Capturing the fast-food landscape in England using large-scale network analysis

Fast-food outlets play a significant role in the nutrition of British children who get more food from such shops than the school canteen. To reduce young people’s access to fast-food meals during the school day, many British cities are implementing zoning policies. For instance, cities can create buffers around schools, and some have used 200 meters buffers while others used 400 meters. But how close is too close? Using the road network is needed to precisely computing the distance between fast-food outlets (for policies limiting the concentration), or fast-food outlets and the closest school (for policies using buffers). This estimates how much of the fast-food landscape could be affected by a policy, and complementary analyses of food utilization can later translate the estimate into changes on childhood nutrition and obesity. Network analyses of retail and urban forms are typically limited to the scale of a city. However, to design national zoning policies, we need to perform this analysis at a national scale. Our study is the first to perform a nation-wide analysis, by linking large datasets (e.g., all roads, fast-food outlets and schools) and performing the analysis over a high performance computing cluster. We found a strong spatial clustering of fast-food outlets (with 80% of outlets being within 120 of another outlet), but much less clustering for schools. Results depend on whether we use the road network on the Euclidean distance (i.e. ‘as the crow flies’): for instance, half of the fast-food outlets are found within 240 m of a school using an Euclidean distance, but only one-third at the same distance with the road network. Our findings are consistent across levels of deprivation, which is important to set equitable national policies. In line with previous studies (at the city scale rather than national scale), we also examined the relation between centrality and outlets, as a potential target for policies, but we found no correlation when using closeness or betweenness centrality with either the Spearman or Pearson correlation methods.