Node in an association network: meaning and application

A network diagram can look persuasive: words inside circles, thick lines and a few prominent nodes. What does each element represent? Without that explanation, a visualisation can be mistaken for an explanation. A node becomes useful research information when you know its underlying data, why it is represented separately and what its connections mean.

A node in an association network is a defined unit, such as a concept or coded meaning, used to represent relationships in research data. It belongs to a model. It is not a brain region or neuron and does not prove that an association causes behaviour. Explain nodes, connections and visual properties separately.

Network models represent units and their relationships. Collins and Loftus helped develop their application to semantic processing. Their account connects concepts; a contemporary diagram of customer responses is not automatically the same model or a neurobiological measurement. Collins and Loftus, 1975

A simple map may contain three layers: original statements, coded concepts and a graphical representation. Keep those layers identifiable. Translating ‘I want someone I can call’ into ‘accessible support’ is an analytical decision. It may be useful, but the coded node is not the participant’s original statement.

Choose the level in advance. A node can represent a word, a clearly defined meaning or an overarching theme. Do not silently mix these levels. A diagram containing ‘price’, ‘trust’ and ‘my manager never responds’ otherwise places an attribute, a broad concept and a particular experience alongside each other.

A codebook makes the choice auditable. Record each node’s definition, inclusion criteria, exclusions and examples. Give it a stable identifier as well as a readable label. The same node then remains identifiable when its wording is clarified or translated. This also helps collaborators discuss a disputed category without confusing a changed label with a changed definition.

Position on the screen also requires explanation. Software may place connected nodes close together to make the map readable. This does not necessarily represent independently measured psychological distance. State when placement comes from a layout algorithm and when distance represents a specifically defined measure.

You might record a relationship when someone responds B to cue A. Alternatively, you might connect concepts appearing in the same interview. These are different rules. The first relationship may be directed; the second may describe co-occurrence within a chosen unit of analysis.

Small World of Words provides an example of systematically collected word relationships. It helps explain a data structure; it does not give every custom network map the same evidential standing. De Deyne et al., 2019

Explain how missing or rare relationships are treated. An absent line may indicate an unobserved relationship, a relationship that was never investigated or an edge hidden by a display threshold. Readers need to know which interpretation applies. Preserve the underlying data even when simplifying the visual presentation.

Centrality has different definitions. Freeman distinguished conceptions involving many connections, positions between other nodes and short paths to other nodes. A centrality score therefore requires a specified measure and network for interpretation. Freeman, 1978/1979

Do not translate a high score directly into a campaign priority. A broadly coded node may have many connections because different responses were combined. Check whether its position survives defensible alternative coding and whether the concept matters in the decision being investigated. Structural prominence and practical importance are separate claims.

A fictional employer investigates reactions to an additional mentoring role. An initial map contains one large ‘support’ node. Reviewing responses reveals different meanings: allocated time, access to an experienced colleague and a clear escalation contact. This example contains no research findings.

The team codes the three meanings separately and retains ‘support’ as an overarching theme. It then examines whether the same participants mention several meanings and which conditions they attach to participation. The revised map is more precise even if it looks less simple.

The recommendation becomes ‘test whether a concrete explanation of time allocation and escalation arrangements answers the remaining questions’. It does not become ‘strengthen the support node’. The former identifies an actionable proposal and makes clear which part follows from the data and which part still requires testing.

1. Define the research question and unit of analysis.
2. Specify what nodes and connections represent.
3. Create a codebook with examples and exceptions.
4. Retain links to original responses.
5. Document direction, weights and thresholds.
6. Provide a legend for size, colour and position.
7. Check sensitive interpretations through alternative coding or additional data.

Preserve the map’s version as well. Splitting a category into two nodes can change counts and network measures. Do not compare the new version as though only participants’ views had changed. In repeat studies, disclose whether the codebook and calculations remained consistent. A visually similar map can still rest on a different analytical definition.

Include nodes that help substantiate the profile, together with their meaning, evidence and uncertainty. An associative target group concerns shared relevant meanings and considerations in a particular choice situation. Identical words on a map do not prove that participants make decisions in the same way.

Use the network to explore patterns and explain findings. It can structure discussion of differences, but it cannot replace explanation. Ask a colleague to restate the legend and an important conclusion in their own words. If that interpretation exceeds the data, revise the presentation rather than relying on a small footnote to correct the impression.

• Treating broad themes and specific statements as equivalent nodes.
• Leaving direction or display thresholds unexplained.
• Interpreting central position as demonstrated causal influence.
• Hiding small samples behind prominent graphical shapes.
• Comparing different codebook versions without qualification.
• Presenting an association map as a brain scan.

A node’s value lies in the auditable connection between meaning and data. Define the level, explain the relationships and document the visual choices. The map can then help readers understand the research and formulate focused questions for the next study.