The authors develop a simulation framework that generates fission–fusion spatial association data from an explicit set of parameters. The principal aim is to bridge the gap between global network metrics produced by social network analysis (SNA) and the array of latent processes that generate observed group-level social structures. By producing synthetic association data from known inputs, the model provides a means to interpret how changes in behaviour, demography, ecology, and sampling translate into variation in commonly used global network measures.
The framework breaks down animal social systems into three conceptual building blocks, each encoded in model parameters:
Each building block can be adjusted independently in the simulations so that the contribution of each class of process to measured network properties can be examined.
The model maps underlying processes to widely used global network metrics within an SNA framework. Examples of the metrics quantified include clustering coefficient, network density, and modularity. By computing these metrics on networks derived from simulated data, the framework permits direct comparison between the parameter settings that generated the data and the values of global metrics they produce. This establishes mechanistic links between behavioural/ecological inputs and the SNA outputs researchers commonly report.
When single parameters were varied while holding others constant, demographic and ecological parameters consistently affected global network metrics. Specifically, changes in group size (demography) and the size and variability of foraging parties (ecology) altered metrics non-linearly and, in some cases, non-monotonically. These results emphasize that simple changes in group composition or the ecological context of associations can produce complex and unintuitive shifts in global SNA measures.
The simulations show that variation in individual behaviour and related observation bias have measurable effects on network metrics. Likewise, varying observation effort alters all calculated global metrics. Because observation effort and detectability shape the raw association data used to construct networks, these sampling factors can introduce biases into downstream inference if not accounted for. The model highlights the magnitude and direction of such effects under controlled parameter changes.
Imposing group sub-structures in the model affects global metrics in predictable ways: explicit subgrouping or constraints that favour within-subgroup associations change measures such as modularity and clustering. These results demonstrate how latent organisational features of a population—whether intrinsic or externally imposed—drive the network-level signatures that researchers measure.
Across simulated scenarios, the authors find that global network metrics are sensitive to multiple, interacting underlying processes. Because metrics can respond non-linearly and sometimes non-monotonically to parameter variation, interpreting differences in network measures across populations or studies without considering the generating processes risks biased or incomplete inference. The framework therefore provides a tool for diagnosing which behavioural, ecological, demographic, or sampling processes are plausible drivers of observed network patterns.
The authors position their model and conceptual decomposition as a means to “open the black box” of animal SNA. By linking global metrics directly to parameterised underlying processes, researchers can (1) probe how specific changes at the individual or environmental level influence group-level structure, (2) assess the sensitivity of commonly reported metrics to sampling design and observation bias, and (3) generate null or mechanistic expectations against which empirical networks can be compared. The paper argues that adopting such a simulation-based approach will help clarify the proximate and ultimate drivers of diverse group-level social structures and reduce misinterpretation arising from unexamined methodological or ecological constraints.