X Close

CBC Digi-Hub Blog

Home

Menu

Integrating Behaviour Change Theories with the Adaptive Decision-Making Framework

By CBC Digi-Hub Blog, on 25 May 2021

Written by Dr Chao Zhang (Department of Psychology, Utrecht University)

For someone new to behaviour change research, the large number of behaviour change theories is both a blessing and a curse. Yes, you have many options in terms of selecting a theory as the basis of your intervention, but you may also get lost in the ocean of theories. When I started my PhD project, I certainly struggled to decide which theory to use as the backbone of my thesis. This was partly because of my working environment. In the Human-Technology Interaction group at a technical university, my supervisors generally had a very pragmatic approach and were not committed to any particular theory, so I had all the freedom to explore. It was also due to my personality and intellectual style. I am not a person who easily commits myself to a particular idea, but prefer to have an overview on things and to connect pieces of ideas together. So the adaptive decision-making framework started as a literature review of behaviour change theories in my first year as a PhD student. With the encouragements from fellow PhDs and senior researchers, who found my review to be really useful, it is now a published paper in the Journal of Medical Internet Research.

Why a new framework for behaviour change theory integration?

While I was researching the literature, my urge to connect theories was accompanied by a growing discontent with many popular theories in the field. I was influenced by several articles criticising traditional behaviour change theories, but no one inspired me more than the 2011 article by William Riley and colleagues. They almost bluntly asked the question in their title: “…are our theories up to the task [of informing digital interventions]?.

One criticism they raised was the mismatch between the temporal granularities of behaviour representations used in traditional theories and digital interventions. Many of the popular theories used today were invented in the 70s or 80s, when there were no technologies available for observing or altering behaviours as they happen in people’s daily lives. For example, the famous “stage model” describes behaviour change as going through 5 stages that usually last for weeks or months, so interventions based on this theory are tailored to these coarse stages. In contrast, an e-health app on your phone can adapt its intervention strategy in almost real-time, as long as there is a need for an adaptation, e.g., to match the momentary context of a user or to tailor intervention messages to the user’s personal progress on a specific day. To inform the design of such apps, we need theoretical frameworks that match to the temporal granularity of lifestyle behaviours and digital technologies.

Another problem with traditional theories is that many of them are variable theories that describe statistical relationships between variables rather than process theories that explain the underlying mechanisms or cognitive processes of behaviour change. Variable theories are often represented visually by nodes of behaviour and behavioural determinants and the links between them that denote statistical or causal relationships. For example, based on the links connecting attitude, intention and behaviour in the Theory of Planned Behaviour, if you know the attitude of a person towards physical exercise at a certain time, you can most likely predict their exercise behaviours in the following period with some level of certainty. However, besides its predictive value, the theory does not tell you much about how to change attitude or intention in order to change behaviour, or what the cognitive processes behind the statistical relationships are. Since I was trained partly in applied cognitive psychology, I was aware of many other psychological theories that focus on processes and mechanisms, such as habit formation, reinforcement learning and decision-making. Models from these areas have certainly been applied to behaviour change interventions, but they are still not at the central stage of the field. Therefore, I was motivated to integrate such process theories into a new framework.

What is the “adaptive decision-making” framework?

When I say “behaviour change” in this blog, I almost exclusively refer to lifestyle behaviour change, not just any behaviour change. This distinction is crucial for understanding the “adaptive decision-making” framework in our paper. Some interventions may target single-time health decisions, such as a decision to receive a vaccination or not. These decisions are very consequential and usually require a lot of thinking from the decision-makers. But changing the behaviour requires nudging people to choose differently only once. However, lifestyle behaviours are about routines, habits and how people behave repeatedly, e.g. on a daily or hourly basis. For example, eating one meal of high-fat food won’t do any harm, but continuously following a high-fat diet increases one’s risk of obesity. Accordingly, lifestyle interventions should target the repeated meal choices in a person’s diet, not just a single meal choice. The characteristics of lifestyle behaviours naturally determined how behaviour change processes are represented in our framework. Instead of “stages” or a diagram of interlinked variables, behaviour change is represented as a series of repeated and interrelated decisions (see Figure 1).

Figure 1. A two-level representation of lifestyle behaviour (change). Reproduced with permission from Zhang et al. (2021).

