Can the average be fractional?

Yes. It is a mean of degree values and need not be an integer.

Does it require a connected graph?

No. Each node's local neighborhood can be evaluated within a graph containing several components.

Why can my neighbors seem more connected than I am?

Connections tend to encounter high-degree nodes more often, but the pattern and its strength depend on the graph and the way nodes are sampled.

AI Glossary / Graphs and symbolic AI / Intermediate

Average neighbor degree

Definition

Average neighbor degree is the mean number of connections held by a node's immediate neighbors. It describes the connectivity of the surrounding nodes, rather than the number of connections held by the focal node itself. The source spelling uses “neighbor.”

What it means

Two nodes can each have three neighbors but occupy very different structural surroundings. One may connect to three lightly connected endpoints, while the other connects to three busy hubs. Their own degrees are equal, but their average neighbor degrees differ.

The measure is local: it examines immediate neighbors and their degrees in the represented graph. It is not the average distance to all other nodes, a count of distinct second-hop contacts or a direct measure of influence.

Interpretation depends on what an edge represents. A supplier linked to diversified partners may face different dependencies from a supplier linked to specialized partners, but degree alone does not measure capacity, reliability or the importance of any relationship. Those require additional evidence.

How it works

In a simple undirected, unweighted graph, list the focal node's neighbors. Find each neighbor's degree in the full graph, add those degrees and divide by the number of neighbors. The edge connecting a neighbor back to the focal node is included in that neighbor's degree.

Repeat this calculation for each node if a node-level map is needed. A separate operation can group the values by the focal nodes' degree classes, producing average degree connectivity. Keeping those output levels distinct prevents accidental comparisons between individual nodes and group summaries.

Directed graphs require a choice of incoming or outgoing neighborhood, together with the type of degree measured for those neighbors. A weighted version gives stronger edges more influence on the mean. Record these choices explicitly. A node without neighbors has no ordinary arithmetic mean for this quantity; software may return a convention such as zero, which should not be mistaken for a measured average over actual neighbors.

Three practical examples

A knowledge-base team distinguishes an article linking to broadly connected reference pages from one linking mainly to isolated notes.
A partnership analyst describes the immediate network surroundings of organizations with the same number of direct partnerships.
A graph-learning workflow includes local neighborhood connectivity as a candidate feature, then evaluates whether it contributes useful predictive information for the specific task.

Think of it this way

Think of asking each member of your small contact list how many contacts they have, then averaging their answers. Your list length and their average list length answer different questions.

The analogy does not imply that you can reach every contact they know, or that all relationships are equally strong. Overlapping contact lists and access restrictions can make a simple connectivity summary very different from practical reach.

Worked example

A local calculation and a weighted alternative

Suppose node P has two neighbors, Q and R. Q has degree 2 and R has degree 6. P's own degree is 2, but its average neighbor degree is (2 + 6) / 2 = 4.

Now assign the P - Q edge weight 3 and the P - R edge weight 1, treating the weights as nonnegative relationship strengths. The weighted neighbor-degree mean is (3 × 2 + 1 × 6) / (3 + 1) = 3. The stronger connection to the lower-degree neighbor pulls the average downward.

This calculation weights the neighbors' ordinary degrees; it does not replace them with weighted degrees automatically. It also does not say P has four distinct second-hop neighbors. Q and R may share contacts, and their degree counts include their links back to P.

Illustrative case study

Fictional case: Elm Knowledge analyses links among employee help articles. Two pages each contain four outgoing links, but one points to general indexes while the other points to focused instructions. Their average neighbor degrees are very different.

The team first assumes that the page linked to indexes must be more useful. User testing reveals that employees performing a specific task often prefer the direct instructions. The analysts retain the graph statistic to describe navigation structure and evaluate usefulness separately through task completion and content review.

The case illustrates a common boundary: a higher neighborhood-connectivity value can reflect broad navigation without indicating better support for a particular user need. The intended task determines how the measure should inform decisions.

Common misconceptions and mistakes

It is simply another name for the focal node's degree. The calculation uses the degrees of its neighbors.
It counts unique people reachable in two steps. Shared neighbors and return edges make that a different calculation.
A high value proves influence. Influence depends on the process operating on the graph, not just local degree counts.
Direction does not matter. Incoming and outgoing relationships can represent entirely different roles.
A weight can be used without interpretation. Frequency, strength, cost and distance are not interchangeable meanings.

Business use

Average neighbor degree can enrich network descriptions and help identify structurally different nodes that a simple degree ranking treats as equal. It is useful for exploration, comparison and feature engineering when its limits are clear.

For business reporting, explain the node and edge definitions next to the metric. Use a consistent observation window and distinguish missing relationships from confirmed absence. If weights represent costs or distances, consider whether giving larger weights more influence matches the business question. Evaluate any predictive or operational use against relevant outcomes rather than treating the statistic as an automatic quality score.

Check your understanding

Node T has degree 3, and its neighbors have degrees 2, 5 and 8. What is T's average neighbor degree, and does that change T's own degree?

Answer: The average is (2 + 5 + 8) / 3 = 5. T still has degree 3; the value 5 describes its neighbors' connectivity.

Related terms

Learn next

Draw a graph with two nodes of equal degree but different neighborhoods. Calculate their average neighbor degrees and list their distinct second-hop neighbors to see why these are different quantities. Then add edge weights with a clearly stated meaning and compare the weighted and unweighted results.

Sources

NetworkX documentation, average_neighbor_degree, provides the node-level formula, weighting convention and directed options: https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.assortativity.average_neighbor_degree.html

NetworkX documentation, average_degree_connectivity, explains the related degree-class summary: https://networkx.org/documentation/stable/reference/algorithms/generated/networkx.algorithms.assortativity.average_degree_connectivity.html

The numerical example and fictional knowledge-base case are original illustrations.

Previous: Average degree connectivity Next: Average node degree