Happier people’s minds seem to return to friends before money — across the US, UK, and South Korea, their spontaneous thoughts were more closely tied to friendship, and those thoughts predicted when they would choose relationships over financial gain

A job offer can be generous on paper and still ask for something the salary does not replace.

That tension sits at the centre of a 2025 study connecting three parts of experience that are usually measured separately: self-reported happiness, the concepts that come readily to mind, and a hypothetical decision between close relationships and financial opportunity.

In an online sample from the United States and United Kingdom, people with higher subjective wellbeing generated chains of words that were semantically closer to “friend”. Their responses did not show the same consistent relationship with “money”. A separate dataset from South Korea reproduced the friendship pattern using a related task.

A few days later, participants from the English-speaking sample faced a scenario about a lucrative job in another country. Those whose earlier word chains sat closer to “friend” said they would require a larger salary increase before accepting a move that would cost them regular access to close relationships.

The sequence gives the finding its appeal: what came to mind first appeared to anticipate what people later said they valued. But the evidence is narrower than a direct recording of ordinary mind-wandering, and the choice involved imagined salary rather than a real employment decision.

“Spontaneous” meant a timed chain of associations

The study was published in Communications Psychology in November 2025 by Won-Gyo Shin and colleagues. Its main sample contained 210 native English speakers aged 18 to 35 from the US and UK, recruited through Prolific in September 2023.

Eligibility required participants to report no current mental health issues. The sample included 114 men and 96 women, with a mean age a little above 28. Most participants identified as White.

They first completed questionnaires measuring positive affect, negative affect and life satisfaction. The researchers standardised and combined those scores into a measure of subjective wellbeing.

Next came the Free Association Semantic Task, or FAST, a method developed in earlier research on self-generated conceptual chains. Participants began with three seed words: “key”, “friend” and “money”.

Starting from each seed, they typed a word or short phrase that came to mind, then responded to the concept they had just generated. They had seven seconds for each answer, plus four extra seconds after a warning if needed. Thirty concepts for each of the three seeds produced 90 responses per participant.

The task captures concepts that are easy to retrieve under time pressure. It is not a passive recording of thoughts during an ordinary afternoon. People deliberately responded to prompts while sitting inside an online experiment, so “spontaneous” refers to the associative flow within that structure.

A language model turned words into distances

A list of responses such as “dinner”, “laughter”, “holiday” and “support” may feel intuitively social, but the researchers needed a repeatable way to compare thousands of words and phrases.

They used Word2vec, a model trained on part of the Google News corpus. It represents each term as a vector with 300 numerical dimensions. Words that appear in similar linguistic surroundings are positioned more closely in that mathematical space.

The team then used two main measures. Cosine similarity estimated the angle between the response vectors and the target concepts “friend” or “money”. Word Mover’s Distance estimated how far one set of words would need to move through semantic space to resemble the target.

A third measure examined convergence. Rather than asking only where a person’s 30-word chain sat on average, it tested whether the sequence moved progressively closer to “friend” or “money” as the trials continued.

These methods do not understand a participant’s private meaning. A news-trained vector model infers similarity from patterns in language use. It can detect that some concepts are usually discussed near friendship, but it cannot know whether “home”, for one person, evokes warmth, conflict or both.

That is why the study is best read as a map of semantic accessibility rather than a literal view into the whole mind.

Positive affect carried most of the friendship pattern

Participants with higher overall subjective wellbeing produced concepts with greater cosine similarity to “friend” and shorter Word Mover’s Distance from it. Their chains also tended to become more friend-related as the trials progressed.

Most of the money analyses did not show a significant relationship with wellbeing. The headline’s contrast comes from that asymmetry: friendship-related semantic patterns tracked happiness more consistently than money-related patterns did.

“Happier” still needs unpacking. When the researchers placed positive affect, negative affect and life satisfaction in the same models, positive affect was the component consistently associated with closeness to “friend”. Life satisfaction and lower negative affect were not.

In other words, the result was strongest for how frequently people reported positive feelings, not for every possible meaning of a happy life. Silicon Canals has previously examined why pleasant feeling and a deeper sense of meaning should not be collapsed into one idea. The new study gives another reason to keep the components visible.

One especially interesting exploratory result appeared when “money” itself was the seed. Participants with more positive affect still showed greater convergence towards “friend”. Even inside a chain initiated by money, their later concepts became more socially related.

That does not mean every happy participant literally thought about a friend before thinking about cash. The researchers analysed group-level variation in semantic similarity and movement, not a simple race between two words in each person’s mind.

