Scale, Context, and Heterogeneity: A Spatial Analytical Perspective on the 2016 U.S. Presidential Election

Abstract

This article attempts to identify and separate the role of spatial “context” in shaping voter preferences from the role of other socioeconomic determinants. It does this by calibrating a multiscale geographically weighted regression (MGWR) model of county-level data on percentages voting for the Democratic Party in the 2016 U.S. presidential election. This model yields information on both the spatially heterogeneous nature of the determinants of voter preferences and the geographical scale over which the effects of these determinants are relatively stable. The article, perhaps for the first time, is able to quantify the relative effects of context versus other effects on voter preferences and is able to demonstrate what would have happened in the 2016 election in two scenarios: (1) if context were irrelevant and (2) if every county had exactly the same population composition. In addition, the article sheds light on the nature of the determinants of voter choice in the 2016 U.S. presidential election and presents strong evidence that these determinants have spatially varying impacts on voter preferences.

本文试图在美国选民倾向的众多影响因子中, 确定空间“环境”的作用, 并把空间环境的作用从其它社会经济因子的作用中分割出来。根据2016年美国总统选举的县级民主党得票率数据, 本文校正了一个多尺度地理加权回归模型。该模型输出两种信息：选民倾向的决定因子具有空间异构性、在某个地理尺度下这些因子具有稳定性。在空间环境影响选民倾向的定量化上, 本文可能是首创研究。本文还展示了2016年选举的两个假想场景：（1）如果空间环境无关, （2）如果各县具有相同的人口构成。此外, 本文阐明了2016年美国总统选举中选民投票的决定因子。有很强的证据显示：这些因子对选民倾向的影响在不同空间上有所不同。

Este artículo intenta identificar el rol que tiene el “contexto” espacial en la configuración de las preferencias de los votantes, y separarlo de los roles que cumplen otros determinantes socioeconómicos. Se hace esto calibrando un modelo multiescalar de regresión geográficamente ponderada (MGWR) de datos a nivel de condado sobre porcentajes de voto por el Partido Demócrata en la elección presidencial de los EE.UU. del 2016. Este modelo rinde información tanto sobre la naturaleza espacialmente heterogénea de los determinantes de las preferencias de votantes, como de la escala geográfica a la cual los efectos de los determinantes son relativamente estables. Quizás por primera vez, el artículo es capaz de cuantificar los efectos relativos del contexto contra otros efectos sobre las preferencias del votante y es capaz de demostrar qué habría podido ocurrir en la elección del 2016 en dos escenarios: (1) si el contexto fuera irrelevante y (2) si todo condado tuviera exactamente la misma composición de la población. Además, el artículo arroja luz sobre la naturaleza de los determinantes de la selección por el votante en la elección presidencial de los EE.UU. de 2026, y presenta fuerte evidencia de que estos determinantes tienen impactos espacialmente variables sobe las preferencias del votante.

Key Words:

• context
• local modeling
• multiscale geographically weighted regression
• voter preferences scale

关键词：空间环境:

• 局域建模
• 多尺度地理加权回归
• 选民倾向尺度。

Palabras clave:

• contexto
• escala de preferencias del votante
• modelado de lo local
• regresión multiescalar geográficamente ponderada

Notes

1 The data are available from the MIT Election Data and Science Lab (see https://electionlab.mit.edu/data).

2 The variables we selected were the product of a long period of experimentation, discussion with political scientists and political geographers, a thorough reading of the literature in this area, and common sense. We cannot, however, claim that the selection is theory based, because there is little acknowledged theory in the social sciences and particularly when it comes to examining the determinants of individuals’ voting preferences. This should not denigrate the research, however; much can be learned from empirical experimentation. Confidence in the variable selection is gained from the very strong replication power of our model both at the global level and at the local level (see below).

3 The data were gathered from the American Community Survey 2012–2016 five-year estimates. No variance inflation factor was greater than 4.

4 There is little evidence to suggest that any major explanatory variable has been omitted in our model, nor what such a variable would be to “explain” the spatial pattern of what we call context here. As described earlier, what is meant by context here is an otherwise unmeasurable effect on people’s preferences for one political party over another based solely on locality (the influence of family and friends, local media, etc.), which is independent of any other measurable effect on voting preference (e.g., income, ethnicity, age, etc.). The pattern that emerges is very similar to that of most people’s mental images of political leanings across the United States. Here, possibly for the first time, we have been able to quantify this effect.

Additional information

Notes on contributors

A. Stewart Fotheringham

A. STEWART FOTHERINGHAM is a Regents’ Professor of Computational Spatial Science in the School of Geographical Sciences and Urban Planning at Arizona State University, Tempe, AZ 85281. E-mail: [email protected]. His research interests include spatial data analytics, spatial processes, and spatial interaction modeling.

Ziqi Li

ZIQI LI is a Visiting Assistant Professor in the Department of Geography and Geographic Information Science, University of Illinois, Urbana–Champaign, IL 61820. E-mail: [email protected]. His research interests include GIScience, spatial data science, spatial statistical learning, and their applications in multidisciplinary fields.

Levi John Wolf

LEVI JOHN WOLF is a Senior Lecturer (Assistant Professor) in the School of Geographical Sciences at the University of Bristol, UK. E-mail: [email protected]. His research interests include probabilistic programming, machine learning for applications in urban geography, politics, sociology, and economics.