X

        Comprehensive methods and data sources can be found here.

        Step 1: Initial modules are computed using various input networks: a separate set of modules is created for networks of commuters, migrants, trips, and social media relationships (Twitter and Facebook-based), using the Fast-Greedy community detection method.

        Step 2: An adjacency matrix of counties is created using a Delaunay triangulation-based triangular irregular network (TIN) to connect adjacent county centroids. Each edge is weighted so that if nodes are both in the same module from step 1, they receive a ‘point’. The modules of interest in Step 2 are: commuters, migrants, trips and states (determined by state boundaries). Social media modules are not included due to low performance in encapsulating COVID-19 cases (see arxiv paper).

        Step 3: The adjacency matrix is re-weighted where each ‘point’ can be weighted from 0-4 in the agreement matrix. For instance, if two counties are in the same module for commuters and migrants only, and the unweighted version their agreement weight would be 2. However, in the weighted version (taking user input weights), if commuters and states are each weighted 1 and 1 and migrants and trips are weighted 0 and 0, the weighted agreement would be 1 (weights can range from 0 – 16).

        Step 4: Community detection is then performed on the weighted adjacency matrix, and the output is returned based on the user inputs.