The South Korean data replicated one half of the finding

To test whether the friendship association extended beyond the English-speaking sample, the researchers analysed an earlier South Korean dataset. It contained 350 participants drawn from two studies conducted in 2019 and 2020, with subjective wellbeing data available for 330.

The Korean participants were younger, averaging about 23, and many were college students. They had completed a related FAST exercise in Korean, but the seed words differed. The chains began with concepts including “family”, “mirror”, “tear” and “abuse”, and participants produced 40 responses per seed rather than 30.

The Korean concepts were translated into English before being converted into the Word2vec representation. Despite those procedural differences, higher subjective wellbeing was again associated with semantic proximity and convergence towards “friend”, but not “money”. Positive affect again accounted for the most consistent part of the pattern.

This is meaningful cross-cultural consistency, but it is not a complete identical replication. The seed words, language, ages and surrounding procedures differed. Only the US and UK group faced the choice between salary and relationships.

The two samples also differed in their overall money-related associations. The US and UK participants showed stronger links to money concepts than the Korean group, but culture cannot be isolated from age or task design here. The paper explicitly leaves that comparison for future research.

The financial choice was an imagined relocation

Within three days of the first session, 192 people from the main sample returned for a second online session. They read a scenario about an offer from a top-tier global company.

The role promised higher pay, regular bonuses and broad financial benefits. It also required moving to another country, an eight-hour flight from friends and family, with uncertain prospects of returning home. The scenario deliberately placed financial gain and close relationships in conflict.

Participants selected the salary increase that would make them willing to accept. Options rose in 20 per cent increments from 20 to 200 per cent, and they could also indicate that accepting remained unlikely at any offered increase.

People whose earlier concept chains were semantically closer to “friend” reported needing a larger increase before taking the role. The association remained after the models accounted for age, sex, education, subjective socioeconomic status and materialism.

The choice model was supported by leave-one-out cross-validation, a technique that repeatedly fits the model without one participant and tests its performance on the person left out. That helps assess whether the pattern depends too heavily on a few observations, but it is not a substitute for replication in a new decision sample.

The boundary is straightforward. Nobody negotiated with an employer, moved country or lost a friendship. The task measured a stated threshold inside a deliberately sharp hypothetical scenario.

Real offers include partners, children, visas, debt, career stage, hybrid work, the quality of existing relationships and the possibility of building new ones. A person may accept a move precisely to support the people they love. The study created a clean trade-off by leaving most of that complexity outside the frame.

The statistical pathway did not settle causality

Higher subjective wellbeing also predicted a greater preference for relationships over money in the dilemma. A mediation analysis found a small indirect path through friend-related semantic similarity: happier participants had more friend-adjacent concept chains, which in turn predicted a higher salary threshold.

Mediation describes how variables line up statistically. It does not establish that one caused the next, particularly when the proposed mediator was not manipulated.

Close relationships may generate positive emotion and make social concepts easier to retrieve. Positive emotion may broaden attention to other people and encourage investment in relationships. Personality, financial security or earlier social experiences could shape both. The processes may also reinforce one another over time.

Older research offers context without resolving the direction. In a 2002 comparison of 222 university students, the happiest group was highly social and reported stronger romantic and other relationships. No single variable was sufficient for happiness, and that study was observational too.

The 2025 study did not manipulate thoughts, improve relationships or track real career decisions. Its main data collection was not preregistered. It therefore cannot show that making a person think about friends would increase happiness, or that happier feelings directly cause somebody to reject money.

“Friends before money” is not a verdict against income

The absence of a consistent money association should not be read as evidence that money is irrelevant to wellbeing. Financial resources can create security, reduce stress, support family members and make time with friends easier to protect.

The paper itself found one limited exception: life satisfaction was positively associated with convergence towards “money” when money was the seed. The result did not repeat across the other money measures, but it fits the possibility that financial resources relate more strongly to how people evaluate their lives than to their day-to-day positive feelings.

There are broader measurement limits. “Friend” stood in for a social world that also includes family, romantic partners, neighbours and community. This matters especially across cultures where family may be more central than chosen friendship.

Both samples were young and demographically narrow. The main study excluded people reporting current mental health issues. Translation added another layer to the Korean analysis, and the English Word2vec model may not preserve every cultural meaning carried by the original words.

The authors have shared the data and analysis code through the Open Science Framework, making the semantic and statistical choices available for scrutiny and reuse.

The careful conclusion is not that happy people disregard money. It is that, in these timed conceptual chains, happier participants, especially those reporting more positive affect, occupied a more friendship-related semantic landscape. Among the people later asked to make an imagined trade-off, that landscape also predicted how much financial gain it would take to pull them away from close relationships.