A null model for Dunbars circles

An individual’s social group may be represented by their ego-network, formed by the links between the individual and their acquaintances. Ego-networks present an internal structure of increasingly large nested layers (or circles) of decreasing relationship intensity, whose size exhibits a precise scaling ratio. Starting from the notion of limited social bandwidth, and assuming fixed costs for the links in each layer, we propose a null model built on a grand-canonical ensemble that generates the observed hierarchical social structure. The observed internal structure of ego-networks becomes a natural outcome to expect when we assume the existence of layers demanding different amounts of resources. In the thermodynamic limit, reached when the number of ego-network copies is large, the specific layer degrees follow a Poisson distribution. We also find that, under certain conditions, equispaced layer costs are necessary to obtain a constant group size scaling. Our model presents interesting analogies to a Bose-Einstein gas, that we briefly discuss. Finally, we fit and compare the model with an empirical social network.

💡 Research Summary

The paper proposes a statistical‑mechanics‑based null model that reproduces the hierarchical “Dunbar circles” observed in ego‑networks. Each layer (or circle) of an ego‑network is assigned a fixed social‑resource cost sᵣ, and an individual possesses a total resource s and an average degree k̄. By maximizing entropy under the constraints of fixed average degree and resource, the authors construct a grand‑canonical ensemble with Lagrange multipliers λ and µ, leading to a Gibbs distribution over the layer‑degree configuration {kᵣ}. The partition function can be evaluated analytically, yielding closed‑form expressions for the expected layer degrees ⟨kᵣ⟩, their variances, and inter‑layer correlations. In the thermodynamic limit (number of ego‑networks N → ∞ while k̄ and s̄ remain constant), the layer degrees become independent Poisson variables with mean ⟨kᵣ⟩ ∝ N e^{λ+µsᵣ}. Consequently, the marginal distributions are Poisson, and correlations vanish.

The authors then examine the condition for a constant scaling ratio between successive cumulative group sizes (the empirical ≈3 factor). By approximating the ratio ⟨nᵣ⟩/⟨nᵣ₊₁⟩ in terms of layer means, they show that a constant ratio requires the cost differences between adjacent layers to be equal (Δ = sᵣ − sᵣ₊₁). Thus, equally spaced layer costs naturally generate the observed scaling. The parameter y = e^{µ} controls whether outer layers contain more links (y < 1) or fewer (y > 1); the former matches typical human social structures, while the latter describes inverse regimes where resources are concentrated in inner circles.

To validate the model, the authors fit it to the Reciprocity Survey dataset comprising 84 undergraduate participants who rated relationships on a discrete scale. By fixing the empirical averages of total degree and total resource, they solve for λ and µ, then generate synthetic ego‑networks. The synthetic networks match the empirical data in overall degree, total resource, layer‑wise means, variances, and especially the scaling of cumulative group sizes across layers.

Finally, the paper draws an analogy between the proposed ensemble and a Bose‑Einstein gas: layers correspond to energy levels, acquaintances to bosonic particles, and the grand‑canonical formalism yields Bose‑Einstein‑like occupation numbers. This connection highlights how concepts from quantum statistical physics can illuminate the emergence of hierarchical organization in complex social systems. Overall, the work demonstrates that a minimal set of assumptions—limited social bandwidth and fixed per‑layer costs—suffices to generate the empirically observed Dunbar circles, providing both a theoretical baseline and a tool for generating realistic synthetic ego‑networks.