Using Data Mining to Refine Digital Behaviour Change Interventions

Abstract

Do Something Different (DSD) behaviour change interventions are digitally delivered programmes designed to help people improve their health and wellbeing by adopting healthier habits. In addition to content addressing specific issues, such as diet, smoking and stress reduction, DSD interventions contain a core component promoting behavioural flexibility. This component helps people practice behaving in ways they currently do not, such as assertively, proactively or spontaneously, and is based on a model developed by psychologists researching the connections between behavioural flexibility and wellbeing. This paper describes how we have used data mining techniques to optimise the design of DSD interventions, in particular the behavioural flexibility component. We present correlation networks and regression models obtained using pre- and post-intervention questionnaire data from 15,550 people who have participated in a DSD intervention delivered by email, SMS or smartphone app. We explain how these results led us to a clearer understanding of the connections between behaviour and wellbeing, using which we have optimised DSD interventions, ensuring that participants concentrate on developing the behaviours that are likely to benefit them the most. Additionally we have used logistic regression to fit a propensity score model, which models how likely it is that each person in the dataset will complete the post-intervention questionnaire, based on their pre-intervention questionnaire data. When we stratify our dataset using these propensity scores, we find that the kind of people who are the least likely to tell us they have completed the intervention, by answering the post-intervention questionnaire, are also the kind of people who will experience the biggest increase in wellbeing from a completed programme.