What an online IQ test can and can't tell you

The first intelligence scales, built by Alfred Binet and Théodore Simon in France and revised for American schools by Lewis Terman, measured children against what children of each age could usually do. Terman explained it this way in 1916: if a 9-year-old "goes only as far as normal 8-year-old children ordinarily go, we say that his 'mental age' is 8 years."

The intelligence quotient turned that into one number. In Terman's words, "the intelligence quotient (often designated as I Q) is the ratio of mental age to chronological age", multiplied by 100. A child whose mental age matched their real age scored 100, and Terman's rule was that a correct scale must give a median of 100 "for unselected children of each age".

Terman also gave the numbers names. In his classification, an IQ from 90 to 110 counted as "normal, or average, intelligence", a band that has stayed roughly where it is ever since.

The ratio works only while abilities keep growing with age. They don't. David Wechsler, who designed an intelligence scale for adults, wrote in The Measurement of Adult Intelligence, first published in 1939, that "beyond the age of 15, mean scores on practically all intelligence scales cease to increase significantly with age".

He put the problem plainly: "A mental age score of 12 years, for example, in a child of twelve does not represent the same level of mental ability that it would in the case of a man of forty."

So a mental age stops making sense for grown-ups. "To speak of an individual having an M.A. of 20 years", Wechsler wrote, "is both practically and scientifically meaningless." Terman himself had set aside any age above 16 when working out an IQ.

Testers had patched the gap by dividing every adult's mental age by the same assumed age. Wechsler noted that adult IQs "have been almost universally calculated by using some assumed constant denominator", and argued that psychologists had been "calculating not intelligence quotients but efficiency quotients".

Wechsler's answer was to drop the ratio and compare each person with people of the same age. As he put it, "an intelligence quotient measures a person's ability relative to those of individuals of his own age group".

To do that, a test needs norms: the scores of a large group of people of each age who took the same test in the same way. A person's score is then converted into a position within their own age group's spread of scores. Wechsler kept 100 as the middle because, he wrote, "the historical definition has set it once and for all at 100". In his scheme, the Average band ran from 91 to 110 and covered the middle half of people.

Such a score, in Wechsler's words, "merely defines his relative position among the group with which he has been compared". Change the group and the same answers produce a different IQ.

The key point is that an IQ is a position, not a quantity. It says where you stand among the people the test was normed on. Without that group, there is nothing to stand among.

The group a test is compared with doesn't stay still. Scores on intelligence tests rose for decades across much of the world, a pattern known as the Flynn effect.

In the first formal meta-analysis of it, Jakob Pietschnig and Martin Voracek pooled 271 samples, almost 4 million people in 31 countries, tested between 1909 and 2013. Full-scale IQ rose by about 0.28 points a year. The gains were larger on reasoning tasks, about 0.41 points a year, and smaller on knowledge-based ones, about 0.21. They were stronger for adults than for children, and they have slowed in recent decades.

A test scored against old norms therefore compares you with a group that would score lower today, which makes everyone look a little better than they are. That is why professional tests are normed again from time to time.

Taking a test again usually improves the score, even when nothing about you has changed. A 2018 meta-analysis by Scharfen, Peters and Holling pooled 122 studies, with more than 150,000 people taking cognitive ability tests up to four times. They found "significant retest effects and no further score gains after the third test administration".

The size of the gain depended on the kind of ability tested, whether the same version of the test was used, the time between attempts and the person's age. For a puzzle test, the reason is easy to picture: the second time, you already know what kinds of puzzle are coming and how to approach them.

• 24 original puzzles of four kinds: pattern grids, number series, logic and shape rotation, ordered from easy to hard.
• Every puzzle is checked by code when the site is built: its rule must hold, and exactly one of the options may follow it. An unfair puzzle stops the site from being published.
• A 20-minute clock. Each right answer scores one point; a skipped puzzle scores nothing, and there is no penalty for guessing.
• An estimated IQ from a fixed table: 12 of 24 is 100, each puzzle is worth about 3 points, and the scale runs from 80 to 135.
• It has not been standardized against a representative group of people. The 100 on our scale is where we placed the middle score, not the measured average of any group.
• The puzzles are never shown after the test, to you or on a shared result, so a retake or a friend's attempt stays fresh.

So read the number as a friendly estimate of how you did on these 24 puzzles, today. It can show you which kinds of puzzle come easily to you, through the breakdown by puzzle type. It can't place you among the population the way a professionally normed test can, and it isn't an official IQ score.

• Take it when you have 20 quiet minutes and won't be interrupted. The clock keeps running if you leave.
• Use the first attempt as your real result. Later attempts are higher partly because you have seen the puzzle types.
• Skip a puzzle that stalls you and come back to it. Every puzzle is worth the same point.
• Look at the breakdown by puzzle type, not only the headline number. It shows which kinds of puzzle suit you.

This is a puzzle test for fun and practice. The IQ shown is an estimate on our own scale, not the result of a standardized IQ test, and it says nothing about your health.