Weighted Eigenvector Centrality
Weighted Eigenvector Centrality (Spectral Prestige in Weighted Networks) · Also known as: WEC, weighted spectral centrality, strength-weighted eigenvector centrality, weighted eigenvector prestige
Weighted eigenvector centrality extends the classic eigenvector centrality measure to graphs where edges carry numerical weights, scoring each node proportionally to the sum of its neighbors' scores multiplied by the connecting edge weights. Nodes score highly not just by having many connections but by being strongly linked to other influential nodes, making the measure sensitive to both tie strength and network position simultaneously.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use weighted eigenvector centrality when your network has meaningful edge weights (collaboration frequency, communication volume, trade flow, co-occurrence count) and you need to identify nodes that are both strongly connected and connected to other well-connected nodes. It is especially appropriate in co-authorship networks, economic trade networks, and brain connectivity analyses. Avoid it when edge weights are ordinal or arbitrarily scaled, when the network is highly sparse or disconnected (the principal eigenvector may not converge or may be dominated by a small clique), when weights represent distances rather than affinities, or when interpretability of a single summary score is insufficient and path-based measures such as betweenness are more relevant.
Strengths & limitations
- Captures both tie strength and global prestige simultaneously, unlike degree centrality which ignores who the neighbors are.
- Theoretically grounded in spectral graph theory; the Perron-Frobenius theorem guarantees a unique, non-negative solution for connected non-negative matrices.
- Sensitive to edge weight heterogeneity, rewarding nodes with strong links to other influential nodes over nodes with many weak links.
- Widely implemented in standard network libraries (NetworkX, igraph, R's igraph package), making it accessible without custom code.
- Applicable to both undirected and directed weighted networks.
- Problematic on disconnected or near-disconnected graphs, where the principal eigenvector can collapse to zero for peripheral components.
- Sensitive to weight scaling: multiplying all weights by a constant changes eigenvalues but not rankings, yet arbitrary weight transformations can distort results.
- Does not account for path length or bridging roles; nodes on important shortest paths may be underrated compared to betweenness-based measures.
- Computationally expensive for very large dense networks; sparse approximations or power-iteration methods are needed at scale.
Frequently asked
How does weighted eigenvector centrality differ from standard (binary) eigenvector centrality?
Binary eigenvector centrality uses an unweighted adjacency matrix, so each neighbor contributes equally regardless of tie strength. The weighted version uses a weight matrix, so a neighbor contributes in proportion to the weight of the connecting edge, making tie strength explicitly part of the prestige calculation.
What if my network is directed?
For directed weighted networks the weight matrix is asymmetric. You can compute left eigenvectors (reflecting in-flow prestige, analogous to a weighted hub score) or right eigenvectors (reflecting out-flow prestige), depending on whether you care about who receives influence or who sends it. This is analogous to the distinction between authority and hub scores in HITS.
Can negative edge weights be used?
The Perron-Frobenius guarantee of a non-negative principal eigenvector applies only to non-negative matrices. Negative weights (e.g., antagonistic relationships) break this guarantee and can produce eigenvectors with mixed signs that are difficult to interpret. Pre-process negative weights by either separating positive and negative layers or applying an absolute-value transformation with caution.
Is it better than weighted degree (strength) centrality?
They capture different things. Weighted degree (node strength) counts total weight of direct ties but ignores who the neighbors are. Weighted eigenvector centrality incorporates global network position recursively. In practice, the two measures correlate strongly in many empirical networks, so if eigenvector centrality fails to converge or the graph is disconnected, node strength may be the safer fallback.
How should I handle very large networks?
Use sparse matrix representations and power-iteration algorithms (available in NetworkX and igraph) rather than full eigendecomposition. Check convergence tolerance and iteration limits. For networks with millions of nodes, consider sampling or community-level aggregation before applying the measure.
Sources
- Bonacich, P. (1987). Power and centrality: A family of measures. American Journal of Sociology, 92(5), 1170–1182. DOI: 10.1086/228631 ↗
- Opsahl, T., Agneessens, F., & Skvoretz, J. (2010). Node centrality in weighted networks: Generalizing degree and shortest paths. Social Networks, 32(3), 245–251. DOI: 10.1016/j.socnet.2010.03.006 ↗
How to cite this page
ScholarGate. (2026, June 3). Weighted Eigenvector Centrality (Spectral Prestige in Weighted Networks). ScholarGate. https://scholargate.app/en/network-analysis/weighted-eigenvector-centrality
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Degree CentralityNetwork analysis↔ compare
- Eigenvector CentralityNetwork analysis↔ compare
- Weighted Betweenness CentralityNetwork analysis↔ compare
- Weighted Closeness CentralityNetwork analysis↔ compare
- Weighted Degree CentralityNetwork analysis↔ compare
- Weighted PageRankNetwork analysis↔ compare