The Eigenvector Friendship Paradox

Not just “your friends have more friends than you” — your friends are more central & influential than you. Watch it emerge in a random graph, then watch it sharpen as the network grows.

The classic friendship paradox (Feld, 1991) says the average person's friends have, on average, more friends than they do — because popular people show up in more friend-lists. This demo explores a 2026 generalization (arXiv:2607.19549 — The Generalized Friendship Paradox for Eigenvectors): replace “number of friends” with eigenvector centrality — the recursive notion of importance where you're influential if you're connected to influential people (the Perron eigenvector of the adjacency matrix, the math behind PageRank). The paper proves that in graphon-generated random graphs, a majority of nodes have lower centrality than the average of their neighbors, and it pins down the exact limiting distribution of that gap. Everything below is computed live in your browser: a real adjacency matrix → real power iteration for the Perron eigenvector → a real per-node eigen-bias. Nothing is faked. The dashed curve is the graphon's predicted limiting shape, computed the same way on a fine grid.

Graphon shape

Slide up and watch the histogram tighten toward the dashed limiting curve — that convergence is the paper's theorem.

Overall wiring probability scale of the graphon.

— of nodes are less central than their neighbors' average
(the paradox — expect well above 50%)
Mean eigen-bias—
Perron λ₁ (Rayleigh)—
Power-iter steps—
Edges—
Mean degree—

The network, colored by eigen-bias

below neighborsabove
eigen-bias < 0 — less central than neighbors (the paradox) eigen-bias > 0 — more central than neighbors node size ∝ eigenvector centrality · hover a node

Distribution of eigen-bias across all nodes

eigen-bias = (my centrality) − (mean of my neighbors' centrality)
live empirical (this graph, N nodes) graphon limiting shape (N → ∞, fine grid) mass to the left of 0 = paradox nodes