Limitations of Centrality

Centrality measures are mathematical tools used to find the most important nodes in a network, but relying on just one can be misleading. Think of it like judging a car's speed solely by its engine noise; loud doesn't always mean fast. In the example above, the 'Gossip' has high visibility (Degree Centrality), but the 'Liaison' holds the structural power (Betweenness Centrality) to control flow between groups. Relying on a single metric ignores the complexity of how information actually moves. Check this: Next time you identify a key player, ask if they are a 'hub' (popular) or a 'bridge' (critical).

Networks are complex, and "importance" is subjective based on what is flowing—whether it is viruses, rumors, or packages. Different measures capture different, often conflicting, types of influence.

The Context Trap High "Degree Centrality" (number of connections) signals popularity, but not necessarily influence. A receptionist talks to everyone but may not shape strategic decisions. Conversely, a "super-spreader" in a pandemic is best identified by degree, making the measure context-dependent.

The Bottleneck Blindspot Measures like "Betweenness Centrality" identify bridges. A researcher connecting Engineering to Sales might have few friends but controls the product launch timeline. If you only count connections, you miss this critical vulnerability.

The Boundary Problem Calculating centrality requires drawing a line around the network. If you exclude external partners, informal advisors, or family members from your data, your calculations for who is "central" might be completely wrong because the measure is sensitive to the network's defined borders.

Practical Implications

• In HR: Don't overlook the quiet employee who connects silos just because they have fewer meetings.
• In Marketing: Don't just pay the influencer with the most followers; find the one connected to diverse, non-overlapping communities to maximize unique reach.

Centrality measures are not universal indicators of importance but are strictly tied to the assumed process of flow across the network.

• Borgatti (2005): Centrality measures imply specific flow characteristics; for example, Betweenness assumes traffic travels on the shortest path (like packages), which fails for processes like gossip that wander randomly [1].
• Landherr et al. (2010): The "Boundary Specification Problem" reveals that missing data or incorrect network boundaries can radically alter centrality rankings, making them unstable in noisy datasets [2].
• Costenbader and Valente (2003): Sampling errors in network data have non-linear effects, where missing even a few central nodes can cause the calculated centrality of peripheral nodes to fluctuate wildly [3].

Static Bias: Most measures analyze a static snapshot, ignoring that relationships change effectively instantly in digital networks.

Node Homogeneity: Measures often treat all nodes as identical processing units, ignoring individual capacity, intelligence, or willingness to transmit, which often matters more than structural position.