 

To give an example of what repeated decision-making means, let’s suppose you just started a new job and you eat your lunch everyday at the canteen of your organisation. Each day, you are exposed to different lunch options and you make your choice based on several considerations, including taste, nutritional aspects, and price. Over time, you explore different foods at the canteen by trying them out, and eventually you end up with alternating between a few of your favourite options based on your personal goals at the time. Your lunch behaviour remains habitual for a while, until you change your personal goal(s). For instance, you may become aware of your health risks after a medical test, and after some deliberation you decide to follow a low-fat diet. You take this goal into consideration when making your lunch choices and gradually you may change your lunch habit.

By considering daily decisions as the building blocks of behaviour change, the temporal granularity of the theoretical framework is matched to that of digital interventions. Even if a person tends to behave in a certain way for a certain time period, the framework allows individual decisions to vary, and at each time a decision is made, it may be affected by a different intervention (or no intervention). This representation of how lifestyle behaviours unfold over time also makes it clear what are the different explananda in a behaviour change process. First, we need to explain how individual daily decisions are made and what factors influence them. These decisions do not occur in isolation, so we also need to explain how a decision and its associated outcomes influence a person’s subsequent decisions. Finally, the adaptive decision-making framework distinguishes reflection-level processes from action-level processes. In addition to the individual daily decisions, occasionally there are also reflective moments when people set up new goals or reconsider their goals based on self-monitoring of daily behaviours. We also need to know how these reflection-level processes work. The two-level representation allowed me to integrate a wide range of theoretical ideas in psychology into a single framework, including decision-making, self-control, reinforcement learning, habit, implementation intentions, and goal-setting (see Figure 2). You can read more about how the framework was developed and its component parts in the published article.

Figure 2. A full representation of the adaptive decision-making framework. Reproduced with permission from Zhang et al. (2021).

 

How can the framework be used?

When I was struggling with revising the paper, my supervisor encouraged me by saying that many researchers would thank me for the paper because I did an important and difficult job for them. Indeed, I think the framework can help many researchers, especially those who just step into behaviour change research, to navigate through the literature more efficiently. Before discussing the framework, our paper also includes a review of important individual theories of behaviour change. There are of course other excellent introductory materials that give a more comprehensive overview, such as the ABC of Behaviour Change Theories, but our paper can be read as a supplement to facilitate a deeper understanding of the connections among the individual theories and one’s own ideas for theory integration.

Several other practical uses of the framework are discussed in the paper, including, for example, the identification of intervention techniques based on cognitive constructs and processes in the framework or combining multiple intervention techniques. Here I want to emphasise one specific use, i.e., the development of computational models of behaviour change based on the framework. For my own PhD work, the framework is the starting point for building computational models for the specific processes in the framework. Many people have argued for the advantages of computational models over verbal theories in behaviour change research (e.g., see this article by Spruijt-Metz and colleagues). A particular practical advantage is that computational models are essentially computer programs, which means they can be readily implemented in digital intervention systems for adaptive interventions. This is the core idea behind the psychological computing approach to digital lifestyle interventions proposed in my PhD thesis. For interested readers, this preprint describes our recent work on modelling habit formation and use of the model for better behaviour prediction in behaviour change trials. I would be very grateful if our framework inspires more research in this direction.

Notes

  1. I use “I” as the pronoun most of the time in this blog post because I want to tell my personal story behind the paper. However, the paper would not be possible without the contribution and support of my co-authors and amazing supervisors – dr. Daniel Lakens and prof. dr. Wijnand A. IJsselsteijn.
  2. I also want to express my appreciation for the PRIME theory by Robert West and colleagues. The PRIME theory is the closest to our work among all theoretical frameworks in the literature. Even though I worked out most of the framework before discovering the PRIME theory, I was certainly assured after reading it that I was going in the right direction.

Biography

Chao Zhang is currently a post-doc researcher in the Department of Psychology at Utrecht University and he is a coordinator of the HUMAN-AI alliance program. He obtained his PhD from the Human-Technology Interaction Group at Eindhoven University of Technology. He has broad interests in topics such as habit formation, behaviour change, cognitive modelling and human-centered artificial intelligence. He is keen on applying theory-based computational models to digital behaviour change interventions.

Email: chao.zhang87@gmail.com

Webpage: https://www.uu.nl/medewerkers/CZhang3

Twitter: @forzazhang

Leave a Reply