attention than warranted. Of course this is an important aspect, but definitely not more important than motivational aspects of behaviour change or the possibility to try out new behaviours.
Taking some of the complexity out neccessarily means losing some information.
The –internal– red and –external– green cycles, the (in some circumstances) accellerators or (in other situations) brake pads of the reflective cycle, have moved to the background in this version. The downside of this is that it makes these insights less easy to incorporate because they are less visible.
There is a tension between usability and exhaustiveness.
Furthermore, the previous version suffered from a bizarre combination of being overly complex AND oversimplifying at the same time. In previous versions, it appeared that peer pressure does not have anything to do with motivation: its arrow points at the ‘discrepancy’ stage of the reflective cycle. In reality, that is just plain wrong.
I also removed the different intervention strategies from the model. The more we used the model to inform behaviour change interventions, the more we realised our focus on norm communication plus feedback plus action planning was not feasible. One solution would be to include more intervention strategies. But this would mean introducing up to 89 (according to some taxonomies) new boxes with accompanying arrows, which would make the model look ridiculous and render it utterly unusable. A more viable way was to remove the intervention strategies from the diagram altogether and try to include them in an accompanying tool, an Atlas of Strategies, so to speak.
The translation of this model into usable tools that fit into designerly practice is a complicated process. I tried building some tools myself, but they were complex affairs with too much text, boxes and arrows. Fortunately, the project partners from practice rose to the occasion and developed a range of tools that they could use within their own design methods. Some of these tools, such as a set of ‘behavioural lenses’, closely follow the model’s logic. Other tools, such as Marie-Louise, a tool to establish target behaviours, and a tool for building a behavioural journey, a customer journey based on behavioural insights, are more loosely connected to the model. And we have used parts of our attempts at an atlas of strategies to extend the second edition of our book ‘Ontwerpen voor Gedragsverandering’ / ‘Designing for Behavioural Change’ — available in Dutch and English as part of our Behavioural Lenses Toolkit.
Meanwhile, the model has arrived at a stable form. The Persuasive by Design-model remains a living thing, subject to change and evolution.
But right now, we think it has reached a stage of maturity in which it can have significant value in the design process.
The constant balancing between exhaustiveness and usability means we have had to make some limiting choices. Choices entail that stuff gets left out. These omissions are the model’s most obvious weak points, that we would love to deal with in future versions. The deletion of the green (social influences) and red (threats to the self-regulatory cycle) layers means that especially the social influences no longer receive the attention they deserve. Also, as mentioned before, the visual separation of automatic and controlled modes of behaviour does not do merit to reality: ‘system one’ never sleeps, and almost all behaviours have automatic and controlled components.
And of course, there is the issue of validity. This model is a great tool to provide a scaffolding to use insights from behavioural science to inform the designs of products, services, and communication campaigns. But its scientific validity is as yet unclear. So far, we have not had many chances to exchange views with behaviour change scholars (other than ourselves) on the value of this model for the accumulation of scientific knowledge.
All tools mentioned in this text are available from June 4th, 2015 from this website. For all tools, English translations are available. Our book on designing for behavioural change is also included in the toolkit. To purchase a copy of the toolkit, or to inform about tools not included in the basic kit, please send an e-mail to email@example.com to find out about pricing (all products are sold at cost), delivery, etcetera. (NB: apparently, the project website is undergoing some changes and it’s not always easy to find the tools. If you’re interested in them and can’t find them on the website, just send an e-mail to firstname.lastname@example.org).
Two more publications featuring the Persuasive by Design Model:
Hermsen, S., Van der Lugt, R., Mulder, S., & Renes, R.J. (2016). How I learned to appreciate our tame social scientist: experiences in integrating design research and the behavioural sciences. in: P. Lloyd & E. Bohemia, eds., Proceedings of DRS2016: Design + Research + Society — Future-Focused Thinking, Volume 4, pp 1375–1389.
Van Essen, E., Hermsen, S., & Renes, R.J. (2016). Developing a theory-driven method to design for behaviour change: two case studies. in: P. Lloyd & E. Bohemia, eds., Proceedings of DRS2016: Design + Research + Society — Future-Focused Thinking, Volume 4, pp 1323–1338.
Are you interested in using the Persuasive by Design–Model in your projects? Please let us know how you used the model and how it helped you (or did not help you). Your findings, results, opinions, and especially the problems you might encounter, help us to further develop the model and its accompanying suite of tools.