Skip to content
Back to search results

Weighted Degree Centrality

Introduced by
Barrat, A.; Barthélemy, M.; Pastor-Satorras, R.; Vespignani, A.
Year
2004
Type
Centrality measure for weighted networks
Family
Network science
Font size

What does it do?

Weighted degree centrality — also called node strength — extends the classic degree centrality measure to networks whose edges carry numeric weights.

Instead of simply counting a node's connections, it sums the weights of all edges incident to that node, capturing both the volume and the intensity of a node's ties in a single, interpretable score.

The intuition behind the method

In an unweighted network, a node's degree counts how many neighbors it has, but a researcher emailing 10 colleagues weekly is not equivalent to one who emails each colleague once a year. Weighted degree centrality captures this distinction: it sums the actual strength of each connection. A high score signals a node that is not merely well-connected but that maintains strong or frequent relationships, making it a natural measure of influence or load-bearing importance in weighted social, biological, or infrastructure networks.

History

The concept of node strength as the weighted generalisation of degree was formally introduced by Barrat, Barthélemy, Pastor-Satorras, and Vespignani in their 2004 PNAS paper analysing the architecture of complex weighted networks, including the global air-transportation network. Their framework became a foundational reference for weighted network science and is now standard material in network analysis textbooks such as Newman's Networks: An Introduction (2010). The measure has since been applied in neuroscience, economics, ecology, and computational social science.

When to use it

Use weighted degree centrality when your network has meaningful numeric edge weights — collaboration frequency, communication volume, trade value, physical flow — and you want to identify the most active or influential nodes without the complexity of path-based measures. It is the appropriate first centrality measure when binary degree would discard important variation in tie strength. Do not use it when edge weights are nominal or ordinal categories (convert to meaningful ordinal scores first), when network structure rather than raw throughput matters most (prefer betweenness or eigenvector centrality), or when the network is unweighted (use standard degree centrality instead).

Limitations and considerations

  • Sensitive to the scale and distribution of edge weights; skewed weights can inflate a few nodes' scores dramatically.
  • Does not account for the position of a node in the global network structure — a peripheral but heavily weighted node may outscore a topologically central one.
  • Ignores the direction of edges in directed networks unless in-strength and out-strength are computed separately.
  • Provides no information about clustering, brokerage, or the reach of indirect connections.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Computationally cheap — O(m) where m is the number of edges — and trivially parallelizable for large networks.
  • Intuitive interpretation: it directly reflects the total volume or intensity of a node's connections.
  • Discriminates between nodes with equal degree but different tie strengths, which binary degree centrality cannot do.
  • Widely implemented in NetworkX, igraph, Gephi, and R's igraph package.
  • Combines naturally with unweighted degree to detect whether high-degree nodes also carry heavy traffic.

Applications

  • Identifying the most active or influential actors in weighted social and collaboration networks.
  • Detecting major hubs in transportation and trade networks by total flow or volume.
  • Ranking brain regions by synaptic or functional connectivity strength in connectome studies.
  • Quantifying prolific co-authorship or citation influence in bibliometric networks.
  • Detecting key species in ecological food webs weighted by interaction frequency or biomass flow.
  • Benchmarking node importance before applying more expensive path-based centrality measures.

How it works

How it works

1

Represent the network as a weighted graph G = (V, E, W) where nodes V are actors or entities, edges E are ties, and the weight matrix W assigns a positive real value w_ij to each edge, encoding frequency, intensity, cost, or some other domain-relevant quantity. Negative weights require careful pre-processing or an alternative formulation.

2

For each node i, sum the weights of all edges connecting it to its neighbors: s_i = sum of w_ij over all j in the neighborhood of i. This quantity is referred to as node strength in the literature of Barrat et al. (2004) and directly generalises binary degree to the weighted case.

Formula
s_i = \sum_{j \in \mathcal{N}(i)} w_{ij}
3

Divide each node's strength by the total weight in the network (or by the maximum node strength) to obtain a normalized score between 0 and 1. Normalization enables fair comparison across networks of different sizes or total weight budgets and is standard practice before ranking or regression analyses.

Formula
s_i^{\text{norm}} = \frac{s_i}{\sum_k s_k}
4

Sort nodes by their normalized strength scores and identify high-centrality nodes. In social networks these are the most active or influential actors; in trade networks, the largest hubs; in co-authorship networks, the most prolific collaborators. Compare with unweighted degree to identify nodes whose influence is amplified or attenuated by tie strength.

Common pitfalls

  • Treating raw weights as comparable across datasets without standardization or log-transformation when weights are heavily skewed.
  • Forgetting to compute in-strength and out-strength separately in directed networks and collapsing them to a single score without justification.
  • Reporting weighted degree centrality without comparing it to binary degree centrality, thereby missing whether high strength is driven by many weak ties or few strong ones.
  • Using the measure on self-loops without deciding whether to include or exclude diagonal weights.
  • Ignoring missing or zero-inflated weights which can systematically underestimate the strength of poorly observed nodes.

Frequently asked questions

How does weighted degree centrality differ from standard degree centrality?

Standard degree centrality counts the number of edges incident to a node, treating all connections equally. Weighted degree centrality sums the edge weights, so a node connected to one very strong tie can outscore a node with many weak ties. The two measures together reveal whether a node's influence comes from breadth or intensity of connections.

Should I log-transform edge weights before computing node strength?

When weights span several orders of magnitude — for instance, trade volumes ranging from thousands to billions — log-transformation prevents a handful of extreme ties from dominating the scores. Always inspect the weight distribution and apply transformation if it is heavily right-skewed.

How do I handle directed networks?

Compute in-strength (sum of incoming edge weights) and out-strength (sum of outgoing edge weights) separately. In-strength often reflects popularity or authority, while out-strength reflects activity or effort. Combining them into a single score discards directional information.

What is a reasonable software choice for computing weighted degree centrality?

NetworkX in Python (strength method via degree with weight parameter), igraph in R or Python (strength function), and Gephi (built-in weighted degree column) all compute node strength with a single function call. For very large sparse networks, graph-tool or GraphBLAS-based implementations are more efficient.

When should I prefer betweenness centrality over weighted degree centrality?

Use betweenness centrality when you care about which nodes control information or resource flow along shortest paths — brokers and bridges in a network. Use weighted degree centrality when you care about raw activity or throughput at each node. They are complementary and often informative when reported together.

Related methods

Sources

  1. Barrat, A., Barthélemy, M., Pastor-Satorras, R., & Vespignani, A. (2004). The architecture of complex weighted networks. Proceedings of the National Academy of Sciences, 101(11), 3747–3752.

    Open DOI reference
    Copied.
  2. Newman, M. E. J. (2010). Networks: An Introduction. Oxford University Press.

    ISBN: 978-0-19-920665-0
    Copied.