A Data Science approach to behavioural change: large scale interventions on physical activity and weight loss

This PhD thesis is a quantitative investigation combining Behaviour Change Science with a Data Science approach in search of more effective large scale, multi-component behavioural interventions for health and well-being. There is limited evidence about how technology-based interventions (including those using wearable physical activity monitors and apps) are efficacious for increasing physical activity and nutrition. The relevance of this research is the systematic approach to overcome previous studies’ limitations in method and measurement: restricted research about multi-component interventions, limited analysis about the impact of social networking, the inclusion of components without sufficient evidence about the components’ effectiveness, the absence of a control group(s), small sample sizes, subjective physical activity reporting, among other limitations. The research was done in conjunction with Tictrac Ltd as the industrial partner, and the UCL Centre for Behaviour Change. Tictrac Ltd builds platforms for the collection and aggregation of personal data generated by the users’ devices and mobile apps. The collaboration with the UCL Centre for Behaviour Change has been instrumental to design, implement, evaluate and analyse behaviour change interventions that impact wellbeing and health. The thesis comprises three areas of research: 1. Computational platforms for large scale behavioural interventions. To support this research, computational platforms were designed, built, deployed and used for randomised behavioural interventions with control groups. The interventions were implemented as experiments related to the behavioural impact on physical activity, weight loss and change in diet. / 2. Behaviour change experiments. The two experiments use the Behaviour Change Wheel framework for behaviour change, intervention design and evaluation. A Data Science approach was used to test hypotheses, determine and quantify the effect of the fundamental intervention components and their interactions. The effective use of tracking devices and apps was determined by comparing the results of ‘structured intervention’ –vs- those of the control group. / Experiment 1: Large scale intervention in a corporate wellness setting. Multi-component behavioural intervention with: control group, self-defined goals, choice architecture and personal dashboards for physical activity and weight loss. The analysis covers network effects of social interactions, the role of being explicit about a type of goal, the impact of making part of team, among other relevant outcomes. / Experiment 2: Identification of critical factors of a technology-based intervention. Multi-component behavioural intervention with simultaneous target behaviours related to weight loss and physical activity, inspired by factorial design for the determination of critical factors and effective components. The analysis comprises: components’ interactions (coach, challenge, team, action plans, forum), non-linear relationships (BMI, change in diet habit), five personality traits, among other relevant results. / 3. Frameworks for future large scale interventions in behaviour change. The implementation of both experiments required an applied use of theoretical and practical principles for the design of the experimental computational platforms. As a result, two frameworks were suggested for future interventions: an implementation framework and a data strategy framework