Compositional Data Analysis in Time-Use Epidemiology: What, Why, How

Abstract

In recent years, the focus of activity behavior research has shifted away from univariate paradigms (e.g., physical activity, sedentary behavior and sleep) to a 24-h time-use paradigm that integrates all daily activity behaviors. Behaviors are analyzed relative to each other, rather than as individual entities. Compositional data analysis (CoDA) is increasingly used for the analysis of time-use data because it is intended for data that convey relative information. While CoDA has brought new understanding of how time use is associated with health, it has also raised challenges in how this methodology is applied, and how the findings are interpreted. In this paper we provide a brief overview of CoDA for time-use data, summarize current CoDA research in time-use epidemiology and discuss challenges and future directions. We use 24-h time-use diary data from Wave 6 of the Longitudinal Study of Australian Children (birth cohort, n = 3228, aged 10.9 ± 0.3 years) to demonstrate descriptive analyses of time-use compositions and how to explore the relationship between daily time use (sleep, sedentary behavior and physical activity) and a health outcome (in this example, adiposity). We illustrate how to comprehensively interpret the CoDA findings in a meaningful way.

Using Compositional Data Analysis to Explore Associations Between Accumulation of Sedentary Behaviour and Physical Activity and Biomarkers of Cardiometabolic Health in Children and Adolescents: A Cross-Sectional Study

“…Participant age and sex were included as continuous and binary variables, respectively. Socioeconomic status (SES) of school location was transformed into deciles, based on the national Index of Relative Socio-economic Disadvantage of the Socio-Economic Indexes for Areas [24] and categorized as low (1-2), mid (3)(4)(5)(6)(7)(8) or high (9 − 10) SES [12]. Data source (LOOK or Transform-Us!)…”

“…Most activities were ubiquitous, except for the VPA longer bouts component, which was monitored as zero for ten participants. A limitation of compositional data analysis is that zero values cannot be included in log ratios, as dividing by zero or taking the logarithm of zero are unde ned mathematical operations [8]. As it can be assumed that the ten participants (1%) with zeros may accumulate some amount of time in longer VPA bouts, just not during the measurement period, these participants were removed from the sample before further statistical analyses were undertaken.…”

“…From a 24-h behaviour perspective, sleep and waking movement behaviours occur on an activity continuum that range from low-to high-intensity activity. A change in time spent in one activity intensity (e.g., moderate-to-vigorous intensity PA; MVPA) will consequently result in a change in at least one other intensity (e.g., LPA; sleep) in this nite 24-h period [7,8]. As such, considering the overall daily time-use composition of total time spent in various activity intensities (i.e., volumes) can yield a different perspective to considering each intensity in isolation [7].…”

“…Using compositional data analysis, the relative distribution of time use between different activities can be examined, classifying activities by their total volume of intensities (e.g., total SED), activity accumulation patterns (e.g., time in longer versus shorter bouts), other attributes such as behavioural context, or by multiple attributes, simultaneously [7,8]. Then, these respective elements of time use can be modelled simultaneously and tested for their combined associations with biomarkers of cardiometabolic health [7][8][9]. This compositional data analysis approach is advanced in that it allows the simultaneous consideration of components that sum to a whole, without statistical problems such as collinearity, commonly encountered in some traditional approaches [7,8].…”

“…Then, these respective elements of time use can be modelled simultaneously and tested for their combined associations with biomarkers of cardiometabolic health [7][8][9]. This compositional data analysis approach is advanced in that it allows the simultaneous consideration of components that sum to a whole, without statistical problems such as collinearity, commonly encountered in some traditional approaches [7,8]. To date, the application of compositional data analysis to studying time use and health has mostly been applied to adults [7,9] rather than in youth [10].…”