Does IQ Predict Income? The Link Explains About 5 Per Cent

The question arrives in two forms and they deserve different answers. Does measured intelligence relate to income across a population? Yes, reliably, and the size of that relationship has been estimated many times. Does an IQ score tell you what any particular person will earn? No, and the same numbers that establish the first answer are what rule out the second.

The number, and what a correlation of 0.23 means

The reference point most researchers reach for is a meta-analysis by Tarmo Strenze, published in Intelligence in 2007, which pooled longitudinal studies that measured ability first and socioeconomic outcomes later. Its estimate for intelligence and income is a correlation of roughly 0.23. Correlations are not percentages and cannot be read as though they were, but squaring one gives a rough sense of shared variation, and 0.23 squared is about 0.05.

So around five per cent of the differences in earnings between people line up with differences in measured ability, and about ninety-five per cent do not. That remaining share is not noise. It contains occupation and industry, hours worked, region, the state of the labour market in the year someone graduated, family wealth, negotiation, health, discrimination, credentials, luck, and the simple fact that some fields pay more than others regardless of who enters them.

Where the effect actually travels

The more revealing figures in the same analysis are the ones that are larger. Intelligence correlates with educational attainment at roughly 0.56 and with occupational status at roughly 0.43 – both considerably stronger than the link to income itself. Read as a chain rather than as three separate facts, that pattern tells a coherent story.

• Measured ability relates fairly strongly to how much education someone completes.
• Education relates strongly to which occupations are open to them.
• Occupation relates to pay, but loosely, because pay within any occupation varies enormously and pay between occupations is set by markets rather than by ability.
• By the time the chain reaches income, the association has been diluted at every link.

This is also why the figure is so often misquoted in both directions. People who want the number to be large cite the education correlation and call it income. People who want it to be zero point at the ninety-five per cent unexplained and call the whole thing meaningless. Neither is reading the table.

The 2026 finding: your specific strengths matter more than assumed

A study by Tobias Edwards and Colin DeYoung, published on 1 July 2026 in Intelligence & Cognitive Abilities, revisits the question with two large American cohorts: the National Longitudinal Survey of Youth 1979, with 11,914 participants, and its 1997 counterpart, with 7,008. Both administered the Armed Services Vocational Aptitude Battery in adolescence, and both followed participants into adult life.

The authors separated general intelligence from three narrower factors – technical, speed, and math-verbal ability – and asked how much each contributed to later education, income and occupational status. Their headline is that the specific abilities carry between 30 and 57 per cent of the importance of general ability. For income the ratio was 0.57 in the older cohort and 0.35 in the younger one. Technical ability in particular clustered people into occupations at about 66 per cent of the strength that general ability did, and among men roughly 80 per cent. The associations survived comparisons between siblings, which is a meaningful control because siblings share much of the family environment that otherwise confounds this kind of work.

What a score cannot tell you about your salary

The gap between a population statistic and a personal forecast is the point at which most coverage of this topic goes wrong. A correlation describes the average tilt of a cloud of points. It says nothing about where any single point sits, and with a correlation near 0.23 the cloud is very nearly round. At every score on the scale you will find people earning a great deal and people earning very little, and the overlap between any two score bands is far larger than the difference between their averages.

There is a measurement caveat on top of the statistical one. Any single test result is an estimate with a confidence range around it, and the range is wide enough that two people separated by a few points are not reliably different. Building a life forecast on a point estimate compounds an uncertain number with an uncertain relationship. Our percentile calculator shows how much of the population sits within a few points of any given score, and the score converter makes the point again from a different angle: the same performance prints as a different number depending on the scale it was reported on. Our explainer on what a cognitive ability test at work actually predicts works through the same problem for hiring, where the corrections applied to validity figures turned out to matter enormously.

What the other ninety-five per cent is made of

It is tempting to read an unexplained share as randomness. It is not. Earnings are set by a long list of things that have little to do with reasoning ability and a great deal to do with structure, and several of them do more work than ability does.

• Occupation and industry. A given level of ability is paid very differently in software, teaching, nursing and logistics, and that gap is a fact about markets rather than about people.
• Hours and continuity of employment. Time out of the labour force, part-time work and caring responsibilities move lifetime earnings substantially.
• Region and cohort. Entering the labour market in a downturn is known to depress earnings for years afterwards, independently of who you are.
• Family background and credentials. Parental resources shape which schools, internships and professional networks are available, on top of any measured ability.
• Bargaining, discrimination and luck – none of which appear in an ability measure, and all of which appear in pay.

The honest framing is therefore not that intelligence matters little, but that it is one input into a process with a great many. A variable accounting for five per cent of the variation in an outcome is not nothing across a population of millions; it is simply nowhere near enough to forecast a person. Most of the public argument about this topic is really a disagreement about which of those two sentences to leave out.

How to read any IQ-and-outcome chart

• Check whether the vertical axis is average earnings for a score band or the earnings of individuals. Averages hide the spread, and the spread is the finding.
• Check whether ability was measured before the outcome or at the same time. Cross-sectional data cannot separate cause from consequence.
• Check whether education is in the model. Much of the ability-income link runs through schooling, so controlling for it or not changes the number substantially.
• Check the country and the decade. Labour markets differ, and the 2026 analysis found the income ratio shifting between two cohorts of the same country.
• Check whether the writer squared the correlation. A great many articles quote 0.23 and then describe it as though it were 23 per cent.

The defensible summary is that measured intelligence is one modest input among many into earnings, that most of its influence appears to run through education and occupational entry rather than directly, and that the specific shape of someone's abilities may matter more than the single summary number has traditionally been given credit for. If you want the underlying construct explained first, start with our piece on what the g factor actually is, then take the test if you want a current estimate of your own.

Common questions