Physical partisan proximity outweighs online ties in predicting US voting outcomes

TL;DR

The study tackles whether offline physical partisan exposure outweighs online exposure in predicting US voting outcomes. It integrates Colocation Maps (offline), the Social Connectedness Index (online), residential exposure data, and ANES survey data, applying spatial lag models, dominance analysis, logit regressions, and predictive modeling with SHAP for explainability. Results show offline exposure explains nearly all variance in county-level vote shares ($R^2$ around $0.97$) and outperforms online and residential exposures across metro and non-metro areas, with offline ties also more predictive of individual vote choices in ANES. The findings highlight the primacy of physical space in shaping political behavior and suggest offline social structure and segregation, particularly linked to education and urbanization, are central to understanding electoral polarization.

Abstract

Affective polarization and increasing social divisions affect social mixing and the spread of information across online and physical spaces, reinforcing social and electoral cleavages and influencing political outcomes. Here, using individual survey data and aggregated and de-identified co-location and online network data, we investigate the relationship between partisan exposure and vote choice in the US by comparing offline and online dimensions of partisan exposure. By leveraging various statistical modeling approaches, we consistently find that partisan exposure in the physical space, as captured by co-location patterns, more accurately predicts electoral outcomes in US counties, outperforming online and residential exposures. Similarly, offline ties at the individual level better predict vote choice compared to online connections. We also estimate county-level experienced partisan segregation and examine its relationship with individuals' demographic and socioeconomic characteristics. Focusing on metropolitan areas, our results confirm the presence of extensive partisan segregation in the US and show that offline partisan isolation, both considering physical encounters or residential sorting, is higher than online segregation and is primarily associated with educational attainment. Our findings emphasize the importance of physical space in understanding the relationship between social networks and political behavior, in contrast to the intense scrutiny focused on online social networks and elections.

Figures (4)

• Figure 1: The three dimensions considered to estimate partisan exposure. (a) A co-location event between two randomly selected individuals from counties i and j is defined as being co-located in the same place for at least 5 minutes, while the Social Connectedness Index accounts for the number of friendships on Facebook between individuals from i and j. Residential proximity considers the nearest 1,000 individuals that registered to vote. (b and c) Co-location probabilities and relative probabilities of friendship on Facebook, respectively, between Jackson County, MO, and all the others. (d) Distributions of partisan exposure by county, including metro and non-metro areas. Note: distributions are not population-weighted.
• Figure 2: Partisan segregation across demographic and socioeconomic factors for each dimension. Maps and population-weighted distributions of the partisan segregation as captured by the Colocation Maps (a), Social Connectedness Index (b), and at the residential level (c). The red solid line represents the weighted average, while the black dashed line indicates balanced social mixing ($0$). For each dimension, the SHAP values distributions, computed from the GB regressions, highlight how the demographic and socioeconomic characteristics of the counties impact partisan segregation. Predictors are ordered by their impact on the final prediction, with points colored red for high and green for low values.
• Figure 3: Relative contribution of the three dimensions of partisan exposure on voting patterns. Physical partisan exposure outweighs online and residential exposure considering all the counties in the contiguous US in both spatial models (a) with $k=7$ ($R^2$) and dominance analysis (d) which considers demographic and socioeconomic controls. The result is consistent in both metropolitan (b and e) and non-metro areas (c and f), employing both OLS models ($R^2$) and dominance analysis.
• Figure 4: Average marginal effects of partisan exposure (online and offline) on vote choice. Offline partisan exposure has a stronger average marginal effect on vote choice than online partisan exposure, both in terms of exposure to Democrats (a) and Republicans (b). Logit models control for respondents' age, ethnicity, educational attainment, and place of residence (metro or non-metro area). The dependent variable is binary, with 0 for voting Democrat and 1 for voting Republican.