As European societies become more unequal, polarized, and diverse, concerns about what holds them together have never been more urgent. For decades, researchers have measured social cohesion – the bonds of trust, cooperation, and shared norms that sustain a functioning society – primarily through attitude surveys: Do people trust their neighbors? Do they feel they belong Yet these surveys overlook what may be cohesion’s fundamental dimension: the actual structure of social relationships, who talks to whom, and who is quietly left out.
Studying social networks at the scale of entire societies has remained out of reach, forcing researchers either to rely on simplified simulations or on data from online social networking platforms. However, together with my co-authors I show that realistic, society-wide networks of intimate discussion ties can be generated by linking two previously separate strands of network research: survey-based personal network data and statistical network modeling.
What’s more, the resulting simulations reveal structural inequalities that attitude surveys are unable to detect. For example, applied to a nationally representative sample from Spain, this approach reveals that the country’s core discussion network lacks the “small-world” properties commonly assumed in sociological theorizing and modeling, and that immigrants are roughly twice as likely as native-born residents to occupy peripheral, structurally isolated positions. These findings provide a fresh sociological approach to empirically grounded, cross-national analysis of how social network structure shapes tolerance, trust, and cohesion.
The idea that societies are held together by invisible bonds is old. Émile Durkheim, the nineteenth-century sociologist, described two forms of solidarity: “mechanical” solidarity, rooted in shared beliefs and customs, and “organic” solidarity, arising from the mutual dependence created by the division of labor. Contemporary sociologists continue to build on this foundation, distinguishing several dimensions of cohesion: the quality of social relationships and mutual trust; the sense of shared identity and belonging; orientation toward the common good; shared values; and equal access to opportunities.
Most empirical studies measure these dimensions using surveys. Tools such as the Social Cohesion Radar administer batteries of questions to representative samples, producing scales that track how trust, civic engagement, or solidarity change over time and differ across countries. While this has been generating valuable knowledge, it leaves the social structure out of the picture:
These surveys measure how people feel about cohesion; they are less equipped to capture how cohesion actually works at the level of a society, through the web of connections between people.
This is where Computational Social Science (CSS) comes to the fore. As a methodological approach, a marriage between social and computational sciences, CSS supplements traditional sociological theorizing with computer simulations and large-scale data analysis. It particularly shines in the context of micro-macro phenomena. It treats social cohesion not as a sentiment to be surveyed but as an emergent property of a social system to be analyzed. Hence, through the lens of CSS social cohesion emerges as a macro-sociological property that is built, or eroded, through myriads of actions and interactions of individuals.
Despite social relationships such as friendships being core dimensions of social cohesion, they have been barely studied at the level of entire societies. Broad networks connecting individuals to hundreds of acquaintances have long been assumed to bind communities together and provide a sense of solidarity. But this assumption has been rarely tested empirically – studying society-wide networks is technically difficult. This is why core discussion ties offer a useful entry point: three to five people with whom a person regularly discusses things that matter to them. These are typically strong, trust-based relationships. At the individual level, they are well-documented: small, kin-centered, and socially homogeneous.
But what does a full web of such relationships look like when mapped across an entire society of millions of people? Who is at the center, and who is at the margins?
These questions cannot be answered by looking at individuals in isolation. Instead, the structure of core discussion networks at the societal level determines how fast norms and information spread, how political views are reinforced or challenged, and whether a society remains integrated or fragmented along social fault lines. Answering these questions requires society-wide network data, and obtaining it is a challenge.
Data from online social networking platforms and administrative registers provide large-scale maps of connectivity, but they do not capture ties that are directly sociologically relevant. They show neighbors, colleagues, and online contacts, but without indicating whether people actually confide in one another.
The chapter I have co-authored, “Simulating an Empirically Informed Population Network of Core Discussion Ties” offers a solution by hybridizing two strands of network research that have previously operated separately.
- The first is personal network surveys: asking a representative sample of individuals about their discussion partners, their characteristics, and the relationships among them. This is an affordable and nationally representative solution but captures only each person’s “local” view.
- The second is statistical network modeling, specifically, Exponential-family Random Graph Models (ERGMs). These models help to understand how people form relationships: who tends to connect with whom and what role similarity and other “local” mechanisms play in shaping the structure of the entire network.
Thus, the innovation is in applying the ERGM framework to personal network survey data, here taken from the Spanish General Social Survey, a nationally representative sample of approximately 5,100 adults, collected in 2013. In the survey, respondents named the people they had discussed important matters in the preceding six months, and described both those people and the connections among them.
The networks simulated using the ERGM framework provide three key findings that would otherwise be invisible:
- A core discussion network is no “small world.” The “small-world” phenomenon – the idea that any two people can be connected in a surprisingly small number of steps (the famous “six degrees of separation”) – has been widely assumed to characterize social networks and is built into most agent-based models. Yet our simulations contradict this. Paths between people in Spain’s core discussion network are long, resembling the distances found in networks of family ties, rather than the shorter distances typical of colleague and online-friendship networks. This means that some norms, behavioral patterns, and information might travel more slowly and unevenly through the social fabric than assumed.
- Echo chambers are structurally embedded. People overwhelmingly discuss important matters with others similar to themselves. Our model confirms assortativity, the tendency to connect according to similar characteristics, by age, occupational status, political views, and religion. Thus, it shows that people born in Spain tend to discuss important matters primarily with other Spaniards; and Catholics talk about significant issues predominantly with other Catholics. What’s more, the network itself reinforces divisions rather than bridging them – it is a finding with important implications for research on polarization.
- Structural exclusion is measurable. Most strikingly, the analysis reveals a form of exclusion that surveys cannot detect: by examining which social groups fall outside the main connected part of the network, the study finds that immigrants are roughly twice as likely as native-born Spaniards to be structurally peripheral, which means being cut off from the main web of conversational ties. Those who are unemployed and otherwise inactive on the labor market show similar patterns. These are not just feelings of exclusion; they are measurable positions in the architecture of society.
In sum, we uncover that social cohesion includes a structural dimension which has been systematically understudied: knowing how people feel about their society is important but knowing who they actually talk to, and who is being quietly left out, may matter even more.
___
The research chapter is available in: Bojanowski, M., Loseva, A., Schuler, P., Böller, S., & Lubbers, M. J. (2026). Simulating an Empirically Informed Population Network of Core Discussion Ties. In M. A. Keijzer, J. Lorenz, & M. Bojanowski (Eds.), Computational Social Science of Social Cohesion and Polarization (pp. 55–82). Springer, Cham. https://doi.org/10.1007/978-3-032-01373-6_3
