Reasoning and age: what changes and what stays

Two intelligences, according to Cattell

In the 1940s, the psychologist Raymond Cattell proposed splitting general intelligence into two parts, and in 1963 he published an experimental test of the idea. Fluid intelligence is the ability to solve new problems that do not depend on what you have learned: finding the rule in a sequence of figures, noticing a relationship nobody taught you. Crystallized intelligence is the knowledge and skills accumulated through experience and culture, such as vocabulary, general information and learned procedures.

In 1967, Cattell and John Horn compared groups of different ages and described the pattern that defined the field: fluid measures were lower in older groups, while crystallized ones held steady or were higher. The two are correlated, but they do not move together with age.

The distinction helps you read any test. Matrices, series and figure rotation lean toward the fluid side; vocabulary and verbal analogies also draw on the crystallized side. IQNorth mixes both, with new problems predominating, as the breakdown of the six areas shows.

The curve of reasoning in adulthood

Timothy Salthouse spent decades measuring these curves. In a 2009 paper, he pooled data from healthy, well-educated adults and showed that average performance in reasoning, processing speed and memory starts to decline as early as the 20s to early 30s, and keeps declining gradually after that. Vocabulary, by contrast, rises until around age 60.

When the decline begins is debated. Longitudinal studies, which follow the same people for years, usually show the decline later than studies that compare groups of different ages. Salthouse argued that part of this difference comes from the practice effect: people who retake the tests every few years do better through familiarity, and that masks the decline. It is the same phenomenon the retest has to discount.

Studies that compare groups have limits too: people born in different decades had different schools, health and environments, as the Flynn effect shows. Neither design is perfect, and both readings agree on the essentials: on average, reasoning about new problems is strongest in early adulthood.

These are averages. The variation among people of the same age is large, and the average curve says nothing about how any particular person's trajectory will go.

With age, average reasoning about new problems declines gradually, while accumulated knowledge holds up for much longer.

Differences between people are stable

Even if the average shifts, relative positions change little. The best-known example comes from Scotland. In June 1932, nearly every Scottish child born in 1921, then about 11 years old, took the same intelligence test at school, in what was called the Scottish Mental Survey. Decades later, Ian Deary and colleagues tracked down some of those participants, by then about 77 years old, and gave them the same test again.

The correlation between scores at 11 and at 77 was around 0.6, and close to 0.7 when corrected for the narrower spread of the group that was found again. More than six decades later, those who did better than their classmates at 11 tended, on average, to keep doing better. The correlation is far from 1, and there is real room for different trajectories, but relative position proved much more stable than many people imagine.

Why we compare by age group

An IQ indicates a position relative to a reference group, not a count of correct answers. If the reference group were only young adults, almost everyone older would show up below 100, not because they reason poorly for their age, but because average reasoning about new problems declines over the years. That is why the classic tests, going back to David Wechsler's scales, compare each person with people in the same age group.

IQNorth does the same. One of the three questions about you is your age, and it changes the comparison; education and self-rating do not change the number. Today's age norms are provisional: they use parameters from ICAR, a public-domain battery administered online to nearly 97,000 people (Condon and Revelle, 2014), corrected because people who take part in online research score higher than the general population, and they follow the age curve from the literature, such as Salthouse's (2009). They will be recalculated with real responses, separately for each language. The method page describes the details.

Ceiling and floor shift with age

Every short test has limits. For adults aged 20 to 29, IQNorth does not distinguish above about 138 or below about 57: when someone gets almost everything right or almost everything wrong, there are not enough questions to separate values beyond that. Because the comparison is made by age, these limits shift in the other age groups, and the report shows yours.

The same performance can therefore correspond to different numbers at different ages. That is exactly what comparing you with people at the same point in life means, and the adjustment has nothing to do with flattery.

See also: How we calculate.