1 Centre de recherche du Centre hospitalier universitaire (CHU) Sainte-Justine, Montréal, Canada.
2 Département de médecine sociale et préventive, École de santé publique de l'Université de Montréal, Montréal, Canada.
3 Office of International Affairs for the Health Portfolio, Public Health Agency of Canada, Ottawa, Ontario, Canada.
4 Le Centre de recherche du Centre hospitalier de l'Université de Montréal (CRCHUM), Montréal, Canada.
5 Cedars-Sinai Medical Center, Department of Medicine, Los Angeles, CA, USA.
6 Department of Family Medicine, Faculty of Medicine, McGill University, 5858 Côte-des-Neiges Rd, Montreal, QC, H3S 1Z1, Canada.
7 Department of Mathematics and Statistics, Concordia University, Montréal, Canada.
8 School of Kinesiology and Physical Activity Sciences, University of Montréal, Montréal, Canada.
9 Department of Pediatrics, University of Montréal, Montréal, Canada.
10 Centre de recherche du Centre hospitalier universitaire (CHU) Sainte-Justine, Montréal, Canada. tracie.barnett@mcgill.ca.
11 Department of Family Medicine, Faculty of Medicine, McGill University, 5858 Côte-des-Neiges Rd, Montreal, QC, H3S 1Z1, Canada. tracie.barnett@mcgill.ca.

Background: We compared the relation between neighborhood features and moderate to vigorous physical activity (MVPA) using linear regression analysis and the more novel compositional data analysis (CoDA). Compositional data analysis allows us to take the time children allocate to different movement behaviours during a 24-hour time period into account.

Methodology: Data from youth participants (n = 409) in the QUALITY (QUebec Adipose and Lifestyle InvesTigation in Youth) cohort were included. Time spent in MVPA, light physical activity, sedentary behavior, and sleep ("24-hour movement behaviours") was measured using accelerometers. Neighborhood data were collected using a geographic information system and through direct observation. In CoDA models, we used orthogonal logratio coordinates, which allows for the association of neighbourhood walkability with MVPA to be estimated with respect to the average composition of all other behaviours within a 24-hour time frame. In baseline linear regression models, MVPA was regressed cross-sectionally on neighborhood walkability. All models were stratified by sex, and controlled for BMI z-scores, pubertal development, seasonal variation, parental education, and neighbourhood safety.

Results: Based on CoDA, girls who lived in more walkable neighborhoods had 10% higher daily MVPA (95% CI: 2%, 19%), taking into account all other movement behaviours. Based on linear regression, girls who resided in more walkable neighborhoods engaged in 4.2 (95% confidence interval [CI]: 1.2, 6.6) more minutes of MVPA per day on average than girls residing in less walkable neighborhoods.

Conclusions: Unlike with traditional linear models, all movement behaviours were included in a single model using CoDA, allowing for a more complete picture of the strength and direction of the association between neighbourhood Walkability and MVPA. Application of CoDA to investigate determinants of physical activity provides additional insight into potential mechanisms and the ways in which people allocate their time.