Millions of people have already collected weeks, months and even years of data about their own health and physical activity levels. The potential is enormous for use in personal applications as well as for public health analysis of large populations at low cost. However, the reality is many people fail to wear their tracker and record data all day every day especially over the long-term. The resulting incompleteness in data poses an important challenge for interpreting long-term tracker data, in terms of both making sense of it and in dealing with the uncertainty of inferences based on it. Surprisingly, there has been little work into defining the problem, its extent and how it should be measured and addressed.
This thesis tackles this key challenge and we demonstrate the need for a term to describe and quantify this challenge. We introduce the term, adherence, which quantifies the completeness in such data. We also offer interface designs that accounted for adherence to support self-monitoring and reflection. Bringing these together, we provide broader definitions and guidelines for incorporating adherence when making sense of long-term physical activity tracker data, both in personal applications and in public health research results.
This thesis is based on three studies. First is a semester-long study of tracker use by 237 University students. Second is a study of 21 existing long-term physical activity trackers and provided the first richly qualitative exploration of physical activity and adherence of such users. It also evaluated the iStuckWithIt, a long-term physical activity data user interface, and reported on insights gained within and as aided by a tutorial and reflection scaffolding. In the final study, we drew on 12 diverse datasets, for 753 users, with over 77,000 days with data and 73,000 days missing to explore the impact of different definitions of adherence and methods for dealing with its implications.