Global, local, and directional spatial autocorrelation of Republican vote share, 2012-2016

• Question. Was the shift in Republican support between 2012 and 2016 geographically random, or did it move in regional blocks?

• Data. County-level presidential results for 3,108 counties in the contiguous U.S.; Census QuickFacts demographic indicators.

• Methods

  • Built a queen-contiguity spatial weights matrix in GeoDa.

  • Measured global clustering with Moran’s I (999 permutations) for 2012 share, 2016 share, and change.

  • Mapped local clusters (LISA) of GOP strongholds and of GOP gains and losses.

  • Ran a directional analysis comparing each county’s movement with its neighbors’ movement, plotted as origin-standardized vectors.

• Findings. The change in GOP share was even more clustered than vote share itself (Moran's I = 0.64). 77% of counties moved in the same direction as their neighbors. GOP gains clustered in 602 mostly rural Midwestern and Appalachian counties home to about 19 million people. GOP losses clustered in 405 faster-growing metropolitan counties home to about 132 million.

• Tools. GeoDa, ArcGIS Pro

• Images. Directional scatterplot; LISA cluster maps of change