MultiLevel Modeling of Space–Time Variations: Exploring Landslide Voting Patterns at United States Presidential Elections, 1992–2016

Much has been written about the polarization of the American electorate and its reflection in its legislatures, but less about its spatial polarization, which Bishop has argued has taken place in parallel with the ideological and behavioral polarization. The extent of that polarization can be assessed, he argues, by identifying the number of landslide counties, those won at presidential elections by margins of 20 percentage points or more. This paper uses a multilevel modeling strategy to explore changes in the number and extent of those landslide counties over the period 1992–2016, relative to both the location of the counties and their population composition. It shows that a county’s population composition was a major determinant of whether it returned a landslide for either party’s candidate at any election—with a clear change in direction over the period for counties according to their level of affluence—but this was by no means the sole determinant. Holding constant those variations there were additional geographies that were more place- than people-specific.