A computational model explains the sudden rise in societal division

As modern communication technology has doubled the number of close social ties people maintain, societies have fractured into deeply divided political camps. A computational model simulating human social behavior shows that simply having more connections past a certain threshold forces communities to splinter into polarized echo chambers. The research was published in the Proceedings of the National Academy of Sciences.

Political polarization has risen sharply in the United States and other Western nations over the past two decades. During this same period, the average number of close friendships reported by individuals has nearly doubled, a trend coinciding with the public debut of smartphones and mass social media networks around 2008.

Stefan Thurner, Markus Hofer, and Jan Korbel, researchers at the Medical University of Vienna and the Complexity Science Hub Vienna, noticed the parallel timing of these two trends. They wanted to test whether the sheer volume of social connections might be driving the ideological divide.

To investigate this, the researchers focused on two well-documented features of human behavior: homophily and social balance. Homophily is the human tendency to seek out and befriend like-minded individuals. When people interact with those who share their views, they often reinforce each other’s beliefs, causing their opinions to align even more closely.

Social balance describes how groups of three people naturally seek comfortable, stable relationship dynamics. A balanced triad occurs when all three people are friends, or when two friends mutually dislike a third person.

Unbalanced triads cause psychological tension or social stress. This happens if one person tries to remain friends with two people who despise each other, or if three people all dislike one another. People naturally try to resolve this tension by either changing their opinions to match their friends or by distancing themselves from people who disagree with them.

To see how these basic social rules play out on a large scale, the research team built an agent-based computational model. In this virtual society, each simulated individual was assigned a mathematical profile representing a set of binary political stances on different topics.

Individuals in the simulation were connected in a network where relationships could be either positive or negative. A positive connection formed if two individuals shared a majority of their stances, acting as a virtual friendship. A negative connection formed if their stances opposed one another, acting as a rivalry or enmity.

The researchers used a mathematical algorithm to randomly select individuals and flip one of their political stances. After flipping an opinion, the model calculated the individual’s new level of cognitive dissonance, measured by how much they now agreed with their friends and disagreed with their enemies.

If the new opinion lowered their overall social stress, the individual kept the new stance. If the opinion increased their social stress, the individual rejected the change, though the model allowed a small probability of keeping it to represent occasional irrational behavior. As individuals updated their views, their positive and negative relationships shifted to maintain social balance.

The researchers ran this simulation multiple times, varying the average number of close connections each individual possessed. They tracked how the distribution of opinions across the entire population changed as the network became more interconnected.

The simulation revealed a sudden tipping point, which physicists call a phase transition. When the average number of close connections remained low, the virtual society maintained a diverse, cohesive blend of opinions. Small friend groups shared similar views, but these groups remained scattered and tolerant of the broader population.

Once the average number of connections surpassed a critical threshold of about five close friends, the societal structure abruptly collapsed into a highly polarized state. The numerous small, diverse clusters merged into two massive, opposing groups of roughly equal size.

Within each massive group, individuals held nearly identical opinions. Between the two groups, individuals held entirely opposite views and maintained strictly negative relationships with anyone on the other side.

Next, the researchers introduced radical influencers to the model. They assigned extreme, unchanging opinions to just two percent of the simulated population. These influencers never adjusted their stances to reduce social stress, but they still formed friendships and rivalries based on ideological alignment.

The presence of these rigid influencers changed the nature of the polarization tipping point. The transition from a cohesive society to a divided one became gradual rather than sudden. However, the splintering into opposing echo chambers began at an even lower threshold, occurring when individuals averaged just four and a half close connections.

To test if their model reflected reality, the research team compared their simulated results to real-world survey data. They analyzed data from large, multi-year surveys, including the Pew Research Center, which track both political ideology and the average number of close friendships among thousands of respondents.

The empirical data showed that ideological polarization in the United States remained relatively low and stable before 2010. Shortly after that time, the average number of close social ties increased, and polarization spiked rapidly.

The researchers found that their computational model accurately predicted both the timing and the magnitude of this real-world polarization spike. The model required no outside factors like economic changes or political campaigns to explain the divide. The mathematical rules of homophily and social balance, combined with an increasing number of social ties, were enough to produce the exact polarization trend seen in the survey data.

The study does present certain caveats. The computational model assumes every individual has the exact same sensitivity to cognitive dissonance and social stress. In reality, human beings display varied levels of tolerance for disagreement and differing levels of political engagement.

The simulation also uses a relatively fixed network structure. While relationships can switch between positive and negative, the underlying web of who knows whom does not fluidly dissolve and rewire over time. The authors note that real social media platforms use recommendation algorithms that constantly suggest new connections, which might amplify the clustering effect beyond what the current model captures.

One concerning finding from the model is a hysteresis effect, meaning the process is difficult to reverse. Once a society crosses the connectivity threshold and becomes highly polarized, simply reducing the number of connections back to the original level will not depolarize the population. A fractured society must drop its connectivity far below the initial tipping point to restore ideological cohesion.

Since reducing human connectivity is likely impossible in the digital age, the authors suggest alternative ways to raise the critical threshold. They suggest that educational efforts could train individuals to have a higher tolerance for opposing opinions, which mathematically prevents the phase transition. Disrupting the reach of radical influencers through content moderation could also delay the onset of extreme societal division.

The study, “Why more social interactions lead to more polarization in societies,” was authored by Stefan Thurner, Markus Hofer, and Jan Korbel.