Modeling roles and trade-offs in multiplex networks

Nikolaos Nakis, Sune Lehmann, Nicholas A. Christakis, Morten MøruparXiv 2025 · arXiv · OpenReview

TL;DR

This paper introduces the Multiplex Latent Trade-off Model (MLT) to understand how people navigate multiple social relationships (social, health, economic). MLT reveals that social ties are deeply embedded and driven by mutual influence (interdependence), while health and economic ties are more about individual status and direct exchange. This has implications for understanding social behavior and designing more accurate network models.

What Problem Does This Paper Address?

Understanding human social life requires looking beyond simple connections. People are embedded in multiple types of relationships simultaneously – think friends, family, colleagues, doctors, and financial contacts. These "multiplex networks" are complex because each type of relationship (or "layer") can serve different purposes and be shaped by different dynamics. Traditional network analysis often treats each relationship type in isolation or focuses on simple metrics like who knows whom. However, this misses how individuals balance their efforts across different relationship domains and how mutual influence (interdependence) plays a role, especially when compared to individual attributes (independence) or the influence of others' status (dependence). The challenge is to model these trade-offs and understand how interdependence manifests differently across various social contexts.

Methods

The core contribution is the Multiplex Latent Trade-off Model (MLT). MLT uses a novel approach based on simplex-constrained latent embeddings.

  • Global Trade-offs: MLT positions each node on a simplex, where each corner represents a distinct social domain (social, health, economic). A node's position on this simplex indicates how it "trades off" its role and activity across these domains. This captures the global structure of interdependence.
  • Local Multi-scale Structure: Within each layer (social, health, economic), MLT models hierarchical community structures. This allows for the discovery of nested communities at different resolutions, capturing local patterns.
  • Social Exchange Theory Integration: The model explicitly incorporates concepts from Social Exchange Theory, defining three modes of interaction:
  • Independence: Tie formation driven by the sender's own attributes.
  • Dependence: Tie formation influenced by the receiver's attributes.
  • Interdependence: Mutual influence between sender and receiver, capturing complex, often asymmetric, dyadic interactions.
  • Link Prediction: The model's effectiveness is evaluated using link prediction, a standard task for assessing network model performance. They compare MLT against a "bias model" that only considers independence and dependence.

Datasets

The study uses a large, real-world dataset from a sociocentric network study of 22,584 people across 176 geographically isolated villages in western Honduras. This dataset comprises three layers of directed multiplex networks: social, health, and economic ties.

Key Insights and Ideas

  • Social Ties are Deeply Embedded: The social layer shows the strongest signal of interdependence. This means social connections are shaped by mutual influence and are less dependent on individual status or simple reciprocity. They are "structurally embedded."
  • Health and Economic Ties are Status-Driven: In contrast, health and economic ties are more influenced by individual status, resources, and direct behavioral engagement. Interdependence plays a subtler role here.
  • Roles are Trade-offs: Individuals don't just have one "role"; they distribute their "role expression" across different social domains. MLT quantifies these trade-offs.
  • Layered Interdependence: The impact of modeling interdependence varies significantly by layer. Social ties benefit most from this, suggesting a fundamental difference in how these relationships are formed and maintained.
  • Hierarchical Structure Matters: The model's ability to capture multi-scale, hierarchical community structures within each layer enhances its predictive power, especially for social ties.

Key Results

  • Link Prediction Gains: The full MLT model (including interdependence) consistently outperforms the bias model (independence and dependence only) in link prediction.
  • Social Layer: Achieves a PR-AUC gain of 15.07% and ROC-AUC gain of 11.81%.
  • Health Layer: Gains are more modest, with PR-AUC of 2.12% and ROC-AUC of 1.37%.
  • Economic Layer: Gains are moderate, with PR-AUC of 6.11% and ROC-AUC of 6.16%.
  • Interdependence Reduces Bias Reliance: When interdependence is modeled, the correlation between node biases (status indicators) and network degree centrality drops significantly, indicating that the model is reallocating explanatory power from individual status to dyadic interactions.
  • Social Dominance: Across all villages, social activity is dominant in both source and target roles, with a notable absence of individuals acting purely in health or economic roles.

Limitations and Future Work

  • Mechanistic Processes: While MLT provides a statistical lens for examining trade-offs and role emergence, it does not directly model the mechanistic processes of tie formation. It captures the statistical patterns but not the underlying "why" of social exchange dynamics.
  • Generalizability: The authors suggest the model can be extended to any network type exhibiting trade-offs, but the current study focuses on human social networks.
  • Future Work: The authors envision extending the model to explore "relational phenotypes" and profiles that align with specific roles, and using metadata to understand emergent structures. They also highlight the potential for applying the framework to other types of multiplex networks beyond social interactions.

Summary updated

Abstract

A multiplex social network captures multiple types of social relations among the same set of people, with each layer representing a distinct type of relationship. Understanding the structure of such systems allows us to identify how social exchanges may be driven by a person's own attributes and actions (independence), the status or resources of others (dependence), and mutual influence between entities (interdependence). Characterizing structure in multiplex networks is challenging, as the distinct layers can reflect different yet complementary roles, with interdependence emerging across multiple scales. Here, we introduce the Multiplex Latent Trade-off Model (MLT), a framework for extracting roles in multiplex social networks that accounts for independence, dependence, and interdependence. MLT defines roles as trade-offs, requiring each node to distribute its source and target roles across layers while simultaneously distributing community memberships within hierarchical, multi-scale structures. Applying the MLT approach to 176 real-world multiplex networks, composed of social, health, and economic layers, from villages in western Honduras, we see core social exchange principles emerging, while also revealing local, layer-specific, and multi-scale communities. Link prediction analyses reveal that modeling interdependence yields the greatest performance gains in the social layer, with subtler effects in health and economic layers. This suggests that social ties are structurally embedded, whereas health and economic ties are primarily shaped by individual status and behavioral engagement. Our findings offer new insights into the structure of human social systems.
BibTeX
@misc{nakis2025modeling,
  author = {Nikolaos Nakis and Sune Lehmann and Nicholas A. Christakis and Morten Mørup},
  title = {{Modeling roles and trade-offs in multiplex networks}},
  year = {2025},
  eprint = {2508.05488},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.05488},
}

Loading related content...

Figure