Does rich-club membership mean a node is wealthy?

No. In this definition it means the node passes a degree threshold.

Are nodes with degree exactly k included?

Not in the supplied definition; their degree must be strictly greater than k.

Does the standard NetworkX routine measure weighted rich clubs?

No. Its documented calculation ignores edge weights; weighted analyses require an explicitly suitable definition.

AI Glossary / Graphs and symbolic AI / Intermediate

Rich Club

Definition

A rich club in a network is a set of highly connected nodes that also connect densely to one another. In the degree-based definition, the rich-club coefficient measures the fraction of possible edges actually present among nodes whose degree exceeds a chosen threshold. Comparison with a suitable random-network baseline helps assess whether that density is greater than expected from degree structure alone.

What it means

Here “rich” means having many connections, not necessarily money, status, or importance. A transport hub, a heavily linked device, or a frequently connected organization can qualify by degree even when its other attributes are unremarkable.

High-degree nodes often have many opportunities to connect, so a dense core is not automatically evidence of an unusual rich-club effect. The raw coefficient describes density within the selected group. A normalized comparison asks a different question: how that density compares with an explicitly chosen reference model. Both the threshold and the reference model should be reported.

How it works

For a simple undirected graph and threshold k, select nodes whose original graph degree is strictly greater than k. Let N_k be their number and E_k the number of edges between them. The raw coefficient is 2E_k/[N_k(N_k−1)], provided at least two selected nodes exist.

Repeat across useful thresholds to obtain a profile rather than relying on one arbitrary cutoff. For normalization, compare with coefficients from randomized reference graphs that preserve the relevant degree structure.

The supplied NetworkX routine uses this degree-based undirected definition, ignores edge weights, and does not support directed graphs, parallel edges, or self-loops. Its normalization uses degree-preserving edge swaps. A zero reference coefficient makes a ratio problematic, and randomization settings affect reproducibility and interpretation.

Three practical examples

Infrastructure analysis: examine whether highly connected network locations also form a densely interconnected group that merits separate resilience analysis.
Collaboration mapping: describe connections among organizations with many partners while comparing the result against an appropriate degree-based baseline.
Graph-learning studies: characterize a dataset's hub structure before assessing whether a model behaves differently on central and peripheral nodes.

Think of it this way

Think of the busiest stations in a transport map. A rich-club question asks whether those stations have many direct links among themselves, not simply whether many routes pass through each one. The analogy omits service frequency, capacity, and geography unless those are explicitly included in the graph or a different weighted analysis.

Worked example

Build a fictional undirected graph with nodes A, B, C, D, E, and F. Include triangle edges A - B, B - C, and A - C, plus A - D, B - E, and C - F. There are six edges. Nodes A, B, and C each have degree three; D, E, and F each have degree one.

At threshold k = 1, select nodes with degree greater than one: A, B, and C. Thus N_k = 3 and E_k = 3. The coefficient is 2×3/(3×2) = 1. Every possible edge among the selected nodes is present.

At threshold k = 0, all six nodes are selected. The coefficient is 2×6/(6×5) = 0.4, which is the whole graph's density. At k = 3, no node qualifies, so the formula has no meaningful density value; this is not a measured coefficient of zero.

For a separate arithmetic illustration, suppose an appropriately defined reference ensemble had a mean coefficient of 0.75 at the chosen threshold. A raw value of 1 would yield a normalized ratio of 1/0.75, approximately 1.33. The 0.75 is stipulated, not calculated from randomizations of this toy graph. The example does not establish statistical significance or a rich-club finding for real data.

Illustrative case study

Fictional case: Oak Network Classroom studies a partnership graph and reports a high coefficient for its most connected organizations. A reviewer asks whether the result is surprising given the degree sequence and how the selected threshold was chosen.

Students add a degree-preserving reference comparison and report the threshold profile, graph construction rules, and randomization settings. They also explain that a binary edge means a recorded partnership, not its value or current activity level.

The case illustrates the information needed to interpret a density statistic. It does not claim that the selected organizations coordinate behavior or that membership in a dense core establishes power, influence, or misconduct.

Common misconceptions and mistakes

Rich-club density is not the same as degree assortativity. Assortativity summarizes degree relationships across edges, while rich-club analysis focuses on connectivity within threshold-selected high-degree groups.

A raw coefficient close to one can occur in a very small group. Report the number of selected nodes rather than presenting the coefficient alone.

Do not interpret a normalized ratio above one as a probability or automatically as statistical significance. It is a comparison with a particular null model. Sampling variability, graph construction, and multiple threshold choices can all affect the strength of a conclusion.

Business use

Businesses can use rich-club analysis as one descriptive view of infrastructure, collaboration, or dependency networks. It can suggest where further investigation is useful, but it does not by itself quantify resilience, traffic capacity, or operational criticality.

Document what nodes and edges mean, whether the graph is complete enough for the question, and why the degree threshold is relevant. Keep raw and normalized results distinct. When normalization fails because a reference coefficient is zero, report that limitation instead of substituting an arbitrary value that would make comparisons misleading.

Check your understanding

Three selected nodes have two edges among them. What is their raw rich-club coefficient?

Answer: There are three possible undirected edges, so the coefficient is 2/3, equivalently 2×2/(3×2).

Related terms

Hub
Assortativity
Degree
graph density
null model
degree-preserving randomisation
network core
centrality

Learn next

Calculate the coefficient at several thresholds on a small hand-drawn graph and record both N_k and E_k. Then study why degree-preserving randomization provides a different comparison from a random graph with only the same total edge count. Keep statistical inference separate from the raw descriptive calculation.

Sources

NetworkX, rich_club_coefficient: https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.richclub.rich_club_coefficient.html McAuley, Costa, and Caetano, The Rich-Club Phenomenon Across Complex Network Hierarchies: https://arxiv.org/abs/physics/0701290 Official definition and primary abstract were read. Graph, stipulated ratio, and fictional case are original; no randomisation was executed.

Previous: Rhetorical structure theory Next: Ridge detection