Five Factors, Found Repeatedly

The method that produced the Big Five is unusual and is most of the reason to take them seriously. It did not start from a theory of personality. It started from a dictionary.

The lexical hypothesis holds that differences between people important enough to matter, over enough time, get encoded in ordinary language. If so, the structure of trait vocabulary is evidence about the structure of the underlying differences. So: extract every personality-describing word, have large samples rate themselves and each other, factor-analyse the result, and see what falls out.

What falls out is five dimensions: openness, conscientiousness, extraversion, agreeableness and neuroticism. The same five have emerged from different languages, different instruments and different decades, which is the strongest thing that can be said in their favour: independent lexical studies in unrelated languages recovering broadly the same structure rules out the possibility that it is an artefact of English or of one research tradition. The convergence is imperfect: some languages yield a sixth factor around honesty and humility, and the fit varies. That is the normal texture of a real but approximate finding.

They are dimensions rather than types, which matters more than it sounds. Everybody has a position on each, and the positions are continuously distributed. That is the basis of the profession's dislike of type indicators: carving a continuum into categories throws away most of the signal and makes anybody near the middle unstable between administrations. Retest studies of popular type instruments find substantial proportions of people receiving a different type weeks later, which follows mechanically from cutting a continuous variable in two. The underlying dimensions such instruments measure are often perfectly reasonable; the categorisation is the problem, and the categorisation is what gets used.

The most important caution is about what the factors are. They describe how behaviour clusters. They do not explain it, and it is very easy to slide from one to the other. Saying somebody is reliable because they score highly on conscientiousness is circular, since the score was derived from ratings of reliability in the first place. The genuine finding is that behavioural tendencies cluster in five broad ways across people and situations. What produces the clustering (temperament, learning history, environmental niches, or some combination) is a separate and much harder question that the factor structure does not touch.

Which is worth remembering when an organisation assigns every employee to one of sixteen boxes and starts building teams from the result. The objection is not primarily moral. It is that if a meaningful fraction of staff would be assigned a different box on retest a month later, any decision made from the label is partly noise, and the confidence with which the labels get used tends to be inversely related to how much anybody has looked at the reliability figures.