This website presents three different ways to compare the spatial structures of different mobility measures over a period of time. The Animated Choropleth Maps page allows you to either play an animation or step through individual weeks to see the spatial structure of our different measures in those weeks. The Interactive Maps page shows the temporal aspects of this animation in a single single view. The Scatter Plots page shows the same points as on our Interactive Maps, just without the spatial aspect to see a different view of the same information.
We have two data sources, Safegraph and Cuebiq, and two variables from each source. One variable is a measure of the distance traveled for people in a county, the other is a measure of the percentage of people in the county that are staying at home. We then computed weekly averages of these variables on the county level for each week in 2020.
We can use the Global Moran's I to test for spatial association between counties. Below are four plots for each data source and variable which show the test statistic of Global Moran's I, calculated for each week in 2020. All of these values are positive and significant which indicates that there is spatial association every week for each variable. This means that we generally see high valued counties next to other high valued counties (hotspots) and low valued counties next to other low valued counties (coldspots).
The vertical lines in the plots show the week that COVID-19 was declared a pandemic by the World Health Organization.
Since we have determined that there is spatial association (hotspots and coldspots) we can now use Local Moran’s I to find which counties are the hotspots and coldspots.
The first accompanying figure shows %-sheltered Cuebiq measure for Utah for the week 4/13-4/19. The percentage value is placed above each county and counties are colored according to this value. Counties with a lower value are colored a darker blue while counties with greater values are lighter. This plot shows what it means to be a cold spot; in the center of the state, we see lower percentages (darker counties) clustered together.
The next plot of Utah shows the results after running the Local Moran test. Counties that are significant coldspots after running the Local Moran test are given a color. The significance test that is used to determine whether the counties are coldspots is based on permutations of values that result in simulated p-values. We then use the False Discovery Rate (FDR) in order to reduce the number of false positives from this simulation. The counties that are significant after the FDR are colored a darker purple and these counties represent the centers of our coldspots.
Cuebiq % Sheltered for Utah 4/13-4/19. Darker shades of blue indicate a lower percentage of the population sheltering.
Counties classified as cold spots in Cuebiq % Sheltered for Utah 4/13-4/19. The ligher shade of purple shows significance at alpha=0.05 for the simulated p-values. The darker shade of purple are the counties that are significant using FDR.