Primary and secondary associations: meaning and application

You ask people what a topic brings to mind and get a list of words back. Those first words show what surfaces immediately, but not which meanings lie behind them. That is why you distinguish between primary and secondary associations: two layers in the same association network.

A primary association is an association at the first level: what someone names straight away for a topic. A secondary association is what that primary association in turn calls up when you present it as a new starting point, at level 1 and possibly level 2.

This Neurofactor working concept lets you read an association network in layers: which meanings are immediately available and which underlying meanings they carry. The level reflects the route by which an answer was elicited, not a ranking of importance or strength.

An association is a link in memory between a topic and something else. Neurofactor uses two working concepts to order those links in layers.

• Primary association: an association at the first level. It is what someone names straight away when asked about the topic.
• Secondary association: an association that a primary association in turn calls up. The primary association becomes a new starting point. Level 1 sits directly below a primary association, level 2 below a level 1 secondary association.

The distinction concerns an answer’s position in the elicitation, not the word itself. ‘Stress’ can be a primary association for one respondent and a secondary association under ‘deadlines’ for another. Each answer is a node traceable to the starting point that called it up.

Primary and secondary associations are Neurofactor working concepts. They are not a validated scientific classification, but build on a long tradition of association research.

In a free association task, participants name the first word that comes to mind for a presented word. In the University of South Florida norms, forward strength is the proportion of participants giving a particular response to a word (Nelson, McEvoy & Schreiber, 2004). The English Small World of Words norms asked for three responses, each to the same word (De Deyne et al., 2019).

That associations call up further associations fits network models of memory. Keller described brand knowledge as a network in which spreading activation from node to node determines retrieval, citing Collins and Loftus among others (Keller, 1993; Collins & Loftus, 1975). These models are context; they do not prove that a primary association sits above a secondary one in memory.

Primary associations show which meanings are most readily available.

One answer per respondent mainly captures the strongest links. De Deyne and colleagues note that single-response frequencies are reliable only for very strong associates, while weaker responses are unreliable or missing (De Deyne, Navarro & Storms, 2013). Several answers to the same topic give a more varied picture, but remain primary associations: they refer directly to the topic.

An example from the Small World of Words norms: for the word brewery, beer accounted for 85 per cent of first responses. A dominant response can block other strong associations; second and third responses help reveal the full distribution (De Deyne et al., 2019).

A primary association has meanings of its own. Two people naming ‘freedom’ may mean different things. Secondary associations reveal this: you present the primary association again and ask what it brings to mind.

This is a deliberate chaining step: one layer’s answer becomes the next layer’s starting point. That differs from continued free association. In the Small World of Words study, participants were told to respond only to the displayed word, not to their previous response; even so, the researchers found evidence of occasional chaining (De Deyne et al., 2019). So separate the two layers explicitly, or they will mix unnoticed.

Secondary associations show which underlying meanings a primary association carries: positive and negative sides or concrete situations. They do not tell you why someone considers something important.

Laddering is an interviewing technique that also works in layers. Reynolds and Gutman (1988) described how an interviewer asks ‘Why is that important to you?’ to move from product attributes to consequences and personal values. It builds on the means-end chain model, which links perceived product attributes to values (Gutman, 1982).

Laddering asks for reasons, from concrete to abstract. A secondary association asks what something brings to mind, without that direction. If you combine both, do not record answers to why questions as secondary associations.

Here, ‘secondary association’ means the Neurofactor working concept, not Keller’s secondary brand associations. An association’s position in the whole network is beyond this article’s scope.

1. Define the topic and the choice context. State precisely what you present and which choice the results should inform. ‘Working from home’ may mean something different in a policy discussion than in a job advert.
2. Ask for the primary associations. For example: ‘What does working from home bring to mind?’ Decide in advance on one answer, a fixed number or a free list, and instruct respondents to answer the topic, not their previous answer.
3. Choose which primary associations to explore further. Let respondents explore their own primary associations, or give everyone a fixed set from an earlier round. The second route makes comparison easier.
4. Ask for the secondary associations. Show the primary association as a new starting point: ‘What does flexibility bring to mind?’
5. Decide on the depth. Only go to level 2 if your research question requires it; each layer costs attention and drifts further from the topic.
6. Measure additional properties separately. Use a separate question or task for strength or valence; see How do you measure associations?.

How many associations you elicit is a design choice per study; there is no fixed rule. Weigh breadth against respondent burden and the number of answers per primary association.

You must be able to trace which answer belongs to which starting point. For each answer, record at least:

• a pseudonymised respondent code;
• the topic and the choice context;
• the level: primary, or secondary at level 1 or 2;
• the starting point the answer responds to;
• the position in the order within that starting point;
• the verbatim text and the coded category.

Merging synonyms is an interpretive decision; keep it traceable. Do not combine primary and secondary associations in one frequency list: their starting points differ. Report them by layer, and secondary associations always per primary association.

What is named first is usually highly accessible. The proportion of participants giving a response shows how dominant it is in a group, like forward strength in free association norms (Nelson et al., 2004).

Yet the first association is not necessarily the most important. A first answer can be a cliché or a socially easy answer, while a later or secondary association weighs more in a choice. How quickly and often something comes to mind is covered under association salience. Measure link strength separately; see association strength.

The level describes the route of an answer, not a ranking of importance. A secondary association is not automatically weaker.

This example is fictional; it contains no research results.

Situation. An insurer wants to revise its hybrid working policy. Employees react differently to proposed fixed office days.

Choice question. Which meanings do employees associate with working from home?

Approach. Employees first name three primary associations for ‘working from home’. They then explore two of their own primary associations at level 1.

Possible interpretation. At the first level, ‘flexibility’ seems clearly positive. The secondary associations show two possible meanings: planning your own day, but also constant availability. A message that promises flexibility can thus evoke both freedom and pressure.

Next step. Test whether this split is widespread and relates to reactions to fixed office days. Until then, it is a hypothesis.

• Recording the second or third answer as a secondary association. An extra answer to the same topic remains a primary association.
• Equating the first association with the most important. Accessibility is not importance; measure importance and strength separately.
• Putting all layers in one word cloud. That erases the layering you wanted to see.
• Reading secondary associations as reasons. ‘Always online’ under ‘flexibility’ shows what someone thinks of, not why they want to work from home.
• Probing too deep. At level 2 or beyond, answers often drift from the topic.
• Adopting a fixed number as a rule. The number depends on your research question, target group and time.
• Keeping only coded categories. Without verbatim text, coding decisions cannot be checked or revised.

The distinction is a practical way to order association data, not a validated model of how memory stores meaning. Network models offer a plausible framework, but the layers arise in your elicitation, not in the brain.

Different wording, order or numbers of answers produce different layers. Patterns apply at group level and do not predict individual thinking or behaviour. And a secondary association does not explain behaviour: that a primary association calls something up does not mean it drives the choice. That requires further research.

Primary associations show what surfaces first; secondary associations show which meanings they carry. The distinction only works if you maintain it consistently in elicitation and data. Read the level as a route, not as importance, and measure strength and valence separately.