Built environment determinants of residents' running activity in Guangzhou, China: what are the nonlinear relationship and pathways?

Although studies have indicated that the built environment (BE) is linked to running activity, they are still uncertain about the nonlinear relationship and mediation effect of BE on this activity in high-density Chinese cities. This study used fitness tracker application, street view, social media text, and other multi-sourced data as bases to establish a framework combining the Shapley Additive exPlanation model, restricted cubic splines, and mediation analysis to address the nonlinearity and interpretability of the relationship between BE and running activity. Results of the empirical analysis are as follows. First, BE was significantly associated with residents' jogging distance (JD). Second, nonlinear relationships and threshold effects between BE and running activity were found. In particular, inverted U-shaped relationship existed between JD and bluespace proportion (BP)/ distance to bus station (DBS), within the effective range of 0.0231-0.350 in BP and 109-1893 m in DBS; S-shaped relationship existed between JD and density of road intersection (RI) / POI sum (PS)/green view index (GVI), within the effective range of 41-63 in RI, 99-183 in PS and 0.137-0.382 in GVI; U-shape relationship between JD and distance to nearest subway station (DSS)/ normalized difference vegetation index (NDVI) was observed, with peak benefits observed at an DSS of 0.628 km or an NDVI of 0.138. Thirdly, mean temperature and relative humidity had moderation effect between BE and running activity. Lastly, sentiment improvement played separate roles in mediating the effects of BE on residents' running activity. Results provide profound insights and boost in the design of additional attractive and sustainable BE for running activity, as well as the development of running-friendly city construction.