UBhave is a project that received funding of £1.5 million from EPSRC to promote cross-disciplinary research. Our aim is to investigate the power and challenges of using mobile phones and social networking for Digital Behaviour Change Interventions (DBCIs), and to contribute to creating a scientific foundation for digitally supported behaviour change. The projects work packages are as follows:
Work Package 1: How do users interact with mobile behaviour change interventions? What features and elements promote effective engagement?
We will design, develop, and evaluate a Smartphone-based application to support weight management. This application will supplement “POWeR”, an existing web-based intervention for weight management. The application will be evaluated in a series of n of 1 studies using in depth qualitative and quantitative methods. We will collect objective data about participants’ actual use of the application in addition to subjective data about their thoughts and opinions about using the application. The first application we will be trialling will include the following features: reminders of weight management plans (e.g. goals, suitable foods, personal motivations for losing weight); selfmonitoring of diet, physical activity, and goal progress; immediate progress feedback (e.g. tailored messages, tailored visual displays); and just in time support via phone-based tools designed to support weight management.Key questions:
- Is it feasible and appropriate to deliver behaviour change interventions by mobile phone?
- What elements of mobile applications are most engaging and effective for people and why?
- Can mobile applications enhance predictors of successful weight management (e.g. attitudes, behaviours)?
Work Package 2. How can mobile phones collect data accurately about human behaviour and efficiently support real-time feedback for DBCIs?
A fundamental issue to be addressed in the development of mobile DBCIs is the optimisation of the techniques for querying information from the mobile phone sensors, processing it on the phones and uploading the resulting data in real-time to the back-end, given the limited resources (in particular phone battery); in a symmetric way, communication protocols between the remote server and the phones should also be carefully designed and optimised to save energy. Data will be collected on the phones and sent to back-end servers where they can be analysed automatically in real-time. On the basis of these inputs behavioural interventions can be delivered through participants' phones and online social networks. The platform should allow for automatic actions and interventions, such as the display of suggestions for changing user behaviour, for example triggered by context events. A key aspect of this work package is the development of inference algorithms for off-the-shelf mobile phones that contain sensors that have not been designed for this kind of sensing activity. Privacy of the data and consent of the users will be pivotal to these studies.Key questions:
- What are the fundamental design principles for developing efficient social sensing components for mobile phones?
- What is the best design for effective real-time feedback mechanisms for behaviour intervention?
- How should existing inference algorithms be adapted and improved in order to support sensing and feedback also considering the hardware characteristics of off-the shelf phones?
Work Package 3. What are the implications of delivering DBCIs through online social networks?
The aim of this work package is to determine how best to employ social-networking websites as a medium for disseminating and implementing interventions and collecting valid intervention data. We will address that aim by recruiting active users of MyPersonality, a Facebook application with several psychology tests and surveys with over 3 million active users, to participate in various behaviour change interventions (e.g. weight management, smoking cessation, self-care for minor symptoms). We will examine 1) participation rates and retention, 2) participants’ experiences of using the behaviour change platform, 3) efficacy of the interventions, 4) individual difference and usage variables that moderate retention and efficacy, and 5) interest in mobile interventions and types of mobile devices participants use. We will compare results from the proposed study with results from previous studies to assess the generalisability and validity of the intervention data.Key questions:
- Do online social networks offer a viable platform for delivering behaviour change interventions?
- How does the efficacy of behaviour change interventions delivered through established online social networks compare to those delivered through more traditional channels?
- What are the characteristics of social network users who are most likely to benefit from the behaviour interventions?
Work Package 4. How can data from real-time sensing and social networks be integrated effectively into DBCIs?
This task is concerned with the integration of online social network data gathering and mobile phone sensing and will build on WP1, WP2 and WP3. We will enhance the mobile weight loss intervention developed in WP1 with mobile phone sensing data collection and online social network data. For example, location (e.g. at home, at work or elsewhere) can be sensed using GPS, physical activity through the accelerometer, and co- location with other people through proximity to Bluetooth devices. Excerpts of conversations can be compared with a library of audio samples to identify emotions such as stress. Online information about users' personality, attitudes and preferences (measured using the ‘MyPersonality’ Facebook application) will be integrated with this information about activity, physical co-location with others, location, and emotions gathered through mobile phone sensing. Mobile phone sensing will be developed as a sub-application to be pushed to millions of users of MyPersonality to install on their devices. One of the most challenging research aspects is the development of algorithms for merging data from mobile phones and information from online social network applications. We will investigate how mobile phone and online social network tools can best be used as feedback mechanisms. A key aspect is the design and implementation of the subsystem for providing feedback to users in real-time.Key questions:
- Can online social network and mobile phone application be integrated in an effective behavior intervention system?
- Can data from this system be used to analyze and improve behavior?
Work Package 5. What new forms of statistical data analysis are needed for this type and scale of data?
The interaction with users in the other work packages will create some very large rich and new data sets. To accommodate the analyses of these data sets, we will develop tools to visualize the sequential use of the various interventions over time. This tool will provide an overview on the behaviour of users, the similarities and differences in their use of various interventions and possible causal relationships between their characteristics and the observed patterns. These factors can be used to promote positive behaviour changes.
Other than the challenge of dealing with very large data sets, we will also encounter the problem of drawing inferences from very small data samples. This problem arises when there is a small trial or low uptake on new specific interventions. These provide, for example, repeated measures on a small number of people over a short period of time with and without interventions. The second statistical tool that we will develop is to detect effects of these interventions given small samples with the use of generalized linear models. We will compare the properties, for example power, of asymptotic techniques based on large sample theory and bootstrap techniques. This tool will also provide guidance on the design of trial studies including how to balance the number of participants and time required.
The third statistical tool we plan to develop involves the analyses of temporal social network measures and spatial and social measures using large data sets. It will be developed in conjunction of our first tool. We aim to uncover various causal relationships between interventions and technological, psychological, demographic and social factors. Longitudinal data will be modelled to understand behaviour changes (both positive and negative) and to accelerate positive behaviour changes.