What IQ predicts and what it does not Image: J. E. Wallace Wallin, 1911, public domain, via Wikimedia Commons

What IQ predicts and what it does not

An IQ test is a collection of short problems: vocabulary, arithmetic, finding the pattern in a sequence of shapes, repeating digits backwards, assembling blocks to match a design. The scores on the different problems correlate with one another, which is the fact Charles Spearman noticed in 1904 and called g, and the total is converted to a scale with a mean of 100 and a standard deviation of 15 for the person's age group. About two thirds of people score between 85 and 115, and about 2% score above 130. Scores on the main tests, the Wechsler scales and their rivals, correlate at about 0.9 with each other and with themselves when the test is repeated, which makes IQ one of the most reliably measured quantities in psychology.

Reliability is not the question most people ask. They ask what the number is good for. The answer comes from longitudinal studies that measured children and then waited.

School

The largest such study followed about 70,000 children in England who took a cognitive ability test at age 11 and the national GCSE examinations at 16. Ian Deary and colleagues reported in 2007 that the correlation between the test score and the examination results, estimated for the underlying abilities, was 0.81, and about 0.7 for the raw scores in English and mathematics. The paper is at the journal. A correlation of that size is unusual in psychology, and it means that a test taken five years earlier accounts for about half the variation in examination results. It also means that half is accounted for by something else: effort, teaching, home, health, and the test's own error.

Work

Frank Schmidt and John Hunter reviewed 85 years of studies of personnel selection in 1998, at the journal, and found that general mental ability predicted job performance better than any other single measure, with a corrected correlation of about 0.5, rising with the complexity of the job. The figure has been revised downwards since: a re-analysis in 2022 that used more conservative corrections put it at about 0.3, with structured interviews and job knowledge tests doing as well or better. Either figure makes IQ useful for an employer choosing among hundreds and nearly useless for predicting how any one of them will do, because a correlation of 0.3 leaves more than 90% of the variation in performance unexplained.

Health and length of life

The finding that surprised the field came from Scotland, where almost every child born in 1921 and 1936 sat an intelligence test at 11, and the records survived. Deary's group traced the children into old age. A meta-analysis by Catherine Calvin and colleagues in 2011, with 16 cohorts and more than a million people, free to read, found that a one standard deviation higher score in childhood went with a 24% lower risk of death over the following decades. Part of the link runs through education and income, part through health behaviour, such as not smoking, and part may reflect that a brain that scores well at 11 is a brain in a body that is working well. The relationship is one of the most replicated in the field and nobody fully understands it.

The scale that moves

In 1984 James Flynn noticed something odd in the manuals of the tests. Every time a test was restandardized, the new sample scored higher on the old version than the old sample had. He collected the data for 14 countries in 1987, in a paper at the journal, and found gains of 5 to 25 points in a generation everywhere he looked. A meta-analysis of nearly four million test takers from 1909 to 2013, by Jakob Pietschnig and Martin Voracek in 2015, at the journal, put the Flynn effect at about 2.8 points per decade, which is about three; the figure the masterclass uses in Unit 11.

Taken at face value, the average person of 1920 scored about 70 by today's norms. Nobody believes that the grandparents were on the edge of intellectual disability, so the rise has to be in something the tests measure that is not the thing we mean by intelligence in daily life: familiarity with abstract classification, schooling, test-taking itself, and the general health and nutrition of children, in some combination that is still argued about. The effect has also stopped in several countries. In Norway, where every male conscript is tested, scores rose until the cohorts born around 1975 and have fallen since, and a 2018 analysis of brothers, free at PNAS, showed that the fall occurs within families, which rules out explanations based on immigration or who is having children. Whatever raised the scores for a century is no longer doing so there.

What it does not do

An IQ score does not measure creativity, wisdom, curiosity, or kindness, and it predicts them poorly; conscientiousness predicts academic grades almost as well as IQ once ability has been taken into account, and the personality traits predict length of life about as well. It is not fixed. The correlation between a child's score and the same person's score as an adult is far from 1, the heritability of the score rises from about 0.2 in infancy to 0.6 or more in adulthood, and schooling raises it: a year of extra education adds somewhere between one and five points, according to a 2018 meta-analysis of natural experiments. And it is not a ranking of worth, which is a statement a scientist should not have to make and does, because the history of the tests includes their use for exactly that.

Differences between groups require the most care of all. Average scores differ between countries and between groups within countries, and the differences have shrunk where the environments have converged, which is what the Flynn effect would lead you to expect. The claim that the gaps have a genetic cause has no direct evidence behind it, and the within-group heritability of a trait says nothing about the cause of a between-group difference, a point that Unit 11 spends a section on because it is so often got wrong.

How to hold the number

The correct way to think about an IQ score is the way a doctor thinks about blood pressure. It is a real, reliable measurement of something that matters, it predicts outcomes across a population with useful accuracy, it says much less about any one person than its precision suggests, and it can be changed. Any claim about IQ that would not survive being rephrased about blood pressure is probably wrong. The one difference is that nobody feels that their blood pressure is a judgement on them, and the number 100 has carried that weight for a century. The best cure for it is to learn how the number is made, which the masterclass does with the normal curve, the reliability figures, and the correlations above, all of them on the key findings sheet.

Study the mind properly

The Psychology Masterclass: weekly units, quizzes, a final exam, and a certificate.
About the program