The lexical hypothesis: how dictionaries gave us the Big Five

What is the lexical hypothesis?

The idea is simple: if a difference between people is common enough to talk about, a language will eventually settle on a word for it. Reliable people get a word, so do careless ones, so do people who enjoy company and people who avoid it. Language is a working record of what a community of speakers has found worth noticing across generations of describing each other, which makes the dictionary a reasonable place to start.

The trait concept behind this has a long history, and the conventional citation for it is Allport's (1937) monograph, the foundational statement of a trait-based account of personality. If traits are real and language encodes them, the structure hiding inside a trait vocabulary should resemble the structure of personality itself, a testable claim rather than a slogan.

How did researchers turn word lists into five factors?

Turning a claim about language into a claim about personality took a specific method. People rated themselves or each other on large sets of trait adjectives, and the answers were run through factor analysis, a technique that groups items which move together into a smaller number of underlying dimensions. If reliable, dependable and careful rise and fall together across raters, they point to one factor rather than three.

The clearest account is Goldberg's (1990) three-study programme. First, 1,431 trait adjectives were sorted into 75 clusters, and ten analyses, each using a different factor-analytic procedure, recovered virtually the same five-factor structure. Second, 479 common trait terms reduced to 133 synonym clusters produced the same five factors again, in self-rating and peer-rating samples, with no sixth factor generalising across them. A third study distilled the material into 100 clusters from 339 trait terms, as candidate markers. The table below lays out that narrowing.

What are the Big Five markers, and where do they come from?

A factor found in one paper is not yet usable by anyone else. Goldberg (1992) built adjective sets other researchers could reuse, testing them with more than 1,000 college students until a set of 100 unipolar terms proved robust across self-descriptions and peer descriptions alike, becoming a standard adjective measure of the five factors.

A hundred words is still a lot to rate. Saucier (1994) scrutinised those markers across 12 data sets and settled on the 40-item Mini-Markers: fewer difficult words, lower correlations between the scales, and reliabilities somewhat lower than the full version, convenient at a cost. Personality adjectives works through those marker words, grouped by dimension.

Did the rest of personality psychology agree on five factors?

One laboratory finding the same structure ten times is convincing. A field agreeing on it is a different matter, and personality psychology spent much of the twentieth century without agreement on how many trait dimensions there were. Digman's (1990) review is the standard reference for how a five-factor account came to be seen as the field's emerging consensus.

Goldberg (1993) tells the same convergence from the inside, as a history of the taxonomy rather than a new study, tracing the idea through Galton, Thurstone, Cattell, and Tupes and Christal. He argues the acceptance owes a great deal to its critics: each attempt to replace the model failed to dislodge it. The Big Five, on that account, is what survived decades of people trying to prove it wrong.

Have lexical studies ever recovered a different number of factors?

Yes, and this is a strength of the method, not a weakness in the five-factor account. Ashton and colleagues (2004) ran standard psycholexical studies in seven languages, Dutch, French, German, Hungarian, Italian, Korean and Polish, and recovered a similar six-factor solution across them: variants of extraversion, agreeableness, conscientiousness, an emotionality factor, an intellect or imagination factor, and Honesty-Humility. HEXACO agreeableness and emotionality are described as rotated variants of Big Five agreeableness and neuroticism, the same variation on slightly different axes.

That does not overturn the five-factor finding. Seven languages produced a six-factor solution, not proof every language does the same. The HEXACO model covers what the sixth factor adds and how it is measured.

What the lexical hypothesis does not claim

• It does not claim a language's trait vocabulary is unbiased. Some directions attract flattering words and the other unkind ones, a fact about speakers, not the trait.
• It does not claim five, or six, is the only number a language could return. A different vocabulary can recover a different number, as the six-factor work above shows.
• It does not claim a handful of adjectives is a good way to measure someone. The marker sets validate a structure across many people, not score one person from a checklist.
• It does not claim that people sort into five or six types. The dimensions are continuous, and most people sit near the middle rather than at either end.

How does Wellington measure the traits this history uncovered?

Wellington is a science-based personality assessment from Therabot Labs LLC. It measures personality as continuous traits, built on the Big Five and HEXACO models, and writes the results back as a personal report rather than a type.

The free Snapshot asks 60 questions, takes about 7 minutes, and covers six dimensions, the five this history produced plus Honesty-Humility: Extraversion, Agreeableness, Conscientiousness, Emotional Stability, Openness to Experience, Honesty-Humility. It does not hand you a list of adjectives; each question is a statement about behaviour, rated on a five-point accuracy scale, steadier than picking a word off a list. The first ten questions are answered before anything else is asked; after question ten you sign in with an email so your answers are saved and your results can be shown.

Every trait is reported as a percentile against the reference sample. The three bands (Quiet below the 30th percentile, Balanced to the 70th, High above) and the Portrait names exist to make the result easier to talk about; the percentile is the measurement, and it is always shown alongside. Norms are provisional until the norming sample is complete. The Wellington Membership, $4.99 a month, adds 336 further questions and reaches 90 traits, including every Big Five facet and all 24 VIA character strengths, then carries the method toward 255 traits. Wellington is a wellness tool for reflection, not a medical or psychological diagnosis, and it does not replace care from a qualified professional. Not for hiring; data can be exported or deleted at any time.

Questions people ask

Who came up with the lexical hypothesis?

No single person. Goldberg's (1993) history of the idea traces it through Galton, Thurstone, Cattell, and Tupes and Christal, with Allport's (1937) monograph as the standard citation for the trait concept it depends on. The five-factor programme people usually mean by the term is Goldberg's work from the 1980s and 1990s.

Is the lexical hypothesis the same thing as the Big Five?

No. The lexical hypothesis is the assumption that a language's trait words track real differences between people. The Big Five is a finding produced by testing that assumption: five broad dimensions recovered from English trait vocabularies (Goldberg, 1990). A different vocabulary can return a different number, as a six-factor solution has elsewhere (Ashton et al., 2004).

Does the lexical hypothesis work in languages other than English?

Ashton and colleagues (2004) analysed lexical studies in Dutch, French, German, Hungarian, Italian, Korean and Polish and recovered a similar six-factor structure across all seven, with Honesty-Humility as a separate factor.

Why did researchers shorten the original word lists?

A shorter list is easier to use. Goldberg (1992) built a 100-word adjective set that held up across self-descriptions and peer descriptions, and Saucier (1994) tested it across 12 data sets to find a shorter 40-word version, more convenient at the cost of somewhat lower reliability.