Real-world social ties outweigh online networks in predicting of voting patterns

A new study published in PNAS Nexus challenges prevailing views about the influence of online echo chambers on political behavior. The research provides evidence that Americans’ political environments in physical spaces—such as where they live, work, and spend time—are more predictive of voting patterns than their online social networks.

Political polarization in the U.S. has increasingly become a topic of national concern. Much of the public discussion has centered on the role of online spaces, particularly social media platforms, in dividing people along partisan lines. These digital environments have been blamed for fostering echo chambers, amplifying misinformation, and reducing contact with opposing viewpoints.

But people also live in neighborhoods, commute to work, and interact in public spaces. These physical environments may shape political views through casual or repeated encounters with others. While earlier research has examined polarization through residential data or online behavior, there has been limited large-scale analysis comparing these different types of exposure. The authors of this study aimed to fill that gap by examining how online, offline, and residential social networks relate to political segregation and vote choice in the U.S.

To compare different types of partisan exposure, the researchers relied on four major data sources. One set measured offline interactions using Facebook’s co-location data, which captures how often people from different counties are physically present in the same place for at least five minutes. This metric reflects passive encounters in public or shared spaces, such as on public transit, in stores, or at events.

A second set measured online social connections through Facebook’s Social Connectedness Index, which calculates the probability that two people from different counties are Facebook friends. These relationships are typically active, selective, and sustained over time.

Free daily newsletter

The third data source involved voter registration records to estimate residential partisan exposure—that is, the likelihood that someone lives near others who are affiliated with the same or opposing political parties. Finally, the researchers used individual survey responses from the American National Election Studies to understand how people’s offline and online networks related to their actual votes in the 2020 presidential election.

The researchers first assessed how strongly each type of exposure—offline, online, and residential—was associated with voting patterns at the county level. They found that co-location data, which reflects in-person physical proximity, was a stronger predictor of how counties voted than either Facebook friendships or residential partisan makeup. Counties where people frequently encountered others who supported a particular party were more likely to vote in alignment with those patterns.

This pattern held up even after accounting for demographic and socioeconomic factors, such as education levels, race, and urban versus rural composition. Offline social networks appeared to explain more of the variation in voting outcomes than either online ties or residential clustering.

Next, the researchers looked at partisan segregation—the extent to which individuals are exposed mostly to co-partisans versus a mix of political affiliations. They found that physical spaces, including both co-location patterns and residential arrangements, were more politically segregated than online networks. Offline segregation tended to be more pronounced in metropolitan areas with higher educational attainment and larger African American populations, both of which were linked to greater exposure to Democratic voters.

In contrast, counties with lower education levels or predominantly rural populations tended to show stronger exposure to Republican partisans. Online networks, while still showing some degree of partisan sorting, were more diverse than offline environments and displayed greater “extroversion,” meaning people were more likely to be connected to others outside their local region.

The final part of the analysis focused on individual behavior. Survey participants were asked about the political leanings of their family, friends, and Facebook connections. These self-reported networks were then compared to their stated vote in the 2020 presidential election.

Again, offline social exposure had a stronger relationship with vote choice than online connections. People who reported that their friends and family were mostly Democrats were more likely to vote for Joe Biden. Those who were mostly surrounded by Republican-leaning offline contacts were more likely to vote for Donald Trump. While online exposure also mattered, its influence on vote choice was noticeably smaller. This pattern remained stable across two waves of data collection, before and after the 2020 election.

The researchers also conducted several robustness checks. For example, when they removed local exposure (connections within the same county) from the analysis, the predictive power of offline networks declined significantly. In these cases, online ties sometimes became more predictive of voting patterns. This finding suggests that local, everyday interactions are a key component of political influence in offline environments.

The study draws on large-scale datasets that offer unique insights, but there are still some limitations. The offline and online network data were derived from Facebook, which may not fully capture the behavior of groups less active on the platform, such as older adults or people in rural areas. While the researchers applied adjustments to improve representativeness, there may still be biases in who is included in the co-location and friendship data.

The analysis also compares different types of networks—casual physical proximity, sustained online friendships, and residential proximity—each of which may involve different levels of interaction and influence. For instance, co-location does not necessarily mean that two people talked or knew each other, and online friendships may vary widely in strength.

Future research could build on these findings by attempting to measure an individual’s complete social network, both online and offline, using a single comprehensive data source. Despite these limitations, the study suggests that while online platforms are an important part of modern social life, the nature of our real-world interactions and the physical spaces we share appear to be more fundamentally linked to our political behaviors.

The study, “Physical partisan proximity outweighs online ties in predicting US voting outcomes,” was authored by Marco Tonin, Bruno Lepri, and Michele Tizzoni.