Leveraging the Popularity of Wearable Fitness Monitors: Advantages of Using Self-Tracking Data to Study Health Behaviors

Introduction: In recent years, the use of self-tracking devices to monitor health behaviors has increased rapidly (Kim, 2014), and studies have suggested that these consumer-level physical activity monitors are valid measures of step count and sleep duration (Ferguson et al., 2015).  With the increased popularity and accessibility of commercial devices to measure physical activity, such as pedometers and accelerometers (e.g., Fitbit, Nike Fuelband, Jawbone Up), comes an untapped source of readily available health information. Further, these devices offer numerous advantages in their affordability and ease of use. However, one of the most salient advantages to using self-tracking data to study health outcomes is the ability to collect intensive longitudinal data on physical activity and sleep cycles. These data allow for more sophisticated statistical methods focusing on the individual-level of analysis to be conducted. When paired with other sources of health information (e.g., demographics, health history, food logs, daily diaries, etc.), data from self-tracking devices can help to uncover insightful information about health outcomes. Thus, the aim of this paper is to discuss the advantages of using self-tracking data in order to study health behaviors, as well as provide insights into what we have learned in the process of collecting and analyzing this type of data.

Methods: Nomothetic statistical methods focus on uncovering universal laws of human behavior through use of aggregated data, whereas idiographic approaches focus on the individual and predicting future behavior from prior patterns. Multilevel modeling (MLM) is one method commonly used to analyze longitudinal data. While MLM allows for the modeling of individual trajectories, parameters are typically examined at an aggregate level (e.g., mean values of intercepts and slopes). Alternatively, time series analysis may be used to model processes on an individual-level, fitting unique models to each individual.

Results: We compare nomothetic (i.e., MLM) and idiographic statistical methods (i.e., time series analysis) in the context of studying health behaviors with intensive longitudinal data obtained via Fitbit activity trackers.

Conclusions: Use of self-tracking technologies can provide intensive longitudinal data at very little cost to researchers interested in studying health behaviors. Further, this type of data allows researchers to obtain individual-level data to investigate models of health outcomes.