The origins of large-scale structure in family networks

Every family is a small world of its own, but when you look at millions of families together, they form a vast, invisible web that connects an entire society. This web is built on the most fundamental human decisions: who we choose to love, who we choose to have children with, and how those relationships change over time. For decades, scientists studying these connections have focused heavily on the idea that people tend to choose partners who are just like themselves—similar in age, education, or where they grew up. This tendency, known as assortative mating, was long believed to be the primary engine driving the shape of the entire family network, determining how closely related distant cousins might be or how quickly information could travel from one end of a society to the other. It seemed logical that if people only married their "own kind," the network would naturally become a collection of isolated clusters, keeping different groups of people apart.

However, a new study using a massive, detailed record of Danish families suggests that this common assumption misses the bigger picture. By analyzing a complete network of six million people and eight million parent-child relationships spanning from 1953 to 2018, researchers discovered that the true architect of the family network's large-scale structure is not who people choose to marry, but rather who they choose to leave. The study reveals that the act of changing partners—having children with one person and then later having children with someone else—is the force that weaves the entire society together. While people do indeed pair up with those similar to them, this similarity plays a surprisingly minor role in shaping the overall map of family connections. Instead, it is the movement of individuals between partners that creates the shortcuts necessary to link distant families, turning a slow-growing, fragmented collection of small groups into a single, interconnected whole.

The researchers began by constructing a digital model of the Danish family network, a dataset so large it covers nearly every person living in the country for over six decades. They wanted to understand how individual behaviors aggregate to create the massive structures seen in the data. To do this, they built a series of computer simulations that could recreate the growth of the family network over time. In their first set of models, they programmed the system to mimic the real world's tendency for people to choose partners who share their background, such as growing up in the same town or having similar levels of education. They expected this "homophily," or the love of the same, to be the dominant force shaping the network. They also tested models where people never changed partners, staying with the same person for every child they had.

The results were counterintuitive. When the researchers ran simulations that included only the tendency to choose similar partners but no partner changes, the resulting network looked nothing like the real Danish family network. It remained fragmented, with families staying isolated from one another for long periods. The connections between families were too few and too far apart. The network grew slowly, and large groups of connected people took decades to form. This suggested that simply choosing similar partners was not enough to explain the speed and scale at which real family networks connect. The models that included partner changes, however, told a different story. When the simulations allowed individuals to have children with a new partner after leaving an old one, the network transformed. These changes created "shortcuts" that jumped over generations and linked families that would otherwise have remained strangers for a long time.

The study found that these shortcuts are crucial. In a network where people only have children with one partner, families connect only when their children grow up and have their own children, a process that can take twenty or thirty years. But when a parent has a child with a new partner, they instantly create a bridge between their old family and their new one. This happens much faster than waiting for the next generation. The researchers observed that in the real Danish data, about 10 percent of parents have children with more than one partner. This seemingly small number of people acts as a powerful glue, connecting hundreds of thousands of otherwise separate family trees. The presence of these "multi-partner" parents reduces the average distance between any two people in the network, making the entire society feel much smaller and more connected than it would be otherwise.

Furthermore, the researchers discovered that the behavior of changing partners is not random. They found a clear pattern: the more partners a person has had in the past, the more likely they are to change partners again. This tendency creates a specific type of structure where certain individuals become central hubs, connecting many different families together. In the simulations, when the researchers programmed the models to reflect this increasing likelihood of changing partners, the resulting networks looked almost identical to the real Danish data. They accurately reproduced the size of the largest connected groups, the speed at which these groups formed, and the distribution of how many relatives a person has at different distances.

In contrast, when the researchers added the factor of partner-choice homophily—the tendency to pick partners who are similar in age, education, or origin—to their models, it made very little difference to the large-scale structure. While it is true that people in the real data do pair up with those who are similar to them, including this factor in the simulations only slightly improved the match with reality. The most important factor remained the act of changing partners. Even when the models perfectly mimicked the way people choose similar partners, they failed to recreate the real network's structure unless they also included the dynamic of partner changes. The study explicitly ruled out the idea that assortative mating is the primary driver of the network's shape, showing instead that it is a secondary effect that operates on a structure already defined by partner mobility.

The implications of this finding extend beyond just understanding family trees. The way families are connected determines how traits like wealth, education, and even genetic risks are distributed across a society. If the network is highly segregated, with families staying isolated from one another, these traits remain concentrated within specific groups. If the network is well-connected, these traits can flow more freely. The researchers used their models to test how well they could predict the level of geographic segregation in Denmark. They found that models which ignored partner changes failed to capture the true level of segregation observed in the real data. Even when these models accounted for the fact that people tend to marry neighbors, they still produced networks that were too mixed and too connected. Only when the models included the realistic behavior of people changing partners did the simulations accurately reflect the degree to which people from the same towns remained clustered together in the family network.

This work highlights a previously overlooked link between individual life choices and the broad structure of society. It shows that the decision to start a family with a new partner, a choice often made for personal reasons, has a profound structural consequence that ripples through the entire social fabric. The study does not suggest that partner choice is unimportant; people clearly do choose partners based on shared backgrounds. But in terms of the grand architecture of the family network, the movement between partners is the dominant force. It is the mechanism that prevents the network from becoming a set of isolated islands and instead turns it into a single, complex, and rapidly evolving system.

The researchers were able to reach these conclusions because of the unique quality of their data. They had access to a complete record of the Danish population, allowing them to see the entire network rather than just a sample. This completeness meant they could track the exact moment a parent changed partners and see how that single event altered the connections between families. They validated their findings by running thousands of simulations, comparing the results of different models against the real-world data. The models that included the cumulative effect of partner changes—where the likelihood of changing partners grows with each new partner—were the only ones that could accurately reproduce the complex patterns seen in the real world.

Ultimately, the study offers a new perspective on how social networks grow. It challenges the long-held belief that similarity is the main glue holding society together. Instead, it suggests that the fluidity of relationships, the willingness to move from one partnership to another, is what truly binds us. By creating shortcuts between distant families, these changes accelerate the formation of a connected society. The findings suggest that to understand the flow of resources, information, or even genes across a population, one must look not just at who people marry, but at how those marriages change over time. The structure of our social world is not a static map drawn by our initial choices, but a dynamic landscape constantly reshaped by the paths we take when we leave one partner for another.