The building blocks of social structure: simulating constraints of social network analysis for inference about group-level properties
This paper presents a simulation model that links global social network metrics to underlying behavioral processes—such as individual dispositions, ecological constraints, and observation biases—to demonstrate how ignoring these factors can lead to biased inferences about animal group-level social structures.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Animals live in complex social worlds where individuals constantly come together and drift apart. To understand how these societies work, scientists often turn to a method called social network analysis. This approach treats a group like a map of connections, where every animal is a point and every time they are seen together is a line linking them. By counting these lines and measuring how tightly they are woven, researchers can calculate numbers that describe the group's overall shape, such as how crowded the connections are or how the group breaks into smaller circles. These numbers are powerful tools for comparing different species or tracking how a society changes over time. However, a critical question remains: do these numbers truly reflect the animals' hidden social preferences, or are they simply artifacts of how the group is sized, how the animals move, or how long the scientists watched them? If the numbers are skewed by these external factors, researchers might draw the wrong conclusions about why a society is built the way it is.
A team of researchers set out to solve this puzzle by building a digital model of a fission-fusion society, a type of social system where the group splits and merges frequently, much like the societies of chimpanzees and bonobos. Instead of watching real animals in the wild, they created a virtual population of individuals, each with specific, hidden preferences for who they liked to be near. They then programmed a computer to simulate how these virtual animals would form temporary parties throughout a day, mimicking the way field researchers observe them. The team ran thousands of these simulations, systematically changing the rules of the game to see how the resulting social maps changed. They tested what happened when the total group size grew, when the size of the daily parties fluctuated, when some individuals were naturally more social than others, and when the researchers watched for different lengths of time.
The simulations revealed that the numbers scientists use to describe social structure are far more sensitive to context than previously realized. The most powerful driver of change was simply the number of animals in the group. As the group grew larger, the calculated connections between individuals became less dense, not because the animals were becoming less friendly, but because there were simply more possible combinations of who could be together. Even when the researchers adjusted their calculations to account for this, the size of the group still warped other measurements in complex, non-linear ways. For instance, the variability in how connected each animal was to the rest of the group changed dramatically with group size, suggesting that a larger group naturally creates more differences between individuals, regardless of their specific personalities.
The study also showed that the behavior of the observers matters just as much as the behavior of the animals. When the researchers in the simulation focused more on individuals who were naturally more social, the resulting social maps looked significantly different, exaggerating the importance of those specific animals and distorting the overall picture of the group. Similarly, the length of time spent watching the group had a profound effect. Short observation periods, such as thirty days, produced highly variable and often misleading results, making the group appear more fragmented and unpredictable than it actually was. The simulations indicated that researchers need to observe a group for at least one hundred and eighty days, or roughly two days per individual, before the numbers stabilize enough to be trusted.
Perhaps most surprisingly, the team found that the presence of tight-knit subgroups, or cliques, did not always make the social map look more divided. The relationship between the number of these cliques and the overall structure of the group was not a straight line. In some cases, having a few distinct cliques made the group look more unified, while in others, it created a sharp divide. The results suggested that real chimpanzee and bonobo groups likely do not form rigid, permanent cliques. Instead, they maintain a flexible structure where individuals tolerate a degree of social uncertainty to keep the entire group connected, avoiding the risk of the society splitting apart permanently.
Ultimately, this work serves as a vital warning and a guide for anyone studying animal societies. It demonstrates that the numbers we calculate from social networks are not direct windows into the animals' minds; they are the product of a mix of true social preferences, the constraints of the environment, and the limitations of our own observation. The researchers did not find a single "correct" number to describe a society. Instead, they showed that to understand what a social network metric really means, one must first understand the specific conditions under which it was measured. By using these simulations to separate the signal of true social behavior from the noise of demographic and observational bias, scientists can begin to open the "black box" of animal social structure and understand the true evolutionary forces that shape how groups live together.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.