13 Research methods, statistics, critical review and evidence-based practice13.13 Network analysis
13.13.1 Nodes, edges and network graphs
3 min read1 figure
Written by Resident Doctor Committee · Published
New in the 2027 syllabus
The College added this topic for exams from January 2027.
What a network represents
A node (also called a vertex) is one entity or variable, such as a person, a symptom or a brain region.
An edge (also called a link) shows a relationship between two nodes, such as social contact or a statistical association.
A network graph shows the whole set of nodes and their edges. What it means depends on what the nodes and edges stand for in that study.
Research setting | Nodes | Edges |
|---|---|---|
Social network | People | Friendships or contacts that people report |
Symptom network | Symptoms or questionnaire items | Estimated statistical relationships |
Brain network | Brain regions | Anatomical connections or functional associations |
Types of network
Direction, weighting and sign are different properties of the same network. One network can be, for example, undirected, weighted and signed at once.
Property | Meaning | Example |
|---|---|---|
Undirected | The edge has no direction | A correlation between two symptoms |
Directed | The edge points from one node to another | One person names another as a friend |
Unweighted | An edge is simply there or not there | Whether two people have contact |
Weighted | Each edge has a number attached | The strength of a partial correlation |
Signed | Edge values can be positive or negative | Symptoms rising together, or one rising as the other falls |
A directed edge does not on its own prove causation. What it means depends on the study design and the model.
A simple undirected graph has no edges looping back to the same node and no repeated edges between the same pair of nodes.
An adjacency matrix is a table that records the connections. Its rows and columns both list the nodes. An undirected graph has a symmetric matrix, because the link from A to B is the same as the link from B to A.
Reading a symptom network
A partial correlation network shows the association between each pair of variables after conditioning on (taking account of) the other variables included.
If no edge is estimated between two symptoms, that does not prove they are unrelated in every setting, or that neither can affect the other.
A cross-sectional network uses data from one time point. Its undirected edges alone cannot show which symptom came first.
For example, an edge between insomnia and low mood may reflect an association that remains after allowing for the other measured symptoms. It does not show that treating insomnia will necessarily get rid of the low mood.
Reading the picture
The legend tells you whether line colour shows sign, line thickness shows weight, and node colour shows a group.
The layout only arranges the nodes so the graph is easy to see. In the usual force-directed drawing, how close two nodes sit on the page is not a numerical measure of association.
A node near the middle of the page is not necessarily the most central node by a calculated network measure.
Where two lines cross, there is no extra node unless the graph clearly marks one there.
Comparing plots needs consistent legends and scales. In two separately scaled graphs, a thicker line in one may stand for a smaller weight than a thinner line in the other.