Few ideas attract as much confusion. On one side, people convinced a score sums a person up; on the other, people convinced it measures nothing at all. Both are wrong, and the gap between them is worth closing.
What the test does measure
An IQ test assesses a set of reasoning abilities under standardised conditions: extracting an abstract rule, transposing it to a new case, manipulating relations between symbols, holding several constraints at once, under time pressure.
The empirical finding that founds the discipline is that these abilities, measured separately, correlate positively: people who do well on number series tend to do better than average on verbal analogies, and the reverse. That regularity, first observed by Charles Spearman at the start of the twentieth century, is what is called the general factor, written g.
g is not a substance in the brain. It is a statistical regularity: the fact that these performances vary together. Its reality as a regularity is not disputed; what exactly it represents is disputed a great deal.
What it does not measure
The list is long and worth spelling out, because it is where the misunderstandings live.
Creativity. A test has one right answer per question by construction. It cannot assess the ability to invent a new one.
Knowledge. A well-built test deliberately minimises this — the whole point of matrices and series.
Practical judgement. Whether to change job, how to defuse a conflict, whether to trust someone. None of it appears in a mental rotation item.
Persistence and conscientiousness. These predict educational and occupational outcomes at least as well as cognitive ability. They belong to other instruments.
What distorts a measurement
A score depends on ability, and on a good deal else.
Practice. Retaking a test of the same type typically gains several points on the second attempt. No ability has changed: the format has been learned.
State on the day. Tiredness, stress, performance anxiety, interruptions. The same person can vary by ten points from one day to the next.
Language and culture. A verbal item taken in a second language systematically understates. Even so-called culture-fair items assume familiarity with the codes of testing — answering multiple-choice questions against a clock is itself a learned skill.
The Flynn effect. Raw scores rose by around three points a decade for much of the twentieth century in industrialised countries, forcing periodic restandardisation. A number only means something relative to a population and an era.
So what is it good for?
Clinically, for identifying particular cognitive profiles: a marked gap between indices, a mismatch with observed school performance, a specific difficulty. There the number is a starting point for a conversation, not a conclusion.
In research, for studying tendencies at group scale — which these measures do properly, unlike what they do at individual scale.
And on a site like this one, for satisfying a legitimate curiosity: knowing roughly where you sit on a given kind of task. A modest use, and the one the tool can actually support.
To see how the number is produced here, our method page sets out the model and its limits.