Closeness Centrality

A network represents people or places joined by links. When there are at least two people, everyone can reach everyone else, and links work both ways, higher closeness centrality means a shorter average route to others. This alone does not guarantee faster communication. Draw four people and their links, then count the fewest steps between each pair.

A graph represents a network using nodes, such as people, and edges, the links between them. A shortest-path distance is the smallest total distance along any route between two nodes. In a connected graph, everyone can reach everyone else; in an undirected graph, links work both ways. For such a graph with at least two nodes, normalized closeness centrality is one divided by the average shortest-path distance to all other nodes. Higher scores mean shorter average distances. [2]

The unnormalized version instead divides one by the total of those distances. Consider a hypothetical four-person network where the CEO, HR Coordinator and Intern each link only to the IT Manager. Each link counts as one step. IT’s distances to the others are one, one and one; everyone else’s are one, two and two. IT therefore has the highest closeness because its average distance is smallest.

Degree centrality counts direct connections. Betweenness centrality measures how often a node lies on shortest paths between other nodes. Closeness measures distance to others. These describe different structural properties: having the most direct connections need not give the shortest average distance. A person connecting separate social groups raises a question about betweenness as well as closeness. [1]

The workplace example is a hypothetical analogy, not evidence about real IT managers or CEOs. Job titles, weekly contact and departmental coverage cannot establish closeness without the actual network. Likewise, assessing a transportation hub requires routes and distances. Short average distances alone do not establish the fastest rumor spread, hidden influence, bottlenecks or better decisions from appointing someone as a coordinator.

Weighted graphs assign numbers to links. For closeness, those numbers must represent distances or costs, and shortest routes minimize their total. Directed graphs have links that can work in only one direction, so specify whether distances run toward a node or outward from it. NetworkX, a graph-analysis software library, uses incoming distances by default. [2]

In a disconnected graph, some nodes cannot reach others, so classical distances to all nodes include infinity. A practical variant needs an explicit rule. NetworkX calculates distances within reachable parts and, by default, scales scores by the fraction of the whole network that is reachable. State the convention when interpreting scores. [2]

• Conceptual theory: Freeman distinguishes degree, betweenness and closeness as separate structural conceptions, associated respectively with activity, potential control of communication and independence from intermediaries. These are conceptual interpretations rather than proof of workplace outcomes. [1]
• Implementation guidance: NetworkX defines normalized closeness using the reciprocal of average shortest-path distance. In a connected, undirected graph with at least two nodes, this uses distances to every other node; higher scores indicate shorter average distances. [2]
• Implementation guidance: NetworkX supports distance-valued link weights and directed graphs, using incoming distances by default. For disconnected graphs, its default convention scales reachable-part closeness by the fraction of the network reachable. [2]
• Empirical evidence reviewed by Freeman: The paper describes earlier group experiments as producing confusing and often contradictory findings and calls for further empirical work on the different conceptions of centrality. It does not establish guaranteed improvements from assigning coordination roles by closeness. [1]

Closeness depends on the network and distances chosen. Job titles and contact frequency alone do not determine its value. A distance-based score does not by itself establish actual communication speed, influence or organizational performance. Weighted links require meaningful distances or costs; directed links require a stated incoming or outgoing interpretation. Disconnected graphs require an explicit convention for unreachable nodes. The employee, social-group and transportation examples are hypothetical teaching analogies, not empirical findings.