Does training raise IQ? What the meta-analyses say

The promise of 2008

In 2008, Susanne Jaeggi and colleagues published a study in PNAS that seemed to settle an old question. Young adults trained on a task called dual n-back: they watched squares appear at positions on a grid while hearing letters, and had to signal when the position or the letter matched the one from a few rounds earlier. After a few weeks, the trained group improved on a matrix reasoning test, and the more sessions they did, the bigger the gain.

Matrix reasoning is considered a typical measure of fluid intelligence, the ability to solve new problems. If a memory exercise improved it, the conclusion seemed obvious: intelligence can be trained. The study was widely cited, and in the years that followed an industry of “brain training” apps grew up promising exactly that effect.

A single study, however good, does not close the question. The samples were small, the comparison group did no equivalent activity, and the reasoning test was given under a shortened time limit. What was still missing was a look at what happened when dozens of studies were added together.

What the meta-analyses found

A meta-analysis pools the results of many studies on the same question and calculates the average effect, taking each sample's size into account. It is the kind of evidence that depends least on the luck of any one lab.

In 2013, Monica Melby-Lervåg and Charles Hulme gathered the working memory training studies published up to that point. The result pointed in two directions. Training improved memory tasks similar to the trained ones, but that gain tended to fade over time. For more distant skills, such as nonverbal reasoning, reading or arithmetic, there was no convincing effect, especially when the comparison group did some control activity instead of doing nothing.

In 2019, Giovanni Sala and Fernand Gobet reviewed a larger body of work covering memory training, video games, chess, music and other programs. The paper's title sums up the conclusion: cognitive training does not enhance general cognition. Once the authors accounted for the quality of the control group and for publication bias (studies with positive results are more likely to get published), the effect on distant skills came out close to zero.

Other reviews reached similar conclusions. In 2016, a team led by Daniel Simons examined the evidence cited by the brain training companies themselves and found improvement on the trained tasks, with little evidence of benefit on different tasks or in daily life. Controlled replications of the 2008 study, such as the one by Thomas Redick and colleagues in 2013, also found no gain in intelligence.

Training improves what was trained. What the meta-analyses do not find is the leap to reasoning in general.

Near and far transfer

Researchers distinguish two kinds of gain. Near transfer is improvement on a task similar to the trained one: someone who practices remembering strings of digits gets a little better at remembering strings of letters. Far transfer is improvement on something of a different nature, like solving matrices after training memory.

Near transfer shows up often. Far transfer all but disappears when the study is well controlled. A likely explanation is that much of the gain comes from strategies and familiarity with the format: people learn to group the items, to ignore that screen's distractions, to handle the anxiety of that kind of challenge. That has value, but it does not change the underlying ability.

• On the trained task: a large, reliable gain.
• On similar tasks: a smaller gain that tends to fade months later.
• On reasoning in general, at school, at work: an effect close to zero in well-controlled studies.

This difference also explains why a test retaken after a lot of training in the same format can go up without anything underlying having changed. That is the practice effect, the subject of the article on retesting and practice.

What school does

There is one intervention with good evidence of an effect on IQ: schooling. Separating cause and effect here is hard, because people who reason more easily tend to stay in school longer. Stuart Ritchie and Elliot Tucker-Drob tackled this problem in 2018 with a meta-analysis that used only designs able to isolate the effect of school, such as changes in the law that raised the minimum school-leaving age, and comparisons between children who, because their birthdays were just a few days apart, started school a year earlier or later.

The estimate was about 1 to 5 IQ points per additional year of schooling, depending on the type of design, and the effect persisted into adulthood. School, though, is nothing like a repeated exercise: it is years of reading, math, vocabulary, argument and varied problems. What sets it apart from an app is breadth and duration, not some trick of format.

The same reasoning helps explain the Flynn effect, the rise in scores over the 20th century, which is usually linked to broad changes in environment, such as more years of school and better health.

Why IQNorth does not promise an increase

IQNorth offers daily training for subscribers: 10 exercises a day, about 8 minutes, across the test's six areas, with a level from 1 to 5 that goes up when you get answers right and down when you miss. The exercises come from our own bank of more than 800 items, kept separate from the test, so that practicing does not turn into memorizing answers.

Based on the reviews above, we do not claim that training raises IQ (Sala and Gobet, 2019). It makes you more familiar with the formats, explains each exercise and shows your level in each area. Your training level is not IQ and is never shown as IQ.

If you want to check for a change, you can use the retest, which uses different questions, discounts the practice effect and only reports a change when it exceeds the margin of error. Most people come out with a stable result, which is exactly what the research predicts.

See also: Daily training.