Fig. 1: Behaviour-based dependency networks between places in cities.
From: Behaviour-based dependency networks between places shape urban economic resilience
a, The dependency wji of POI j (for example, a restaurant) on POI i (for example, a college) is computed as the proportion of the intersection of individuals that visit both the college i and cafe j (nij) on the same day within 6 h (Tc) and within 1 step (Nc), out of the total count of individuals who visit cafe j, nj. Note that dependencies wij and wji are asymmetrical and bidirectional. A visualization of total in-weight (\({w}_{i}^{\mathrm{in}}={\sum }_{j}{w}_{\!ji}\)) for all POIs in the Manhattan area in New York is shown. The colours show the POI category and the node sizes show the total in-weight. b, A network diagram showing the average dependencies between POI subcategories in Boston (other cities shown in Supplementary Note 2). Each node is a POI subcategory, and the three largest outgoing dependency edges are shown for each node. The node sizes show the in-degree of the constructed network. c,d, A probability density distribution of the in-weight \({w}_{i}^{\mathrm{in}}\) (c) and out-weight (\({w}_{i}^{\mathrm{out}}={\sum }_{j}{w}_{ij}\)) (d) per node in the five cities, labeled with different marker shapes. The total in-weight \({w}_{i}^{\mathrm{in}}\) has a substantially larger variance compared with the out-weight \({w}_{i}^{\mathrm{out}}\), indicating the existence of nodes with a large attraction of dependencies, such as New York University (NYU, annotated in a). e, Average dependency \(\overline{{w}_{ij}}\) of all POIs i and j with a Haversine distance of dij. Average dependency decays with dij with a slope of \(\overline{{w}_{ij}}\propto {d}_{ij}^{-1.49}\). a was designed using icons from Flaticon.com created by Freepik and Education. The maps were produced in Python using the TIGER shapefiles from the US Census Bureau48.
