A massive geographically weighted regression model of walking-environment relationships
Abstract
Many studies aim at identifying environmental correlates of walking in order to identify specific potential levers for tackling the medical burden of physical inactivity. The links between environmental characteristics and walking behaviors are usually context specific. While local studies fail to embrace a global overview of these contexts, global studies hide the context scale patterns. In this study, we applied a geographically weighted logistic regression (GWR) on a large area (whole of France) to explore spatial variations of the relations between five environmental variables and walking for leisure and errands purposes among 40,480 French adults. This approach allowed us to adopt a global view of local patterns of relations and to highlight spatial contexts (defined through a clustering of GWR odds ratios) where combinations of correlates varied. Specifically, clustering algorithms on the GWR odds ratios led to 9 and 6 clusters for walking for leisure and errands, respectively. Some clusters were characterized by a particularly strong effect of population density, whereas others exhibited low effect of vegetation cover rate. Chi-squared tests indicated that these clusters were associated with type of urban areas (Paris, major urban poles, periurban areas, small urban poles, isolated areas) for the two types of walking. Beyond its methodological contribution - providing a method to handle large data samples into GWR analyses - this study offers key elements to practitioners and policy makers to target relevant contexts and environmental features for promoting daily walking.